2.13.26: Three dominant models of today’s vertical AI horserace

the familiar path, roll-ups, & industry disruptors...

Hello!! Happy almost long weekend. This week’s musing is last week and this week rolled into one, so it’s a bit longer of a take. It’s been a while since I’ve done a good ole vertical software deal, but I’ve been thinking and closely watching the space evolve, so wanted to put down some thoughts on how I see the market today evolving. So without further ado. And as always, love folks thoughts here!

---

As public SaaS market caps continue to free fall, the VC market continues to roar, investing in vertical and horizontal AI applications at frothy valuations. In particular, there has been a ton of focus on Vertical AI as the next frontier, particularly as horizontal models continue to improve and fears of AI eating everything move from hypothetical fear to imminent reality. I took a step back as a long term vertical software investor to take a feel for the market. It feels to me that there are three predominant business models - vertical du jour- at this moment in time. Each represents a different approach to capturing value in this brave new world. The three dominant approaches are AI-enabled SaaS (I know it’s a dirty word, but that is really what it is), AI Roll Up, and AI-native replacements.

Model 1: AI-enabled SaaS - The Familiar Path

What’s the play? Sell AI-powered software to incumbent businesses using a fairly traditional and repeatable sales, PLG, or combo motion. I think of this as “next-generation SaaS” and the most legible to the broader market. It’s taking a proven vertical software category and supercharging it with AI capabilities. The argument why it’s better than just your plain ole vSaaS company is your driving dramatic operational efficiencies and overtime replacing labor (oftentimes outsourced labor).

Data Dynamics: You are dependent on customer data that you don’t own. The AI improves by processing proprietary information. Each deployment might be isolated given contracts and competitive concerns, which over time can limit ability to build truly differentiated moat. You of course can build generalized models, but most valuable insight (should at least!) stay within each customer’s instance. Thus, there’s an inherent ceiling to the “data”, value really is accrued at the workflow layer.

Strengths of Model:

• De-risked GTM: You aren’t really re-inventing the wheel. It’s the same GTM as SaaS was. Buyers understand procurement, have existing budgets (in most cases), and it’s a predictable sales cycle (albeit we’re seeing it accelerate in most cases).

• ROI Story: Reducing costs, driving revenue, improving efficiency easily translates to the buyers. There is now a pull to replace headcount with AI, so spend and budgets are becoming larger and the ROI equation should look different than SaaS iterations of the past.

• Predictable Scaling: once the initial GTM is cracked, you should be able to scale fairly predictably.

• Lower Regulatory Compliance: you operate as a vendor. Not a regulated entity (e.g. selling to a wealth management company vs. becoming a wealth manager yourself)

• Easy for VCs to understand: It’s the next iteration of SaaS. VCs understand it very easily and the funding path is straightforward.

Potential Weaknesses:

• Innovation Ceiling- you’re bound by the confines of your customers in many ways. Customers want to make existing processes more efficient vs. reimagining the way they are doing entirely. This could obviously change over time, but for now I think this still holds true.

• Build vs. Buy- as AI “eats” more and more software and it continues to be easier to DIY, there is a risk that differentiation erodes, commoditization increases, and ultimately customers may want to build themselves.

• Displacement risk- there’s a risk that reimagining space overall makes your end customer at risk (e.g. reimagining a brokerage business vs. selling to a brokerage business).

• FDE dependent model- Oftentimes, this software is custom and quite implementation heavy, so FDEs have become the way to scale. This can work, but its expensive and can be difficult to scale.

What’s an example of this? Any software players selling into the industry so think Fulcrum in our portfolio who sells to brokerages or Harvey selling to law firms.

Model 2: AI-Roll Up- PE gets a facelift

What’s the play? Largely popularized by a few larger firms, but the TLDR is you acquire businesses (usually small) and deploy AI to improve unit economics across a portfolio. You essentially buy revenue and data to then use AI to create operational leverage that the businesses individually couldn’t do.

Data Dynamics: Data almost certainly lives in legacy systems and or in tribal knowledge, but the beauty of this model is YOU own the data- every transaction, customer interaction, operational know how is in your control. In theory, this should create a very real compounding data advantage. Competitors can’t access this level of a unified dataset - at least not in the same granularity. BUT this data is messy and highly consistent. It’s not being formatted and piped into your vertical AI software as model 1. Standardizing data collection, cleaning historical records, and building unified schemas is often the biggest challenge. Not to mention all of the operational complexities you inherit in this model.

Strengths of Model:

• Immediate revenue and scale- you are buying existing cashflows and customer base, not building from ground 0. In theory this should enable you to grow non-linearly as you acquire more assets.

• Proprietary data/product flywheel- Because you are living the problem, you should have access to the most granular data and pain points to quickly optimize. Your product(s) should be better than off the shelf products bc the goal is to drive up margin and optimize quickly.

• Defensible Moat- this isn’t an easy playbook to duplicate IMO, but if you can successfully scale it, you have very real defensibility. You should have network effects, real operational scale, and proprietary data. If there’s a physical component there’s an added level of difficulty to compete against.

• Margin expansion story- if you are able to successfully improve margins by 30-40%, you’ve created very real enterprise value.

• Multiple arbitrage opportunities- the belief on this one is that if you can truly expand margins and prove out an AI-efficiency thesis, you may be valued like a tech company and at a larger scale than just a PE roll up. That being said, this has yet to be proven and is definitely an assumption investors in the approach are making.

Potential Weaknesses:

• Valuation/End State ambiguity- we just don’t quite know how these models will shake out because they are relatively new for venture. The question is if over time, you get valued as a tech company or a roll-up (EBITDA multiple usually lower than a tech revenue multiple). Upside is obvious, but downside is your just valued like a roll-up and then your VC valuation is all lopsided relative to what market will price you at.

• Capital intensity + debt as vehicle- just as this is operational complex, this is more financially complex. You need more capital to finance assets and likely are going down a combo debt + equity path. It’s not just as straight forward as hopping on the VC capital wheel.

• Integration & operational complexity- The difficult part isn’t just the AI, it’s about standardizing operations, integrating all of the systems or rip/replacing them, cultural alignment, people management, and maintaining quality/consistency across the assets.

What’s an example of this? Any roll ups so think Titan for ISPs or EqualParts for brokerages. Basically anyone rolling up industries.

Model 3: Become the thing to do the thing- AI disruption play

What’s the play? Build a fundamentally new version of insert vertical business here. Use AI to deliver a better product than incumbents with 10x better economics and customer experience. No selling software to incumbents or rolling them up, you are fundamentally re-envisioning

Data Dynamics: You’ve got a cold start problem in a way that the other two models don’t. You aren’t inheriting data or building on top of someone else’s data, so data generation should be a real focus whether its partnerships with data providers, incentivizing early adopters to contribute data, etc. Once you can get the data flywheel going, you own 100% of the data AND you can architect from day one the way you want it to be. Unlike model 1 where data is fragmented across customers or model 2 where you are dealing with messy legacy data, your data can be clean, purpose-built, and unified. You just need to reach a scale where this matters.

Strengths of Model:

• Maximum Value Capture- you own the full stack baby, no splitting economics. It’s all yours for the taking.

• Compounding Advantage- every customer you get and every workflow you improve should compound and benefit your system. You aren’t selling it to someone else or improving an existing process, it’s all ground up so advances should compound quickly.

• Unconstrained by legacy processes- No retrofitting needed. Your designing from first principles. Not what “should” be in place.

• Disruption of Incumbents- It’s the classic Kodiak example. These “new” models in some way are the toughest to compete because incumbents would have to disrupt their own business model.

• Valuation Upside- In theory, this model of the 3 should be valued the highest. You are building the next-generation technology leader (think Stripe vs. legacy payment processor).

Potential Weaknesses:

• Cold start problem- pretty self explanatory, but you are starting from scratch building everything, both product and GTM.

• Regulatory hurdles- in regulated industries, you’ll need to get licenses and certain compliance which can take a long time and/or require significant capital reserves. Also, in many industries, regulation varies from state to state.

• Market education/trust- You’ll need to get end customers to trust a fully AI-native approach, particularly if you are building in a high stakes domain like finance or healthcare. There’s some level of market education that will need to be done.

What’s an example of this? Examples of this are like a Corgi Insurance, WithCoverage (which just announced a round), or some of these newer players who are creating AI-native hedge funds.

So which model is best?

The reality is it really depends. It depends on the vertical your going after, the timing, the competitive landscape, and what you are uniquely positioned to excel at. I use it often, but what is your unfair secret? That should dictate which model you go towards. Broad strokes:

• Model 1 (selling software) makes a ton of sense if the industry has entrenched buyers with budget, existing vendors are weak or outdated, and AI provides clear leverage without requiring a whole new business model. Also, if you are exceptional at sales DNA, this could be a great model for you.

• Model 2 (rolling up) makes sense if you are playing in highly fragmented market with unsophisticated (usually smaller) operators and you have access to large sums of capital or you believe your an exceptional fundraiser. This makes sense if you believe scale + AI creates defensibility that pure software can’t achieve and you have the very real finance + operational chops to get a bunch of businesses all moving in one unified direction.

• Model 3 (ai-native player) makes sense if incumbents have structural advantages that your AI-first approach can eliminate, there are some regulatory or other tailwinds, and you have access to large amounts of capital for what will presumably be a longer build. The old model is fundamentally broken and you believe the new model will fix it.

If it doesn’t feel like I’m planting a flag on one winning model…well that is intentional. I just don’t believe one model will win. BUT I do think many companies are choosing a model whose physics don’t match their market.

Picking the right model is a critical starting point and you certainly can’t apply the same metrics, capital strategy, or expectations across them as they are fundamentally different businesses.

If I had a crystal ball, this is what I would predict…

Model 1 will produce breakouts per category, but many will converge toward SaaS-like end states or get subsumed by AI labs or horizontal players..

Model 2 will create some interesting and valuable outcomes, but forcing venture-style expectations onto what is partly an operational and financial engineering play will create very real friction. This model demands real operational expertise and so a software only mindset I don’t think will translate quite as many think it will.

Model 3 will generate the largest winners, but also will have the highest casualty rate. The upside is enormous, but so is the execution risk.

Vertical AI isn’t just one playbook and in today’s market, I am thinking about it as three. Founders who understand the differences and build accordingly can be the ones who actually capture the value.

And with that a tune to cap us off…

Stay weird. Stay curious.

-CBR