Contents · 7 min read
01 / 05 · The data
When years of data are still only exhaust
In short
Years of operating data can look like a strong moat. But stored records create no new edge on their own. First ask whether the firm has built them into a usable asset.
A hospitality point-of-sale incumbent held years of sales and payroll data from customer work. Top chains ran on its product. Yet those records remained operating exhaust. They did not improve its next choice.
The firm had real strengths. Top chains ran on its product. Its software sat inside their daily work. Its staff knew the trade well.
Yet the firm did not build those records into a structured asset. The hospitality point-of-sale case says it treated them as operating exhaust. The data came from running the product. It did not help the product learn.
For the POS firm, the problem was not a lack of data. It lacked a system that turned past use into a better next choice.
That gap matters more when rivals can match the product itself.
02 / 05 · Feature gap
Why owning more data no longer protects you
In short
AI makes software features easier to match. A large store of old data does not stop that. Ownership helps only when the firm turns the data into an edge customers can feel.
Shipping more features still matters. Customers need the core work to run well. New features can also close real gaps.
For the POS firm, that work once bought time. A rival had to match years of product work. It also had to learn the trade.
AI cuts that lead. As AI for B2B SaaS explains, rival software now costs less to build. A funded entrant can match features much faster.
The incumbent still has more records. It still has more use history. But a rival can copy what the screen does without that past.
The vertical software CEO briefing therefore puts domain knowledge ahead of feature count. The useful part is what the firm learned about customer work. The records are only the source.
For the POS firm, those records may still help. The next test is whether the firm may use them inside the work.
03 / 05 · Permission
When does the data become hard to copy?
In short
The data becomes hard to copy when customers permit its use inside their work. Possession is only the first test. Each permitted use must also improve the next choice.
Run the compounding moat test.
Possession
Does the firm hold records from real customer work? For the POS firm, the answer is yes. Its system has years of operating data.
Permission
May the firm use that context inside the workflow? A record can sit in the product without being open to a new use. The firm needs a clear answer before it builds.
Learning
Does each use improve the next choice? Reading the same record again does not deepen the moat. The product must keep what it learns.
For the POS firm, the first yes is real. Yet it does not settle the other two. Data volume cannot make up for a missing customer yes.
The place where the asset sits matters too. The vertical software data briefing warns that a dataset built on a vendor system may serve the vendor. The firm needs to keep control of the asset it builds.
Permission gives the POS firm the right starting point. It must now make each permitted use teach the firm something.
04 / 05 · Learning loop
How does each exchange improve the next choice?
In short
Capture the signal in each customer exchange. Keep what the firm learned, then use it in the next choice. Over time, that history becomes firm knowledge a new entrant cannot copy.
The useful signal does not come from the product alone. Support calls show edge cases. Sales calls show what buyers ask for. Customer teams hear needs that users never put in a ticket.
The vertical software CEO briefing treats each exchange as domain signal. The firm can give that signal a clear form. It can then use it to train its own system.
For the POS firm, a support note may show why an order flow failed. A later case may show the same issue at another site. The firm can keep the reason and use it in the next product choice.
This also keeps knowledge inside the firm. Much of the real edge may sit with senior staff. They know why the data model works. They know which requests point to a wider need.
If that reasoning stays in their heads, it can leave with them. Your Features Are No Longer the Moat shows the same risk across B2B software. Customer knowledge often stays personal when the firm should hold it.
For the POS firm, the asset is now clear. It is the history of customer work and what the firm learned from it. An entrant cannot copy exchanges it never had.
That leaves the CEO with a product choice.
05 / 05 · The priority
Make the learning loop the product priority
In short
Put the permitted learning loop ahead of another race to match features. The CEO should set this aim. Each team then needs an owner and a test.
Shipping features is still useful. It keeps the core product fit for use. It may also answer a real customer need.
But feature speed is a weak place to compete. An AI-native rival began with a newer build. It can often ship faster.
The other path is deeper customer knowledge. The firm keeps what each exchange teaches it. It then makes that knowledge part of the product. The two paths for B2B software lead to very different firms.
For the POS firm, the feature race cannot repair years of missed learning. It needs to turn current use into a better next choice. Each customer can then add depth a rival must build from scratch.
Five steps put that choice into the product plan:
CEO: set the priority
Put customer learning ahead of matching every rival feature.
Product lead: map the work
Show where customer context enters each key workflow.
Customer lead: mark permission
Record where the firm has a clear customer yes.
Data lead: build the memory
Keep the signal and the reason behind each choice.
Leadership team: test the loop
Ask how each release improves the next customer choice.
The board should track the same shift. Feature count shows what shipped. Firm-owned customer knowledge shows whether the moat grew.
For the POS firm, the aim is no longer the largest dataset. It is a product that earns the right to learn from real work.
The full vertical software briefing shows what that choice means for the firm.
Build the learning loop
The CEO’s task is to turn customer work into a permitted learning loop that improves the next product choice. The vertical software briefing shows why feature speed protects less as AI lowers build costs, and how customer exchanges can become domain knowledge the firm retains. Use that industry view to decide where the loop belongs in the product plan and what the leadership team should fund.
What AI changes in vertical software, written for the CEO. Free to read, no form.
Questions CEOs ask about this
Does proprietary industry data still protect a vertical software company when every competitor can add AI?
Yes, a vertical software company can still have a data moat. The CEO must fund the permitted learning loop now, or its data stays unused while rivals match features. Customers must allow their work context to be used. Each use must then improve the next choice. The decision is whether both conditions hold before calling the data a moat.
Which data actually creates defensible customer value?
Data from real customer work matters when it improves a later choice. Sales records alone are not enough. The edge comes from linking use history to what the firm learned.
Can foundation models make our existing moat less valuable?
Yes. They make rival software faster and cheaper to build, so a long feature history protects less. They do not give an entrant the customer exchanges it never had.
Do workflow permissions matter more than data volume?
Yes, when the large dataset cannot be used inside the work. A smaller set of permitted context can help the product learn. A large store without permission stays still.
How should we turn customer data into a compounding product advantage?
Capture each customer exchange as domain signal. Keep the reason behind each choice, then use it in the next one. Store that knowledge inside the firm so it survives staff changes.
What data uses require explicit customer trust and consent?
The evidence for that split is not available in this pack. This piece makes no claim about which uses need explicit trust or consent.
Founder, GRAIL
Johan has spent fifteen years in management consulting and executive leadership, including as Chief Product Officer of a SaaS company. He founded GRAIL to help European leadership teams work out what AI changes at the core of their business. He is the author of The Execution Revolution and the host of the ThinkRoom podcast.
Related reading
Sources
- AI in Vertical Software: why the feature race is hard to win and how customer exchanges can deepen the moat. (accessed October 8, 2026)
- What's Changing in the Moat You Ship: the hospitality point-of-sale case and the gap between an asset and operating exhaust. (accessed October 8, 2026)
- Your Features Are No Longer the Moat: the choice between feature speed and deeper customer knowledge. (accessed October 8, 2026)
- What's Changing in Your Data Story: why the data asset should stay under the firm's control. (accessed October 8, 2026)
- AI for B2B SaaS: why lower build costs weaken the feature moat. (accessed October 8, 2026)
Next scheduled review: February 9, 2027