AI and exit value · Answered by Golden Section from more than 400 B2B software companies observed
Yes, but only data of a particular kind. Proprietary data protects a company when it is embedded in a workflow customers run every day and carries context a general model cannot get from a cold start: years of industry transactions, compliance records, integrations and the reasons behind each exception. Data that sits in a warehouse and could be exported, bought or scraped is an asset, not a moat. The test is whether AI on your data produces a better outcome, in the customer's own units, than AI without it. If it does, build AI into that workflow and price the outcome. If it does not, your defense is the workflow and the trust, and rebuilding around AI will not create a moat; a rebuild is right only in extremely rare cases. Start by listing the decisions your data makes better than a general model could.
Treat data as a moat when AI running on it beats AI without it on a measurable customer outcome and the data refreshes through daily use. Extend the product you have toward that outcome rather than rebuild it (a rebuild is right only in extremely rare cases), and keep a small team working on what could replace the core.
| Metric | Value | What it means | Source |
|---|---|---|---|
| Vertical companies that froze, Q2 2026 | 0 in a 43-company review | public software companies reporting a budget freeze in net-new and expansioninsulation held across vertical names, but only those that priced AI as a product were re-rated | Golden Section, publishedWhere the AI Budget Lands |
| Where AI builders are aiming | about 70% focused on vertical AI; 43% of AI products are vertical applications | share of AI builders and products, ICONIQas cited in our research: the AI-native attack is aimed directly at vertical software, so the vertical label alone protects nothing | Golden Section, publishedInvesting in Software addendum, September 2026 |
| AI-native retention | GRR 40%, NRR 48% | median gross and net revenue retention, 200 AI-native companiesChartMogul 2025 panel as cited by Golden Section, against 82% net retention for B2B SaaS in the same panel | External benchmarkChartMogul 2025, via Investing in Software addendum |
| Premium for AI that shows up in revenue | about 2–3x revenue multiple | AI-validated names over AI-exposed laggardsServiceNow near 7x revenue against Pegasystems near 2.5x, Q2 2026 | Golden Section, publishedWhere the AI Budget Lands |
| Engineering and product allocation | about 45% core / 45% AI-native features / 10% skunkworks | share of engineering and product resourcesour suggested split for incumbents; the 10% builds what could cannibalize the core | Golden Section, publishedThe Enhancement Doctrine, March 2026 |
When the interface becomes natural language, a moat built on a well-designed screen gets shallow quickly, and a model can rebuild a workflow wrapper in weeks. What it cannot reproduce from a cold start is a decade of transactions in one industry, the compliance logic around them and the integrations that keep them current. That is the argument of The Enhancement Doctrine, and the Q2 2026 earnings season supported half of it: vertical companies were insulated, but only those that turned AI into something customers buy were paid for it.
So a data moat has to be proven, then sold. Start from the customer ROI you already deliver and find the decision your data improves. Run it through product management so the plurality of customers, not the loudest one, shapes it, and give AI its own swim lane on the roadmap. Then price it with the AI pricing play. A rebuild is right only in extremely rare cases, and almost never for a company with real data, because it spends years recreating the workflow that holds the data in place.
This company is invented. A vertical software company at $4M in annual revenue runs scheduling and compliance for home-health agencies and holds nine years of visit records across 300 customers. The founder is weighing a full AI-native rebuild. Instead, the team tests one question: can a model using its visit history flag missed-visit risk better than a general model given the same day's schedule? On 20 agencies over a quarter, its version catches meaningfully more missed visits before they happen. The company ships the risk flag inside the existing scheduling screen, prices it per agency, reports it as an AI revenue line, and gives two engineers a skunkworks brief to build the agent that could someday replace the scheduler.
A product whose value is an interface over public or commodity data has no data moat to defend; there a deeper rethink is the honest answer, and this is one of the extremely rare cases where a rebuild can be right. Data that contracts or regulation prevent you from using to improve the product cannot be the moat either, however large it is.
From the Golden Section mistakes list, each paired with the play that prevents it.
An AI feature built without subject experts and customers rarely finds the decision your data actually improves.
Selling an AI roadmap as a moat before it works is selling phantom product, and customers remember it at renewal.
A company that treats its data as protection and does not adapt the product leaves the opening to AI-native entrants.
In the order we would run them. Each is on its own page, most with a free Excel template.
Names the monetized outcome the data has to improve before anything is built.
Keeps the AI work tied to the plurality of customers and to industry expertise rather than to the loudest request.
Gives AI its own lane alongside core and reliability work so the split stays deliberate.
Turns the data advantage into a priced product, which is what the market paid for in 2026.
Puts AI revenue and AI-account retention on the dashboard so the moat has a number.
Development plays Product management, roadmap and engineering plays for building AI into the product you already have.
Start from the outcomes customers already pay for and list the decisions inside them that your data could improve. Give AI its own lane on the existing roadmap, attach a measurable customer outcome to each item, and keep the split near 45% core, 45% AI features in existing workflows and 10% skunkworks. Review it quarterly with the rest of the roadmap, not in a separate process.
Proprietary data that refreshes through daily use, a workflow customers cannot pause, regulatory and integration depth, and AI priced against an outcome the customer can name. AI-native companies grow fast but, in the panel data we cite, retain far worse; defensibility shows up in retention, not in the demo.
Almost never. In our view a rebuild is extremely rare, and rarer still if you hold real workflow data. A rebuild spends years recreating the workflow that keeps the data in place, while AI-native entrants attack your customers. Extend the product toward the outcome AI improves, and fund a small skunkworks team to build the version that could replace the core, so you are the one who does it.
Extending a real data position into priced AI features is a structural product change that takes quarters to show in revenue, which is the kind of change minority equity is built to fund. We invest in vertical companies whose data and workflow AI makes stronger.
Growth equity →Reviewed by Dougal Cameron, CEO & Co-Founder on 2026-10-22. Golden Section observations are labeled separately from external benchmarks and illustrative arithmetic.