The Last Moat in B2B SaaS: What Your Customers Tell You That AI Cannot

Your customers are your moat.

I asked a room of B2B marketers last week what their company has that a competitor could not copy by Friday. Same product category. Same access to Claude and ChatGPT. Same Gong recordings of their own deals. Same playbooks scraped from LinkedIn. What is left? The silence ran longer than usual. Then one of our team in the back said, “Our customers.” That answer is the entire argument I want to make in this piece.

There is a small experiment that has been going around AI circles, and it has stuck with me. Researchers asked every major model to pick a random number between one and ten. GPT picked 7 ninety-two times out of a hundred. Claude, ninety. Gemini, all hundred. Humans pick 7 about a quarter of the time, three times what chance would predict, because the ends feel too obvious, the middle feels too neat, and 7 sits in the sweet spot of feeling spontaneous without being so. The models learned that from us. Now they do it with more confidence than we ever did.

That is a small portrait of what is happening to B2B SaaS marketing right now. Every team is reasoning on the same handful of models, trained on overlapping data, optimized against similar benchmarks. Your competitor’s ICP doc, persona deck, outbound sequence, value prop pillars, content calendar, are all one good prompt away from looking exactly like yours. The intelligence is no longer scarce. The frameworks are no longer secret. The playbooks are no longer hidden.

For about fifteen years, copying your competitor’s everything and making it slightly better was a viable strategy. That stopped working when everyone got access to the same copying engine.

So what is left?

Your customers. Specifically, what they say in your sales calls, what they ask in your support tickets, what they argue about internally before they sign, what they tell their peers in private, and what they refuse to pay for no matter how good your demo is. None of that is in any model’s training set. None of it can be reverse-engineered from your website. It happened yesterday, in your context, with your buyers. That is the only moat left that compounds.

The argument against this

Before I push further, the strongest case against what I just said. Some readers will argue that technology, data, and people are also moats. Each has a kernel of truth. Each is overstated.

Technology is a six to twelve month lead today, and shrinking each quarter. Whatever you ship in Q1, someone reverse-engineers and ships in Q3. Data is a head start, not a moat. Most B2B SaaS datasets sit inert in a warehouse and accumulate without getting sharper. A moat compounds. Data only compounds when someone interprets it. People are the closest thing to a real moat, but the best ones leave. They build the wall at one company and pour a partial copy at the next. Talent is a flywheel, not a wall.

The customer relationship is different. Every sales call you record adds a new sentence your competitor cannot read. Every support ticket adds a new failure mode they cannot anticipate. Every churn conversation adds an objection that is not in their training data. The asset compounds quietly while everyone else is downloading the same templates.

Five tests that turn customer access into a moat

In about forty SaaS companies I have worked with at Kalungi, I have landed on five tests that separate a useful persona from a paper exercise. Hit four out of five and you will outpace your competitor’s AI-generated version comfortably.

1. Voice. The persona must use your customer’s words, not your marketing team’s words. If your VP of Finance persona talks about “driving efficiency through optimized workflows,” you wrote it. If they talk about “the close grinding to a halt every month because three people are still copying numbers from the CRM into a spreadsheet,” your customer wrote it. The fastest path to voice is to listen to recorded sales calls and steal the exact phrasing. The exact phrasing is what no model can manufacture for your category.

2. Tangibility. Personas full of adjectives are noise. Personas full of numbers are signal. “Frustrated by reporting” is a placeholder. “Spends six hours every Monday rebuilding the same pipeline view because the CRM cannot filter by stage and rep” is a persona. Measurement is how the brain separates signal from background. If you cannot quantify the pain, the pain is not really a pain yet.

3. Story. Behind every good persona is a customer who could narrate the last time this problem hurt and the next time it will. When I interview a customer for a persona, I do not ask “what are your challenges.” I ask “tell me about the last time this went really badly, and the next time you expect it to.” Stories carry the context that bullet lists drop on the floor.

4. Trigger. Most of your market is not in market. About five percent are actively shopping. Ninety-five percent are not, and most of your competitors are dogpiling on the same five percent. The persona has to capture what makes the other ninety-five look up. A new VP joining. A regulatory deadline. A board meeting where revenue per employee got benchmarked against a peer. A funding event. Triggers are how you move from demand capture to demand creation, and they only surface when you have spent time with real customers.

5. Testable. A persona is a hypothesis until you use it. Hand your persona to two SDRs and ask each of them to build a list of one hundred matching prospects. The lists should overlap by at least sixty percent. If they do not, the persona is too vague or too poetic. The discipline of running personas through a sourcing test catches more nonsense than any review meeting.

Where the signal already is in your company

Most B2B SaaS companies sit on a treasure of customer signal and ignore it. The audit takes a Friday afternoon.

1. Recorded sales calls. Gong, Chorus, Fathom, Read.ai, Fireflies. Pick one. Your reps are already capturing the language your customers use at the highest-stakes moment of the relationship. Most of that audio is auto-summarized into nothing. The transcripts contain your value proposition, written by your customer, free of charge. Pull a sample of fifty deals, half won and half lost, and read every transcript for one job-to-be-done, one phrase, one trigger. You will rewrite your homepage by Tuesday.

2. Support tickets. Customers tell you what is broken in a support ticket more honestly than in any survey. Tag the last three months of tickets by job-to-be-done and you will find the gap between what you sell and what they actually use you for. Sometimes those are the same. Often they are not, and the gap is where your next case study lives.

3. Review sites. G2, Capterra, TrustRadius. Read the three-star reviews first, not the five-star ones. Three-star reviewers are the ones who took the time to be honest. They will tell you precisely where your value proposition leaks.

4. Internal surveys, sent to the whole company. The receptionist who fields every misrouted call has a perspective on customer sentiment your VP of Sales does not have. Filter by role afterward, in analysis. The point is to maximize the signal you capture, not to pre-select for the answers you expect to hear.

5. Customer interviews. Thirty to forty-five minutes, recorded with consent, structured around the Why/How Ladder I describe in Syntropy. Start with the customer’s own words. Climb up with “why does that matter” to surface the strategic outcome. Climb down with “how do you do that today” to expose the workflow. The combination forces them off autopilot.

How to deploy what you capture

Harvesting signal is half the work. The other half is mapping it to the three personas in any B2B buying committee, the model I describe in T2D3 as P1, P2 and P3.

P1 is the user who experiences the pain. They are usually not the buyer, but they are the gateway. Without P1 there is no business need. Marketing to P1 is about wow content: pieces that make them notice they have a problem they have been normalizing.

P2 is the decision maker. They are the champion who carries your case into a procurement meeting, and they need help looking like the hero who did the homework. Marketing to P2 is about how content: ROI calculators, comparison guides, buyer’s guides, templates they can take back to their team.

P3 is the executive sponsor or potential blocker. You will rarely meet them directly. Marketing to P3 is about now content: urgency, risk of inaction, peer comparisons. P3 content is what P2 forwards with a one-line message that says “we should move on this.”

The reason this matters in an AI world is that the WOW-HOW-NOW framework itself is now generic. Every model knows it. But the inputs to each layer, the actual phrases that wow your P1, the actual workflow your P2 wants to learn, the actual urgency that flips your P3, none of those can be inferred from a prompt. They have to be heard from your customers.

What you stand to lose

I have been watching two cohorts of SaaS companies over the last eighteen months.

The first treats their customer base as a moat to be cultivated. They invest in transcription, they read tickets, they survey the whole company, they interview customers monthly. Their content sounds like nobody else’s content, because it is built from their customers’ actual words. Their close rates have held even as their categories have crowded.

The second cohort treats AI as a content factory. They are publishing more, faster, with prettier visuals, and they are converging on the same handful of titles, the same five LinkedIn hooks, the same three case study templates. Their content is no longer competing on what it says. It is competing on volume. Volume is a cost game, and the cost is heading toward zero.

The difference is not who has better AI. They have the same AI. The difference is what the AI was pointed at. One group fed it the customer voice they had captured. The other group cut the customer voice out of the workflow and let the model run on commodity inputs.

The asymmetry is going to widen. As more of the playbook converges, the marginal value of having something a competitor cannot generate from public data goes up, not down. Your customers are that something.

Discussion items

1. How many customer interviews did your marketing team complete last quarter, and where did the insights end up?
2. When you read your last persona doc, would you bet it was written by your team or generated by an LLM in twenty minutes?
3. What percentage of your homepage copy can be traced to a specific customer phrase from a recorded call?
4. If you handed your ICP to two SDRs today and asked them each to source one hundred matching companies, how much overlap would you see?
5. Where in your stack does customer signal currently die? Sales call recording with no review process? Support tickets with no tagging? Surveys you never run?

Questions to ask

1. Are we mining our recorded sales calls for language, or are we letting them auto-summarize into the void?
2. When did we last interview a customer who is not a reference, just to learn?
3. Which competitor has noticeably better customer-grounded content than we do, and what specifically are they doing differently?
4. Are we testing personas by use, or only by review meeting?
5. What would change in our roadmap if every product manager listened to one hour of support tickets every week?

Sources

• T2D3 Second Edition, on ICP, P1/P2/P3 and the WOW-HOW-NOW model: t2d3.pro
• Syntropy, on signal versus noise and the Why/How Ladder: t2d3.pro/syntropy
• Finish Line Fridays, on OKRs for customer research and content as a moat: t2d3.pro
• Kalungi resources, including the B2B SaaS ICP template: t2d3.pro ICP template