In the mid-2000s, Verisk, faced with limited growth opportunities in an industry it had come to dominate, began placing bets beyond its core, confident that its experience building and managing datasets in one domain would carry over into others. It purchased a series of healthcare targets starting in 2004, then splashed into residential mortgages in 2008 through the acquisition of Interthinx, expanding that foothold with smaller add-ons that together formed the basis of a budding financial-services business, which would later include Argus ($425mn; 2012), a provider of credit card data that banks used for marketing purposes. Argus then crescendo’ed into the $2.8bn acquisition of Wood Mackenzie, a supplier of data and analytics to the energy and natural resources industries.

This all turned out exactly as well as you’d expect. The healthcare assets, a persistent drag on margins and growth, were divested in 2016. Argus’ high-teens organic growth decayed to low-single digits while its profitability collapsed. Wood Mac’s margins flatlined and growth atrophied in the midst of E&P headwinds. Verisk rid itself of these non-core, underperforming divisions, in no small part thanks to D.E. Shaw’s activist agitations.

In its Value Creation Plan, D.E. Shaw laid out a plan for Verisk to grow EPS by 35% over the next 3 years, which, combined with multiple expansion (23x to 29x), would drive the share price 72% higher:

Well, 3 years have now passed. Last year’s EPS of $7.16 fell way short of Shaw’s $10.80 target (and of Consensus expectations too for that matter). The stock briefly kissed $313 on multiple expansion but has sold off over the past year on slower-than-expected organic growth and on, you guessed it, AI fears.

Before getting to that, it is worth briefly recapping what the slimmed down Verisk looks like today. It is now a pure-play insurance business, with roughly 60% of its portfolio tied to commercial lines, 40% to personal lines1. It gets around 70% of revenue from primary insurers, the balance from brokers, reinsurers, managing general agents (MGAs), and claims professionals. Around 80% to 85% of revenue is subscription-based, the other 20% to 25% scales with insurance premium growth. Most major clients are locked in through 3-year contracts with embedded price escalators.

Though it serves a cyclical industry, Verisk is bond-like in its consistency. With the sole exception of 2020, when it grew 5%, the company has organically grown revenue by 6% to 8% every year since its 2009 IPO. The cycle barely registers: revenue growth averaged 6.8% during soft markets, when underwriting capacity grows and pricing declines, and 7.3% during hard markets. Management expects more of the same, with 6% to 8% organic growth – split across pricing (3.5%-4.5%), cross/up-selling (1.5%-2%), and new products (1.5%-2%) – outpacing the insurance industry’s mid-single-digit growth over time. With 25bps to 75bps of annual margin expansion, EBITDA and EPS are expected to grow by 7%-10% and double-digits, respectively. A classic compounder profile.

What Verisk is best known for is aggregating detailed operational data from insurers and turning that data into standardized policy forms and loss benchmarks that are used as a starting point to build or update products. So rather than drafting every policy and pricing framework from scratch, an insurer can adjust Verisk’s templates and benchmarks to its own risk appetite and strategy. This core underwriting solution is surrounded by other datasets, including labor and material cost data for sizing the cost to rebuild a home or commercial property (360Value), and catastrophe models for gauging the probability and financial impact of wildfires, severe storms, and other extreme events in a given area (AIR).

In Verisk’s claims processing division, roughly half of revenue comes from insurers, the other half from third-party adjusters, contractors, and other ecosystem partners. Here, Verisk primarily offers two solutions.

The first, Xactware, which makes up the core of Property and Restoration Solutions (14% of total revenue, ~half of claims revenue), helps put on a number on physical damage and repair work. When a hailstorm damages a roof, a burst pipe floods a house, or a fire tears through a kitchen, insurers, adjusters, and often contractors turn to Xactware to estimate the materials and labor a repair will require and what it should cost.

The second, anti-fraud tools, comprising ~1/3 of Verisk’s claims-related revenue, are used to spot suspicious claims activity and detect fraud after a loss event. Verisk’s primary fraud fighting solution, ClaimSearch, is the world’s largest database of P&C claims information. Insurers comb through its 1.9bn+ records to uncover links across claims records and assess, for instance, whether someone has a history of filing lots of claims or whether their address or phone number has surfaced in other suspicious cases. Alongside ClaimSearch sit other anti-fraud tools, such as a digital media database of more than 600mn images – contributed in part by 5 of the top 10 carriers – that insurers use to evaluate suspicious claims photos (management cites research showing 30% of US consumers, and 55% of Gen Z respondents, consider it acceptable to manipulate submitted photos to maximize a payout).

These foundational datasets, undergirding Verisk’s underwriting and claims products, are complemented by workflow software and analytics that make use of it.

While many have excoriated Verisk’s misadventures in healthcare, financial services, and energy, its moat in insurance data and analytics has rarely come under question. Well, there was that brief flare-up in the mid-2010s, when some feared competitive encroachment by insurtech start-ups, a threat that ultimately proved toothless and passed without incident. But now investors are worried that AI could commoditize the job of gathering data and synthesizing it into usable insight, rendering Verisk’s moat obsolete. S&P, Moody’s, and other vendors of data and analytics have also sold off on this fear.

Subscribe now

Read more

A few months ago, I got around to reading Barry Diller’s memoir, Who Knew. The guy’s career basically toured through every phase of modern media - talent agency, broadcast television, movie studio, fourth network (Fox), cable/home shopping - and culminated in an internet holding company we know today as IAC.

IAC, founded in the midst of the dot-com mania, positioned itself as a sober counterweight to the era’s frenzy:

I started with no business model other than liking companies that had some actual revenue and profits as opposed to sexy-sounding ideas and big talk. I turned an icy eye on the early schemers and hustlers who spent huge marketing dollars to build audiences but had no path to actually making any money. We survived all that because of our values of fiscal prudence, but we were also always willing to take a flier on a good idea, though we would never bet the company on a single transaction.

Over time, and again with no foresight, we did evolve into a unique business model: the anti-conglomerate conglomerate. If you own dozens and dozens of disparate businesses, you’re certainly a conglomerate. And we’ve housed as many as sixty different brands. But I had become opposed to the concept of agglomeration. Bigness for bigness made no sense to me. When I decided a business was sufficiently developed, we’d spin it out into an independent entity. As we built up all these entities, I thought managing them centrally wasn’t the best way. Once they achieved some scale they ought to stand on their own – make their own decisions with independence from the mother ship. We’ve done that now with ten separately traded public companies.

The payoff to this approach has been decent but lopsided, with a few extraordinary outcomes (Match, Expedia, and Live Nation/Ticketmaster) intermingled with a long tail of mediocre-to-disappointing ones (LendingTree, Angi, and Care, among others).

But even the successes can be retrospectively painted as failures depending on your starting point. Vimeo was acquired last year for $1.4bn – a roughly 4x return on the ~$326m IAC put into it across 15 years of ownership, but a steep markdown from the $5bn at which it raised primary capital in January 2021 and the $8bn valuation at which it was spun off later that year. Match’s $8.5bn market cap sits well above the roughly $3bn it paid to assemble its dating portfolio, but far below its 2021 peak of $50bn. HSN, another IAC property, was acquired by QVC/Liberty in 2017 in an all-stock deal valued at $2.6bn. But QVC recently filed for Chapter 11, casting doubt on the viability of the linear-TV home-shopping model in a world of e-commerce and declining traditional TV viewership. LendingTree could have been a great outcome had you sold it in 2019, when its valuation his $5bn. Not so much today, with its $540mn market cap languishing below the $626mn to $734mn IAC paid for it.

IAC’s two most significant and indisputable wins so far are Expedia and Ticketmaster. Expedia today commands a ~$28bn market cap vs. IAC’s ~$5bn cost basis1. Ticketmaster, which merged with Live Nation in 2010, has been even more lucrative. IAC shareholders’ stake in the combined company would be worth ~$14bn2 today against an initial investment of $678m, excluding the $1.4bn that IAC realized just prior to Ticketmaster’s spinoff.

Reading the memoir, one gets the sense that Diller often found himself torn between the prestige of running legacy media and the excitement of whatever was emerging at the edges of it. At various points in his storied career, Diller stepped away from industries at their peak and into messier or less mature ventures. He left ABC for Paramount when network TV was still dominant and resigned as CEO and Chairman of Paramount to launch a challenger network, Fox. After Fox ballooned into a credible fourth network, he again resigned as CEO and Chairman in pursuit of QVC at a time when home shopping was mocked:

Subscribe now

Read more

Here’s roughly where things stood at IAC in September 2022, when I last wrote up the company:

There was a time, not too long ago, when an investor could treat IAC as a collection of binary early stage bets anchored by a few more mature and profitable entities. To most, IAC’s 12% passive minority stake in MGM, acquired near the COVID lows, was a weird one-off opportunistic gambit that could be marked to market. The real diligence was saved for Angi’s, DotDash, and Vimeo. I don’t think many felt compelled to deep dive into casinos because like, whatever, the $1bn MGM investment comprised just ~9% of IAC’s market cap at the time and it felt like most of the upside was going to come from IAC’s digital assets, not a passive minority stake in a mature brick-and-mortar casino operator.

Psych!! Now Angi’s is melting away as its fixed price offering, launched to great fanfare 2-3 years ago, struggles to find product-market fit. DotDash no longer expects to hit its $450mn EBITDA target next year given the weakness in brand advertising. Vimeo has lost ~90% of its market cap since being spun off last May, as losses have widened, growth has dramatically decelerated off tough COVID comps, and investors have soured on unprofitable growth concepts.

Meanwhile, MGM shares have nearly doubled from the ~$17 price at which IAC first accumulated shares in 2q and 3q 2020. IAC has since added to its MGM stake, first at $45 and more recently in the low-$30s. With valuations in digital growth assets wrecked and its own stock price sliced in half, IAC could have opportunistically acquired another digital lottery ticket or more aggressively repurchased (more of) their own shares. But no. They increased exposure to MGM instead. Today, the $2.1bn MGM position accounts for 35% of IAC’s market cap, making it the second most valuable asset in the IAC complex after DotDash.

Much has changed in the intervening years. Bluecrew, an unprofitable marketplace for temp work, was sold to EmployBridge in 2022 for $50mn in cash and equity. A few years later, Mosaic, a collection of mobile subscription apps, was acquired for $160mn by Bending Spoons, an Italian app developer.

A few of IAC’s more prominent properties were jettisoned as well. Care.com, an online marketplace connecting families with caregivers, was sold by IAC in March 2026 to private equity firm Pacific Avenue Capital Partners for roughly $320 million, about 36% below what IAC had paid to acquire it in 2020. As COVID tailwinds passed, Care’s underlying product deficiencies came unmasked. Subscriber count declined for “8 or 9” consecutive quarters through late 2025.

Angi was another problem child that management at long last offloaded. Many of you will be familiar with this saga. After a series of unsuccessful attempts to drive engagement, Angi – which had traditionally earned fees for matching homeowners to service providers – launched a fully managed booking platform, following the lead of other online lead generators, including Yelp and Zillow, who had similarly assumed greater responsibility for facilitating transactions further down the funnel as Google and Meta siphoned away an ever-larger share of leads up top.

By fixing the price of services, Angi gave homeowners a more convenient offering but put itself on the hook for any cost overruns. The results were disastrous. Angi “lost an embarrassing amount of money” as the fixed-price model was extended to complex, higher-value renovation and roofing jobs. The premise that Angi could use cost data from jobs intermediated through its marketplace to price work accurately turned out to be a fantasy. This strategic misadventure coincided with degrading cost discipline and ongoing culture clashes between between Angie’s List and HomeAdvisor, who continued to operate with separate sales forces and product types.

As a fully independent company, ANGI is now “shrinking to grow”, filtering out lower quality service pros and cutting costs to improve match rates and profitability, even at the expense of revenue. I don’t think I would ever invest in ANGI…too much brain damage, especially with AI presenting fresh unknowns. BUT! It’s worth noting that EBITDA (including stock comp) has steadily improve – $16mn → $75mn → $111mn → $125mn – in each of the last 4 years. Having lost 60% of its value this year, the stock now trades at just under 4x trailing EBITDA with 2x turns of net leverage.

Free of its flailing, non-core businesses, IAC is now left with the following assets:

Subscribe now

Read more

Today, a quick post on a small company with a colorful origin story.

We don’t often think about the paper trail we leave behind as we move through the world. The DMV records, property deeds, utility bills, corporate filings, liens, and bankruptcy petitions that trace the arc of our lives are stored across a diffuse set of public and private registries. This data is vacuumed up, loaded into a single searchable system, and linked together by companies that, as it turns out, comprise an incestuous micro-industry spawned out of Boca Raton by a cast of characters orbiting around one man: the legendary (or notorious) Hank Asher.

A high school dropout from Indiana, Hank founded a house-painting business in Boca Raton at age 18, sold it in 1982 when it reached $10mn in revenue, and then did what any self-described adrenaline junkie with a pilot’s license living South Florida at the time would have done – smuggle marijuana and cocaine to and from Central and South America, obviously. This supposedly brief 7-week stint never resulted in a conviction and concluded with him working as a DEA informant.

The next decade of Asher’s life through 1992 is shrouded in some mystery. Whatever else Hank was up to during that period, he also developed enough expertise in computing to launch Database Technologies, which ran clusters of PCs in parallel to mine data. Its flagship product, AutoTrack, ingested the records of Florida’s DMV and was widely used by law enforcement agencies and insurers to find people and tie them to assets, claims, and fraud-related clues.

DBT was taken public in 1995 and acquired by ChoicePoint 5 years later for roughly $462mn. ChoicePoint, one of the industry’s key players today, was then purchased by Reed Elsevier, the parent company of LexisNexis, who in order to satisfy the FTC’s concerns of overlapping investigative-data assets, divested CLEAR, a ChoicePoint product that went on to become a notable ID resolution competitor.

By then, Asher was no longer involved. After news of his drug-smuggling past spread through the law enforcement community, the FBI and DEA halted their contracts with DBT in 1999, and Asher was pushed out and forced to sell his stake back to the company for $147mn. He immediately founded a rival data mining firm, Seisint (short for “seismic intelligence”), whose crown jewel, Accurint, competed directly with ChoicePoint. Derek Dubner, now CEO of Red Violet, joined Seisint in 1999 as corporate counsel, and Ole Poulsen – later Red Violet’s Chief Science Officer – was hired as CTO.

Subscribe now

Read more

Payment processing stocks on the whole have had a rough 12 months:

For some of these names, one can propose idiosyncratic reasons that might explain the sell-off – Shift4 is an opaque and levered serial acquirer; Global Payments and Fiserv are sclerotic, share-donating incumbents; PayPal took an ambitious strategy and executed it disastrously. Adyen, by contrast, operates a unified, organically developed platform, unencumbered by debt and guided by operators with skin in the game. But its stock, too, has collapsed. Having swooned 65% from the highs, it is not much higher than it was 6 years ago, even as payment volumes have sextupled and EBITDA has quintupled. This can sometimes happen to a stock that starts off at ~125x earnings!

Adyen’s commercial success has been supported by 2 unique pillars that should be familiar to most of us by now.

First, unlike legacy merchant acquirers, who spent decades assembling a patchwork of disparate providers through M&A, Adyen created, entirely in-house, a single unified that extends across all its markets. Whereas First Data and Global Payments, which operate more like an archipelago of payment platforms, must build the same product or feature multiple times to accommodate different code bases, Adyen need only do so once.

Second, whereas most peers, including modern PSPs like Stripe, still rely on aggregators and BIN rentals to serve different markets, Adyen holds banking licenses in every country in which it operates, giving it greater visibility into transaction details, which in turn translates to superior authorization rates. As I explained in some thoughts on Adyen:

Combining the gateway with merchant acquiring and processing means Adyen can underwrite merchants according to their own compliance standards and save on fees they would otherwise pay a third party acquirer. But more importantly, it gives them greater visibility into and control over the payments flow.

As an acquiring bank with direct access to card rails, Adyen can iteratively tweak messaging details in an authorization request to improve the chances that the issuer bank will accept it. If a payment fails due to a connection issue at the level of the acquiring bank or processor, they can more easily see and rapidly fix it than most other payment service providers, who obviously can’t inspect the systems of the acquiring banks they rent BINS from. Adyen, more so than any other PSP, is very opinionated about owning the full stack. They would rather not process payments in a country at all than gateway to third party acquirers, whose variable performance dilutes their exacting service standards.

These do not, on the surface, seem like much of a moat. Stripe, First Data, or any other processor could have made the same architectural and strategic choices early in their lives. I suspect one underappreciated reason they didn't is that the payments industry is saturated with FOMO. There is always a plausible-sounding urgent narrative for why BNPL or crypto or digital wallets or some other new thing needs to be squeezed into your roadmap now. Whether due to a cautious European temperament or the personality quirks of its founders, Adyen stands apart in its restraint. As competitors acquired targets in a frenzied attempt to keep pace with the latest themes, Adyen dedicated itself to strengthening its two foundational pillars, recognizing that a unified tech stack and vertical integration would remain relevant regardless of what new payment fashions emerged.

That discipline has produced a genuinely differentiated product. But Adyen prices for it accordingly, which has invited a familiar critique. Some have argued that in a competitive US market where payments is often treated as a commodity, Adyen’s relatively high processing fees are unsustainable. That concern came to a head in August 2023, when the company swiftly shed ~60% of its market cap after management cited increasing competitive pressure in North America. Net revenue growth there had decelerated sharply from 45% in 2h22 to 23% in 1h23. PayPal was partly to blame, having effectively given away Braintree processing at cost in an effort to drive volumes toward its higher-margin branded checkout button. It didn’t help that EBITDA margins were simultaneously compressed by an aggressive hiring spree that had not yet been absorbed.

Subscribe now

Read more

A common rebuttal to the “AI will take all the jobs” narrative is that even if AI can do a given task better than a human, a job is more than just a collection of discrete tasks. Lawyers don’t just draft contracts and flag unusual clauses. They also negotiate with stubborn opponents and decide which risks are worth haggling over. The better AI gets at the contract-drafting part, the more lawyers can focus on sections of the jagged frontier that still call for a human touch.

The issue for Upwork is that the remote freelancers on its marketplace are often hired to do a narrowly scoped, well-defined task. A marketing executive within an organization has a job. They design an ad campaign, decide which channels to lean into, and make judgment calls around budget, timing, and creative direction. The freelancers they hire handle have discrete tasks. They create ad copy, spin up logo concepts, and mock up landing pages. The obvious existential threat for Upwork is that AI will grow ever more capable of doing the tasks currently outsourced to freelancers, rendering its marketplace obsolete.

Naturally, Upwork counters that AI not a threat but a growth opportunity. CEO Hayden Brown insists, “AI is definitively positive for us. In fact, it is one of our most exciting and growing tailwinds” and mounts the usual defenses. AI is creating new categories of work even as it obsoletes old ones. Demand for prompt engineers, LLM application engineers, AI agent workflow engineers, model quality analysts, and data labelers have blossomed in the past several years.

Management is also quick to point out that its marketplace intermediates relatively more complex, value-added, recurring engagements. Projects priced at $300 or less, which are heavily made up of transactional jobs like writing and translation that are most at risk of AI substitution, comprise just 3.5% of gross service volume. Its GSV per active client comes in at around $5k compared to just $300 for Fiverr.

Subscribe now

Read more

Upcoming posts (in no particular order): Upwork, Adyen, PayPal, Gartner, Verisk, maybe Visa/Mastercard

As discussed in last week’s scuttlebit, Danaher’s life sciences coverage can basically be boiled down to 3 types of products: 1) diagnostic instruments and assays that hospitals and labs use to identify diseases, 2) research instruments and consumables that scientists use to study those diseases and discover drugs to fight them, and 3) tools and equipment that CDMOs and biopharmas use to manufacture those drugs. Thermo Fisher, nearly twice Danaher’s size by revenue, envelopes even more of the life sciences space. Whereas Danaher limits itself to products, Thermo Fisher provides services as well, running clinical trials on behalf of biopharma clients and manufacturing drugs for them too.

Here’s a run-through of what’s been happening in each of Thermo’s division over the last several years:

Life Sciences

The Life Sciences segment, created through the $15bn acquisition of Life Technologies in 2014, is made up of 3 business units: Biosciences, Genetic Sciences, and Bioproduction.

Thermo’s bioproduction business is much smaller than Danaher’s in both absolute terms (~$4bn1 vs. $6bn) and as a percentage of total revenue (9% vs. 25%). It boasts a strong position in cell culture media (nutrients that keeps cells alive and productive in the bioreactor) and single-use technologies (disposable bags and tubes), and maintains a decent, though not leading, presence in purification (chromatography resins and columns that isolate proteins of interest). Last year’s $4bn acquisition of Solventum’s Purification and Filtration division plugs the gap between those two steps, adding filtration technologies that remove cells and debris from the harvest before purification begins.

Subscribe now

Read more

Let’s start with Danaher, specifically its Diagnostics division, which sells instruments and associated consumables that hospitals use to diagnose diseases. Though accounting for 41% of revenue and EBITDA, it seems to draw far less attention than Danaher’s two, relatively smaller segments.

And I get it. Unlike Life Sciences, the locus of recent M&A, Diagnostics has been starved of acquisitions for the better part of a decade. And compared to Bioprocessing, it competes in a far more crowded arena, with formidable rivals across multiple testing categories, and enjoys weaker regulatory and process-validation lock-in.

But Diagnostics is also where Danaher’s vaunted kaizen values of continuous improvement, codified as the Danaher Business System, has arguably made its biggest mark on acquired targets. I mean, they’ve crushed it here:

(for reference: Radiometer manufactures blood gas analyzers that provide ICU and emergency room physicians with rapid measurements of oxygen, carbon dioxide, pH, and electrolyte levels in the blood to determine whether a patient’s lungs, circulation, and metabolism are functioning as they’re supposed to.

Leica Biosystems sits in the pathology workflow, its instruments and reagents used by oncologists to check for cancer in tissue samples collected in biopsies or during surgery.

Beckman Coulter is the backbone of “core labs”, providing products for routine blood and urine tests that are run in huge volumes.

Cepheid specializes in molecular diagnostics, particularly PCR testing through its GeneXpert platform. Its products are used to identify the genetic signatures – DNA, RNA – of pathogens responsible for diseases like tuberculosis, COVID-19, flu, and MRSA. See below the relative revenue contributions in 2023; I don’t think these have materially changed in the past two years)

Danaher’s 2024 Analyst Day gets into the gory details. Specialty Diagnostics (Radiometer + Leica) has grown revenue by hsd in the decade from 2013 to 20231 and generates returns on invested capital of more than 20%.

Beckman Coulter went from no revenue growth and paltry 12% EBITA margins to now growing msd+ (ex. China), with ~20% EBITDA margins and dd+ ROICs. By 2023, it generated twice as much new product revenue as it did 2018, having introduced more automation for lower-volume small and mid-sized labs and narrowed the menu gap to Roche and Abbott with assays for bloodborne viruses like HIV and hepatitis, as well as cardiac biomarkers linked to heart failure2.

Notwithstanding these gains, it’s fair to say that Beckman Coulter is a mustier asset than some of Danaher’s other diagnostic brands. Cepheid, for instance, enjoys a far stronger reputation in molecular diagnostics than Beckman does in routine lab diagnostics, a category that has become increasingly commoditized over time.

Subscribe now

Read more

Admitted and non-admitted insurance differ not only in the types of risks they cover and in the flexibility of underwriting, but also in how coverage is sourced and placed. In the admitted market, retail brokers, acting on behalf of insureds, typically go directly to standard carriers, like a Travelers or State Farm, to secure coverage. By contrast, the E&S market introduces an additional layer of intermediation, wholesale brokerage, that sits between retail brokers and carriers1. Wholesalers exist for two primary reasons:

1) Expertise: retail brokers tend to specialize by client and industry vertical. They know enough to guide clients through general liability, auto, workers’ comp, and other core coverages, but rarely possess deep expertise in every line, especially funkier stuff like certain cyber or employment practices liability, where coverage can hinge on highly nuanced policy language. So they turn to wholesalers who live and breathe those specialized exposures and understand which carriers or MGAs have appetite for which risks, at what price, and on what terms. Put another way, retail brokers tend to orient by client and industry (real estate, construction, healthcare) while wholesale brokers generally specialize by coverage type (cyber, directors & officers, environmental).

2) Carrier access: most retail brokers cannot deliver enough volume in a given niche exposure to justify a direct relationship with E&S carriers, who generally do not want to manage thousands of small distribution relationships. Many carriers prefer – and some effectively require – that business flow through wholesalers, who can aggregate submissions from a broad network of retailers. Large retail brokers like Aon and Marsh can often place business directly, but even they still rely wholesalers for oddball risks that fall outside core carrier relationships.

For example, imagine a retail broker representing the owner of a large apartment complex. The broker can comfortably place standard coverages like property, general liability, workers’ compensation, and umbrella, but may be less equipped to source coverage for a more specialized exposure, such as pollution liability tied to mold or asbestos. In that case, the broker would turn to a wholesaler with expertise in environmental lines, who would then approach the E&S carriers it knows have appetite for that risk, gather quotes and proposed terms, and help the retail broker evaluate the trade-offs among them. The carrier pays ~15%-20% of premiums as a fee to the wholesaler and the wholesaler in turn pays ~10% of that to the retailer, keeping ~5%-10% for itself.

In the flow I’ve described – client → retail broker → wholesale broker → carrier – the wholesaler places risk, but isn’t involved in any underwriting decisions. But there are other entities in the E&S ecosystem who are. They go by different names and have somewhat different responsibilities, but what they share in common is broad authority, granted to them by an insurance carrier, to underwrite and bind coverage.

Managing general agents (MGAs, or managing general underwriters, MGUs) design specialized coverage for complex or awkward classes of risk that standard carriers often avoid…higher-hazard retail businesses like liquor stores, cannabis dispensaries, and nightclubs; specialty contractors like as roofers and scaffolders; and habitational risks like older apartment complexes or coastal mobile-home communities. Many present themselves as product-line specialists for specific industries. Ryan’s stable includes MGUs that cover niche risks like post-harvest agriculture property (Agrisk Underwriters); amusement, entertainment, and leisure industries (Alive Risk), and commercial auto for hazmat trucking operations (Freberg Environmental).

Program managers are another subset of the broader delegated-authority world. Like MGAs, they operate under the authority granted to them by carrier partners. But whereas MGAs primarily execute underwriting and binding within a carrier’s existing framework, program managers function more like product owners in that they also design the coverage offering, develop underwriting rules and rating logic, coordinate reinsurance, and otherwise oversee the day-to-day operations of a cohesive program built for a narrowly defined niche (from here on out, I’m going to refer to all delegated underwriters as MGAs. Just keep in mind that this includes different flavors, like program managers).

One reason E&S carriers outsource underwriting and binding authority to MGAs in the first place is that MGAs are often entrepreneurial specialists that build deep expertise in narrow pockets of risk that are not worth the resources for a carrier to develop internally. In the cannabis industry, for instance, dispensaries may have unique cash handling and security protocols while processing facilities may introduce unique fire and explosion hazards, with ventilation and suppression standards that vary by state. It is often faster and more efficient for carriers to access such risks through MGAs than to invest behind the infrastructure to do so themselves.

Another, more structural consideration is that carriers are limited by portfolio management considerations, which might prohibit further exposure to, say, roofers in a wildfire-prone region. An MGA, by contrast, is less constrained by its own balance sheet and can continue underwriting that business so long as it can place the risk with carrier partners that still have appetite and capacity for it.

MGAs and wholesale brokers play complementary roles in the E&S ecosystem. A wholesaler’s job is to place risk sent to them by retailers and sometimes those risks are so niche that, absent MGAs who specialize in underwriting them, they might never be placed at all. At a high level, then, MGAs function like distributors from the insurer’s perspective but look more like underwriters from the wholesaler’s.

From the point of view of US E&S insurers, direct premiums break down roughly as follows:

Only ~20% of E&S premiums are directly placed by retail brokers. Nearly of the other ~80% is intermediated through MGAs and wholesalers with and without binding authority.

Subscribe now

Read more

But as with the non-ratings and non-index parts of S&P, Moody’s Analytics has also come under scrutiny as a business vulnerable to AI disruption. Moody’s Investor Service, the ratings division of the company, accounts for 70% of total profits. But that still leaves 30% AI-exposed, so investors are concerned.

Moody’s Analytics is composed of three sub-segments: Decision Solutions, Research & Insights, and Data & Information:

In my S&P write-up, I explained:

Credit & Risk Solutions, the third subsegment of Market Intelligence, commercializes research and data related to S&P’s credit ratings, and provides tools to measure credit risk and analyze how sensitive a portfolio or company is to changes in credit conditions. It think about C&RS as S&P’s way of double dipping on its dominant ratings franchise. It realizes revenue once by issuing a bond rating, and again by selling detailed research reports justifying the ratings to banks, insurers, asset managers, and corporate treasury teams who invest in those rated bonds.

Research & Insights is the Moody’s equivalent. It monetizes the content that MIS generates in the producing credit ratings. Around 2/3 of its revenue comes from research reports that analysts write on rated companies, the remainder from credit scoring tools that clients use to spot credit deterioration, dig into the drivers of credit risk, and run scenario analyses.

Subscribe now

Read more

Wix was conceived to democratize the creation of websites at a time when doing so required access to developers. Its Wix Editor presented a drag-and-drop interface that let anyone arrange design elements into a professional-looking web presence, no code required. Over time, Wix expanded beyond well site design, adding capabilities – accepting payments, managing reservations, handling bookings – that transformed websites into business operating systems . Those capabilities were bundled into industry-specific solutions – Wix Hotels, Wix Restaurants, Wix Stores – and could integrate with proprietary and third-party marketing and productivity apps.

Wix also pioneered technology that was, at the time, genuinely cutting-edge and unique. In 2016, before the words “LLMs” and “vibe coding” were introduced to the public lexicon, it launched Artificial Design Intelligence (ADI), which based on the answers to a series of questions – what kind of site are you building? are you accepting payments? – would scour the web and Wix’s own proprietary library for relevant images and text, and assemble a finished website in minutes.

Several years later, Wix began climbing upmarket, courting the very web design and development agencies it had originally set out to disrupt. It created a workspace, Wix Studio1, for them to build reusable components and manage projects workflows, and even offered a lifetime revenue share on the subscriptions and payments of every client they brought to the platform.

But now, the fear is that AI is reshaping the competitive landscape and eroding much of Wix’s technical differentiation. The release of Claude 4.5 Opus last November appears to have marked a turning point in coding agent capabilities, sparking fears that users could trivially spin up on their own what they pay a dedicated software vendor for today at lower cost.

Though the marketplace for website builders has long been crowded, during the mid-2010s a handful of players broke out from the pack and began converging on the same strategy of being a one-stop shop for payments, productivity, CRM, and other operational functions. GoDaddy, whose roots were in domain registration, joined Wix and Squarespace in site creation and business apps. Shopify solved online selling better than the rest – not only enabling payments and online checkout but also fulfillment and discovery – and became the default option for independent e-commerce.

Around this time, a number of no/low-code web app builders like Webflow, Bubble, and Framer gained some notoriety as well, though these were generally thought too complex for the SMBs that Wix addressed. And more recently, AI-first builders like Lovable and Bolt have begun collapsing much of the creation process to a single prompt. Here, for instance, is what Lovable produces within minutes when I asked it to “create a website for booking haircut appointments”:

The bear case writes itself. AI has flattened whatever technical differentiation legacy site builders once had. With generative frontier models powering design, competitive parity is now just a thin UI away. But it’s worth considering how Wix might come through the other side of the AI revolution in tact.

Subscribe now

Read more

Moody’s and S&P Global have this year traded as a blended average of two different businesses: a deeply moated ratings franchise that is largely insulated from AI disruption, and a data and analytics operation that is squarely in its crosshairs. In the current AI panic, each stock has sold off roughly in proportion to its exposure to the latter.

I don’t think anyone is confused about what makes credit ratings a GOAT business. But just in case, here’s what I wrote about it back in 2017:

…ratings underpin the risk weightings that banks attach to assets to determine capital requirements, dictate which securities a money market fund can own, and, in ostensibly surfacing the credit risk embedded in fixed income securities, make it easier for two parties to confidently price and trade, enhancing market liquidity.

Because of such industry-wide adoption, a debt issuer has little choice but to pay Moody’s for a rating if it hopes to get a fair deal in the market: an issuer of $500mn in 10-year bonds might pay the company 6bps ($300k) upfront, but will save 30bps in interest expense every year ($15mn over the life of the bond)….and each incremental issuer who pays the toll only further reinforces the Moody’s ratings as the standard upon which to coalesce, fostering still further participation. This feedback loop naturally evolves into a deeply entrenched oligopoly.

Moody’s and S&P enjoy what I described back then as a “standards moat”. The durability of their credit ratings have less to do with the relative accuracy of default and loss assessments than by the role they play as a ubiquitously adopted benchmark across the financial ecosystem. A startup could conceivably use AI-powered algorithms to create a ratings methodology that was just as accurate and Moody’s and S&P’s would still remain entrenched.

A similar logic holds for S&P’s equity index business. From [SPGI, MCO] Market infrastructure: part 2:

Likewise, the utility of equity indices like the Russell 2000, FTSE 100, S&P 500, and The Dow Jones, is derived from consensus rather than from absolute standards of reasonableness. The Dow Jones Index methodology, which weights its 30 constituents by share price, is routinely mocked as non-sensical, but its absurdity is more feature than bug as it suggests that a more sensible index couldn’t just come along and displace it the same way that a better technology might, in theory, replace entrenched network effects.

The credit ratings business should be buoyed by several durable growth themes over the foreseeable future. Some $5tn of debt needs to be refinanced over the next four years, roughly double the dollar volume seen in 2018. The swelling stock of private credit will increasingly demand risk scores too, as regulators and investors press for comparability with public debt markets; Moody’s private credit ratings revenue grew 60% last year. Infrastructure funding, including AI data centers, supplies a further tailwind. Ratings revenue is lumpy and unpredictable year to year, but over extended periods has reliably tracked the high-single digit growth rate that management has long guided to. I don’t expect the future to look much different from the past.

S&P’s Indices segment, meanwhile, should continue to be fueled by market appreciation and the same two secular tailwinds – active-to-passive migration and thematic benchmark creation – that have powered double-digit revenue growth over the past decade, with a potential third growth kicker emerging in the form of private market index products.

Excluding the soon-to-be spun off Mobility segment, Ratings and Indices1 together account for ~60% of segment-level EBITA. The remaining 40% is split across Market Intelligence and Energy. The latter two segments are a mishmash of data and analytics, and both were dramatically enlarged through the $44bn acquisition (~26x EBITDA) of IHS in 2022 (for more, see my March 2021 writeup: Thoughts on the S&P Global / IHS Markit merger):

Market Intelligence is a portfolio of products that management divides into three subsegments. The first and largest is Capital IQ, which provides financial and market data – including deep sector-specific data related to banking, insurance, real estate, and utilities, picked up through the $2bn acquisition of SNL – that investment professionals and sell-side analysts use to research companies, build valuation models, check estimates, and track markets, either by logging into a dedicated Bloomberg-like terminal or by piping it directly into their own systems.

The second is Enterprise Solutions, software that banks and asset managers use to manage loan portfolios (Wall Street Office), monitor private company holdings (iLEVEL), and manage the process of structuring and trading syndicated loans (ClearPar and Debtdomain).

Credit & Risk Solutions, the third subsegment of Market Intelligence, commercializes research and data related to S&P’s credit ratings, and offers tools that measure credit risk and analyze how sensitive a portfolio or company is to changes in credit conditions. It think about C&RS as S&P’s way of double dipping on its dominant ratings franchise. S&P realizes revenue once by issuing a bond rating, and then again by selling detailed research reports justifying the rating to banks, insurers, asset managers, and corporate treasury teams who buy those bonds.

S&P’s Energy segment combines Platts and IHS Resources, and serves energy producers, refiners, utilities, trading firms, brokers, and basically any commercial entity that traffics in commodities. It publishes thousands of daily benchmark prices – across oil, fuels, gas, chemicals, food and other commodities – that are ubiquitously referenced in contracts and used to settle trades. It also sells news and analytics to explain supply and demand trends, and offers data and software that drillers use to find promising oil fields and estimate how productive those fields might be.

In response to investor fears over AI, S&P released this helpful exhibit, which I’m going to use as a prompt for thinking about where the company is most and least exposed:

Subscribe now

Read more

A lot has happened since I wrote up Veeva nearly five years ago. At the time, Veeva, the dominant biopharma CRM, remained locked in a bitter dispute with IQVIA, the dominant vendor of data related to healthcare professionals and organizations (OneKey). The conflict dates back to 2016, when IQVIA blocked pharma customers from uploading its data into Veeva’s data management and analytics products (Network and Nitro, respectively), in turn diminishing the utility of Veeva’s CRM. Veeva, in turn, blocked IQVIA from pulling data out of its EDC system, the software used to capture and manage patient data during clinical trials. Veeva sued IQVIA on anti-trust grounds; IQVIA sued back, alleging unlawful use of data. The clash frustrated pharma customers, who wanted to freely combine IQVIA’s data with Veeva’s software.

Each responded by building its own version of the other’s core functionality. IQVIA elbowed its way into CRM through various acquisitions….Incentia, 360 Vantage (which, like Veeva CRM, was built on force.com), Apparture, Cegedim, and finally another force.com-based solution called Orchestrated Customer Engagement (OCE).

Veeva, meanwhile, began building its own “Data Cloud”, composed of four major pillars:

1) OpenData: a continuously maintained dataset of healthcare professionals – contact details, specialties, and hospital affiliations – scraped from hospital websites, online directories, and other public sources.

2) Compass: anonymous, longitudinal data on US patient prescriptions, procedures, and diagnoses, used to segment and target HCPs and alert sales reps to changes in prescribing behavior.

3) HCP 360: expands on OpenData by adding behavioral and consent data – how do doctors want to be reached? how often do they participate in webinars? which medical topics do they engage with? – so that reps know who they are allowed to reach out to and through what channels.

4) Veeva Link: a “curated LinkedIn on steroids”, Veeva Link pulls data from PubMed, social media, hospital websites, and clinical trial registries to map out more than 1mn key scientific and medical influencers across different therapeutic areas and show how they’re connected.

Crossix, acquired by Veeva in 2019 for $550mn, is also part of the broader data estate – providing longitudinal datasets of prescriptions, claims, and clinical data on more than 300mn patients – but is more commonly treated as a marketing analytics layer that integrates with the other four data sets to measure media spend and target digital ad placement to life sciences audiences. For instance, it can link TV ad exposure to subsequent specialist visits and therapy starts, or connect a patient’s visit to a branded website back to their physician and trigger a notification to the pharma rep assigned to that doctor.

Data is often considered the lifeblood of enterprises, even stickier than the CRM and MDM it flows through, since replacing it means reworking countless downstream systems. But aggregating and curating it requires a different skillset than building software and IQVIA’s repeated attempts to create a rival CRM ultimately concluded on a low note in April 2024, when management agreed to license its OCE CRM software to Salesforce:

As part of the expanded partnership, IQVIA will license the OCE CRM related software to Salesforce, and the parties will collaborate to accelerate development of Life Sciences Cloud for customer engagement expected to be available in 2025. IQVIA will continue to market the OCE CRM product and will support its nearly 400 global OCE customers in 130+ countries through 2029. IQVIA will work with Salesforce to jointly market the new offering.

With IQVIA no longer competing with Veeva in software, both companies settled all lingering legal disputes and struck long-term partnership agreements, allowing the more liberal use of IQVIA’s data in Veeva’s software and vice-versa, reopening for Veeva commercial opportunities that had been curtailed by its lack of access to IQVIA's must-have datasets.

Yet, in retreating from CRM, IQVIA simultaneously breathed life into Salesforce as a competitor to Veeva while simultaneously positioning itself as the mission-critical data provider to both. Veeva, inversely, pitched its software as the center of gravity, with datasets – its own, IQVIA’s, and others’ – as interoperable complements that customers could mix and match at their discretion.

And the bargain was hardly symmetrical. Notice that while IQVIA abandoned its CRM ambitions, Veeva didn’t return the favor. It pushed on in data and made far more progress here than IQVIA ever made in software. Crossix, in particular, has been a standout, more than tripling its revenue since acquisition and now growing at roughly 30% annually, driven by surging demand for online audience targeting. Management believes it could eventually rival Veeva’s CRM business in size by 2030.

But even as Veeva worked to develop datasets in-house to reduce its dependence on IQVIA, its CRM was still not entirely its own, built as it was atop Salesforce Lightning. In time, Veeva created its own underlying chassis, Veeva Vault, which enabled all apps built on it to share data, content, and workflows, prefiguring an ambitious push toward a unified architectural foundation that could serve as a single source of truth from drug development through commercialization. Management began by migrating its R&D-related software to Vault, then in 2023 ported CRM to the platform as well. Freed from Salesforce Lightning’s restrictive architecture, Veeva could build out new commercial modules, like Campaign Manager (running and measuring digital marketing campaigns), Service Center (viewing interactions with HCPs), and Patient CRM (tracking patient enrollment and verifying benefits).

In short, Veeva historically faced two barriers to its commercial offerings. The first, IQVIA’s data use restrictions, limited its use industry-standard datasets. The second, an OEM agreement with Salesforce, stymied the development of certain applications. With both obstacles now removed, there is nothing holding Veeva back from reaching its full commercial potential.

Well, nothing technical at least. Migrating from Salesforce-supported CRM to Veeva’s re-platformed version, Vault CRM, is still a heavy and costly lift for customers. Veeva insists that clients who choose Salesforce are in for a riskier and even more burdensome transition, arguing that its rival’s life sciences CRM is an immature product with few reference customers that requires extensive, region-specific customization. But Veeva has already lost several large clients to Salesforce and expects to retain just 14 of the top 20 biopharmas when it previously claimed 18, suggesting that some clients prefer a more bespoke solution, are drawn to Salesforce’s extensive support ecosystem, or find themselves swept up in its exuberant AI messaging.

Apart from facing direct competition from a formidable rival, Veeva must also contend with the reality that CRM is no longer the growth opportunity it once was. It already commands more than 80% of biopharma CRM – closer to 70% once the Vault CRM transition runs its course – leaving little room to grow from share gains alone. Meanwhile, the clients it serves have trimmed their sales forces by nearly 10% over the past several years, and AI-driven productivity gains could spur even further reductions at some point.

Given that context, management is quick to downplay CRM’s importance to Veeva’s future. A stable, deeply penetrated offering with limited growth prospects, CRM is going being overtaken by Veeva’s other product lines, falling from 75% of total revenue a decade ago to just 20% today, and expected to reach 10% by 2030.

Veeva’s growth is increasingly powered by two other clouds – Development Cloud and Quality Cloud – which organize content and streamline workflows tied to drug development. Like Commercial Cloud (sales and marketing software) and Data Cloud (Veeva’s proprietary datasets), both are built on the Vault platform.

(Data Cloud may eventually have broader integrations with Veeva’s R&D apps but, for now, it primarily goes hand-in-hand with Commercial Cloud. In fact, Veeva reports Commercial Cloud and Data Cloud under a single reporting segment, Commercial Solutions, that comprises ~47% of total subscription revenue).

The above image belies the sprawl of Veeva’s offerings. Each of these 4 pillars contains suites and each suite contains applications, more than 50 in total, in varying stages of maturity. Wrapped cross it all is a business consulting service that helps clients integrate these components into a coherent life sciences cloud.

The sheer number of Development and Quality apps and the alphabet soup of acronyms assigned to them is enough to confuse even a diligent analyst. Rather than cataloguing each offering in isolation, it’s more instructive to walk through a typical clinical trial workflow and see where they slot in along the way (product names in bold).

Subscribe now

Read more

Having just written up Watsco, I thought I’d offer some thoughts on revisit Pool Corp. given the similarities between the two businesses.

Pool Corp is a distributor of pool supplies and equipment. It sources ~200k SKUs from 2k+ suppliers1 and sells them to 125k customers, who fall into one of three groups: service professionals, ~80% of Pool’s revenue, consist mostly of small, family-owned operators that maintain, repair, and install residential pools. Retailers make up another 14%, with about two-thirds coming from sales to roughly ~5k specialty pool stores and the the rest tied to Pinch-A-Penny’s 303 franchise locations, which sell products to homeowners and in some cases service their pools as well. Commercial customers – hotels, multifamily developments – represent the remaining 4 percent2.

(Pool also distributes irrigation and landscaping supplies through Horizon, which it acquired in 2005, as well as hardscapes and other outdoor living products through NPT. But seeing as North America Pool, including sales to Pinch A Penny, comprise 90% of total revenue, the pool business will remain the focal point of this post).

The market for wholesale pool product distribution is fairly consolidated, with about half the industry’s revenue coming from two players. Pool is the largest by far, with what I estimate to be 35% of a $14bn wholesale market (residential + commercial). Heritage Pool Supply – a subsidiary of SRS, a Home Depot-owned3 rollup of building products distributors that derives ~2/3 of revenue from roofing – is the other relevant national peer. While just ~40% of Pool’s size by branches and 33% by revenue4, it appears just as, if not more, ambitious, having opened as many stores as Pool from 2021 to 2024 on a much lower base and acquired a greater number independent distributors to boot. Besides those two giants, there are several regional competitors of note, like Baystate in the Northeast, and Horner and Gorman in the Florida.

As with Watsco, Pool’s scale confers two major advantages. First, its national heft allows it to negotiate more favorable product rebates than smaller competitors and to do so across a broader range of SKUs. Second, its dense Sunbelt branch network5 makes it the most convenient and reliably stocked option for pool professionals, who typically operate on a just-in-time basis and will often begin their mornings at a Pool branch to replenish supplies, sometimes even visiting several more times during the day as they uncover unexpected issues from one job to the next.

As well as being close by and deeply stocked with relevant products, Pool’s branches are also staffed by competent folks who understand the products, a critical consideration seeing as 70% of transactions take place inside a branch. Know-how is preserved through unusually long tenures, averaging around 7.8 years across all employees, more like 21 years at the management tier. Given how much speed and availability matter to pool pros, branch-level bonuses are tied to turnaround times at the counter and product stockouts. The combination of store density, volume-driven product rebates, and operational excellence yield excellent returns, with branches typically hitting mid-teens operating margins and ~25% ROICs within four years of opening.

Another noteworthy parallel with Watsco is that Pool doesn’t just supply contractors with product, but increasingly builds the software they use to run their businesses. That effort began in the mid-2010s with POOL360, which let pros search inventory and place orders online. In 2023, Pool expanded the software toolkit with POOL360 WaterTest, which generates water-treatment recommendations from pH, chlorine, alkalinity, and other chemistry readings taken by a technician or retail associate. WaterTest not only provides fast, consistent testing but also pulls through high-margin sales since its recommendations – surprise! – prioritize Pool’s private label chemical lines (Regal, EZ Chlor, Life).

Finally, a year later, Pool launched the most ambitious of its pro-facing tools, POOL360 Service, which is pitched as an “operating system” for pool-service operators, used to dispatch technicians, optimize routes, schedule recurring orders, provide quotes, invoice and collect payments, and so on. The idea is that by embedding itself in a service professional’s daily workflow, Pool reduces the incentive to shop elsewhere and captures a greater share of their supply spend. So, a typical job might look something like this: a service operator receives a request and dispatches a technician through POOL360 Service. At the customer’s home, the technician finds cracked filters and malfunctioning underwater lights. He runs a water test with a handheld photometer and uploads the results to WaterTest, which converts the readings into precise chemical prescriptions and dosages. Still within Service, the technician documents the work, generates an electronic invoice, checks the stock of filters, lights, and chemicals at nearby Pool branches, and places a replenishment order before leaving the property.

But though a dense branch network, skilled store associates, and software tools insulate Pool from competitive inroads, they do no immunize the company from the macro. Although Pool has long pitched itself as a predictable, maintenance-oriented business, with revenue largely sourced from recurring service work, the revenue and margin swings of the past five years offer a reminder that there is more going on beneath the surface. The degree to which Pool’s fortunes track the housing cycle becomes clearer when you look at the mix of work being performed at any given time.

In 2024, Pool’s revenue mix by application was as follows:

Maintenance and repairs – 64% (non-discretionary)

Renovation and remodel – 22% (somewhat discretionary)

New construction – 14% (discretionary)

Maintenance and repairs – the light upkeep required to keep a pool swimmable – is the least economically sensitive source of demand and new construction is the most, while R&R sits somewhere in between. Chemicals – the single largest product line for Pool, accounting for 13% of its revenue – along with filters, repair components, and other small-dollar items fall firmly within the maintenance category. But other products can land in different buckets depending on the reason they’re being bought.

Subscribe now

Read more

Remitly was founded on a crunchier premise. Its founder, Matthew Oppenheimer, was working on mobile banking initiatives for Barclays Bank in Nairobi when he experienced firsthand how inconvenient and expensive it was to send money back home to people who depended on remittances to meet day-to-day needs (”we send life-changing transfers”). Remitly began by facilitating remittances from the US to the Philippines in its first few years, then expanded to India and Mexico. More than a decade later, those original three destinations still anchor the business, accounting for roughly half of Remitly’s revenue.

Wise positions itself as a cold utility, optimizing for cost and speed. Remitly, by contrast, stresses the “emotional weight” of getting non-discretionary funds to their destination on time. Saving a buck or two on a $200-$300 transfer matters less than ensuring mom and dad have the money to pay for food and medicine. As Remitly’s Chief Business Office recently put it: “When you deliver the most trusted, reliable, fast, simple and delightful experience, price becomes just one part of the value equation. Customers don’t stay with Remitly because we are the cheapest. They stay because we are the most trusted, transparent and fair.”

Trust. Management repeatedly points to various proxies for this north star…4.8+ stars on Google Play and iOS, a Trustpilot score of 4.6 with over 5mn reviews, 94% of transactions completed within an hour, 63% instantly. And having proven itself in low-value, emerging-market remittances, Remitly is expanding in several different directions at once.

It has been gentrifying toward higher-value senders – $1,000+ transfers in select corridors like US-to-India – whose share of total volume has risen 2-3 points over the past year to roughly half of all monthly send volume. Large transactions tend to generate more lifetime value than small-dollar transfers, even at lower take rates and less predictable transaction patterns, as they exhibit lower fraud rates and skew toward digital disbursements and away from expensive cash pickups. The dilutive effect of high-value transfers explains why management rightly discourages take rate as a KPI. Realizing 0.5% on a $10,000 transfer is plainly superior to earning 3% on $500.

Remitly has also started targeting freelancers and micro-enterprises, a lower-churning segment that overlaps heavily with the core customers that Remitly already serves but that transacts at twice the size and frequency.

Alongside its move into new customer segments, Remitly launched a burst of new products last year:

Remitly Wallet: a multi-currency stored-value account where users can hold fiat and USDC.

Remitly Card: a debit card linked to Remitly Wallet that allows users to spend balances through Apple Pay or Google Pay (physical cards not yet available, but coming this year) on everyday purchases at home or abroad.

With Wallet and Card, consumers can store value across currencies, convert them as needed, and spend locally without having to open separate bank accounts in each country. Remitly earns revnue on Wallet the same way it does on any cross-border transfer, by taking a slice of converted volumes, and monetizes Card through interchange fees.

Remitly Flex: a send-now, pay-later cash advance, no interest cash advance designed to smooth cash-flow gaps so customers can help recipients cover urgent needs. Remitly carries these receivables on its balance sheet (for now) and limits credit risk by capping advances to $250, setting short repayment windows, and offering the product only to “trusted senders” with established remittance histories. Since the program’s inception, charge-offs (120+ days past due) have been immaterial and 90+ delinquencies are negligible, though it’s worth noting that roughly 10% of Flex’s balances are non-current:

Remitly One: a subscription-based membership that management likens to Amazon Prime, designed to encourage habitual use of multiple products by offering perks across them. For example, Remitly One subscribers earn rewards on Wallet balances, including a 4% annual boost on USD holdings. They can repay Flex balances over 90 days instead of over 30 and their Flex recipients get instant access to funds rather than having to wait three days.

In its recent Investor Day, Remitly broke down the unit economics of a Flex transaction from a Remitly One subscriber, which works out to a take rate of 6.2%, compared to 1.5% for the core consumer business. After taking loan loss provisions into account, management claims the take rate is still “at least double” the core.

(”Illustrative allocation of $9.99/month Remitly One membership fee to Flex, after other benefits; 1 Assumes 3.5% take rate, 2. Assumes 35% of cross-border revenue, 3. Assumes 10% of outstanding balance, 4. Net take rate = RLTE / Send Volume, 5. ROIC calculated as net take rate dollars associated with the Flex product as a percentage of average outstanding Flex balances”)

The fanfare around these budding initiatives and Remitly’s ongoing success growing customers and send volumes have been overshadowed by persistent fears over the impact of stablecoins. Management has tried flipping the narrative, framing stablecoins as a technology that reinforces, rather than disrupts, Remitly’s offerings.

They classify stablecoin use cases into two categories: Infrastructure and Wallet. The Infrastructure use case has primarily to do with treasury management. One annoying problem that Remitly faces is that it operates 24/7 but its bank partners do not. If Remitly expects strong demand for PHP on Saturday afternoon, it needs to make sure accounts at its Philippine bank partners, who are closed on weekends, are funded with enough pesos by 5pm Friday (or whenever the settlement cutoff time happens to be). But in doing so, Remitly is shouldering the risk that the USD/PHP exchange rate moves against them between when they pre-fund PHP in advance of weekend demand and when they must replenish the PHP currency pool when the FX market reopens. As Remitly explains in its 10K:

To enable disbursement in the receive currency, we prefund many disbursement partners one to two business days in advance based on expected send volume. Foreign exchange rate risk due to differences between the timing of transaction initiation and payment varies based on the day of the week and the bank holiday schedule; for example, disbursement prefunding is typically largest before long weekends.

It can hedge this risk or keep credit facilities on standby to meet transfer demand, but neither of those options are free and the more illiquid and volatile the local currency, the more acute these issues become. In certain corridors, stablecoins give Remitly a way to move value instantly between markets, even over the weekend when most banks are closed, and convert to PHP much closer to when payouts actually happen.

Subscribe now

Read more

In the world of property & casualty insurance, “admitted” lines are more like index funds, characterized by homogenous, well-understood risk parameters that are underwritten over and over again at massive scale. They fit within the investment equivalent of “style boxes”. Carriers file standard policy forms with regulators and their policies are backed by state guaranty funds. The auto policy on your Honda Civic falls in this category. So might, say, a small business policy for an Italian restaurant in a non-cat zone.

By contrast, “excess and surplus” (E&S) lines assume liabilities that are hard to place in the admitted market due to some quirky hazard or complexity that requires unusual terms or exclusions…that might include, for example, a sketchy nightclub; a cannabis dispensary; a contractor working on a high-rise exterior near wildfire-prone areas; a Florida condo association overseeing buildings with unresolved structural issues. Regulatory oversight still applies, but the rules are looser. Rates aren’t pre-approved, forms can be customized, and coverage can be tailored by carving out the ugly bits, demanding higher deductibles, or imposing tighter limits.

There is no hard and fast distinction between admitted and E&S exposure; as certain risks become better understood and more actuarially tractable, they often migrate from the E&S market into the admitted side. Cyber risk and ridesharing endorsements are two such examples. Premiums flow the other way too. Where E&S was once considered a market of last resort, it has increasingly become a primary destination for risks that may have historically been placed with admitted carriers. You see this happening with homeowners insurance in catastrophe-prone states like California, Florida, Texas, and Louisiana. Admitted insurers, constrained by regulated rate filings, are unable to reprice fast enough to respond to exacerbating loss trends, forcing them into retreat. E&S carriers, who enjoy rate and form flexibility, are stepping in to fill the void.

Management points to a 30-year history of E&S outgrowing the admitted market and, according to A.M. Best, E&S premiums as a share of P&C premiums has more than doubled over the last 20 years, from 4.3% in 2001 to 10.1% in 2021. Within commercial lines, where Kinsale predominantly plays, E&S share is even greater:

Source: Ryan Specialty 2024 10-K

Founded in 2009 by CEO, Chairman, and ~4% owner Michael Kehoe – who previously ran James River Insurance, a competing specialty carrier, from 2002 to 2008 – Kinsale Capital focuses on the less competitive end of the commercial E&S market, serving commercial small/mid-sized accounts1. In 2024, ~2/3 of its gross written premiums came from casualty lines, ~1/3 from property.

Insurance is a commodity service where the cheapest quote usually wins. As such, you generally want to bias your investment search to the lowest cost players (see my write-ups on Admiral and Protector Forsikring). But determining which carriers enjoy a sustainable cost advantage isn’t necessarily straightforward because, as Buffett once quipped, insurance accounting is like a self‑graded exam. Carriers estimate their own loss reserves and, whether due to incompetence, wishful thinking, or fraud, can make a book of business appear more profitable than it actually is for a while.

This is particularly true of “long tail” risk, where claims may be paid out many years, even decades, after a policy is written. Asbestos exposure is perhaps the most canonical example of a long-tail catastrophe. Workers exposed to asbestos from the 1940s through the 1970s fell afflicted with causally related diseases, like mesothelioma and lung cancer, decades later. I remember seeing late night TV commercials recruiting plaintiff for class action asbestos lawsuits in the 1990s. Claims are still being filed today.

But there are more garden variety cases. A general contractor might find themselves on the hook for construction defects that don’t show up until five years after the job is completed. Or a radiologist could be sued for misreading a scan of what years later turns out to be cancer. Actuaries set aside some percentage of premiums as reserves to account for such scenarios. But it’s hard to predict the magnitude of those risks so far in advance. Medical costs, legal fees, and jury awards could be far higher when claims are ultimately paid. Expanding theories of liability and rising jury sympathies (sometimes referred to as ”social inflation”) can dramatically lift claim costs over time.

Generally speaking, for most casualty line (not stuff like toxic torts and environmental liability) the more time that passes after a policy is written, the less likely surprise reserve developments are to materialize, as an ever growing proportion of ultimate losses will have already been reported and adjudicated. A chronic abdominal pain that a patient experiences today is unlikely to trace back to a gallbladder removal from 5 years ago, as complications from that procedure would almost certainly have manifested much earlier.

Short-tail lines are vulnerable to sudden adverse events too, but claims typically surface and resolve quickly. When a hurricane rips your house apart, trust me, you will know immediately and won’t wait 5 years to report the event. It’s the lingering uncertainty that makes long-tail casualty so insidious. Injuries take time to manifest, liabilities unfold slowly, and claims can deteriorate long after the policy period has closed.

No approach to assessing reserve adequacy on long-tail lines is foolproof. But two commonly accepted methods provide useful signal.

The first involves looking at loss development triangles. A loss triangle, as featured below, shows how the insurer’s estimate of ultimate losses2 for policies written in a given “accident year” changes over time. If an insurer is reserving accurately and conservatively, subsequent adjustments to a given accident year’s initial loss estimates (initial “loss picks”) should be modest, and when changes do occur, they should skew downward, reflecting a better-than-expected loss experience that allows some portion of reserves to be released.

Subscribe now

Read more