The paranoia every B2B product CEO is now carrying, and mostly not naming out loud: will someone deliver my product lighter, AI-enabled, faster — and do I need to invest in new product, or in making sure the customers I already have never leave? This is the operational question the AI shift forces, with the CRM category as the canary.
If you run a B2B product company at any scale — early-stage SaaS, growth-stage platform, publicly traded enterprise software — the following is probably somewhere between running quietly in the background and keeping you up at night.
Question one. Is someone, somewhere, about to deliver my product — or something close enough to it that customers will switch — lighter, faster, AI-native, and at a fraction of my price? Not necessarily a well-funded venture competitor. Maybe a team of four that used LLMs to compress eighteen months of engineering into three. Maybe even my own customers, building an in-house tool they would previously never have considered building.
Question two. Do I invest the next product cycle in new features — the obvious path, the one my product leaders are pushing — or do I invest it in making sure my existing customers never leave? Because if a lighter alternative does appear, the customers who stay will be the ones for whom leaving is more expensive than the subscription, and that depth of embedding is not built by roadmap slides. It is built by a deliberate, compounding investment in stickiness.
Question three. What has become table stakes in my category that I am not shipping yet? Not differentiators — table stakes. The things that eighteen months ago were competitive advantages and are now the floor. Agent-operable surfaces. LLM-consumable data exports. Context-aware defaults. If I am not shipping those, I am quietly being filtered out of evaluations I do not even know I am in.
These three questions used to be a five-year strategic consideration, revisited at off-sites. They are now an operating horizon. Product CEOs who are not asking them weekly are running the company on last year's model of the market. The market has moved.
The CRM category is useful not because it is unique but because it is representative. Salesforce is a ninety-billion-dollar company. HubSpot is a category-defining product. Pipedrive, Zoho, Dynamics, Freshsales, Monday, ClickUp, and a long tail of vertical CRMs have built real businesses on the same underlying model. And yet the foundation that category is sitting on has four cracks in it, each of which compounds the others.
Crack one. Most customers use ten to fifteen percent of the features they pay for. This has always been true to some degree. What is new is that the customer now has a clear mental model of that fact, because an LLM will tell them — in thirty seconds — which features they are using, which they are not, and what the price-to-utilization ratio looks like compared to alternatives.
Crack two. Every meaningful configuration change in a mature CRM requires a specialized administrator. Every custom field, every workflow rule, every integration adjustment. The administrator is expensive, the turnaround is slow, and the customer resents both. The customer's CFO quietly notices how much the "total cost of CRM ownership" has drifted above the license line item.
Crack three. The data model is highly generalized to serve every industry, which means every customer feels like it is nearly but never quite built for them. The gap between the generic model and their actual business is bridged by the administrator — which amplifies crack two — and by the internal team rebuilding reports outside the CRM, often in spreadsheets, often in ways leadership does not see.
Crack four. The pricing model is per-seat in an era when teams are actively being restructured by AI. Finance teams are now looking for SaaS licenses to rationalize. Per-seat pricing is structurally disadvantaged in that conversation, because the buyer's headcount forecast has more downside risk than it has had in a decade.
Any one of these is manageable. A vendor can address crack one with better adoption tooling. Crack two with no-code admin features. Crack three with vertical versions. Crack four with outcome-based pricing pilots. But all four compounding is how categories get quietly disrupted — not by a single superior competitor, but by the customer waking up to a new economic reality.
The CRM customer is not leaving for a better CRM. They are leaving for a custom system their own team built in six weeks that covers exactly what they actually use — and nothing they don't.
The reason enterprise buyers reliably bought Salesforce or HubSpot instead of building their own was never primarily about the features. It was about the math. The cost of building a robust, secure, compliant, multi-user, multi-workflow, integrated, performant customer data system was enormous. The expertise to do it well was rare and expensive. Maintenance was its own ongoing line item. The incumbent vendors amortized all of that cost across thousands of customers, which is why the subscription was a bargain relative to the true in-house cost.
LLMs did not eliminate these costs. But they collapsed the first two meaningfully. A competent product-minded operations leader, paired with two mid-senior engineers and an LLM coding assistant, can now build a custom version of a mid-market firm's most-used CRM workflows in six to ten weeks. It will not have the robustness of Salesforce. It will not have the breadth. It will not have the plug-and-play integration ecosystem. It will not have the enterprise-grade audit trail. It will not have twenty years of edge-case handling.
And it does not need to. It only needs to cover the ten to fifteen percent of features the firm actually uses, configured exactly for their business, with pricing that is a one-time build cost plus ongoing maintenance rather than a per-seat subscription that scales with the org chart.
The incumbent's instinct is to dismiss this. "They will regret it in eighteen months when they hit a compliance requirement we have already solved." That may be true. It is also irrelevant to the CFO signing the decision this quarter. The CFO is making a decision against a subscription line item that is visible in the budget every month, with a total that has been growing faster than revenue. The hypothetical future compliance cost is not.
The strongest remaining moat most incumbent SaaS platforms have is not the feature set. It is compliance. SOC 2 Type II. ISO 27001. GDPR. HIPAA. FedRAMP for the public sector. Industry-specific certifications for financial services, healthcare, pharma, defense. The cost of building and maintaining these is meaningful for a small player, and it amortizes efficiently across a large installed base, which is why enterprise buyers still default to incumbents even when the product itself is unremarkable.
This is why the custom-build-an-in-house-CRM move is, today, mostly a mid-market phenomenon. Enterprise buyers cannot responsibly run customer data on an uncertified internal system. The compliance requirement is what forces them back to the shortlist of mature vendors.
Now consider the scenario where a well-funded third party packages compliance as a service. An opinionated infrastructure layer — somewhere between a hosting platform and a managed security provider — that a smaller product company or even an internal IT team can adopt to inherit enterprise-grade compliance posture. SOC 2 Type II out of the box. Audit logs by default. Data residency controls baked in. A turn-key path from "we built our own thing" to "we built our own thing, on infrastructure that makes the CISO comfortable."
This is not speculative. Versions of it already exist in adjacent categories. What is speculative is when a fully productized, enterprise-credible version ships in the core B2B data-hosting space. When it does, the moat that keeps enterprise buyers defaulting to incumbent platforms weakens substantially. Customers who previously had no choice suddenly have one.
The question for an incumbent product CEO is not "will compliance-as-a-service ship in my category this year." It is "if it ships, how much of my enterprise retention depended on it being unavailable?" If the honest answer is "a lot," then that retention is not moat. It is regulatory drag. Regulatory drag always eventually eases. The firms that planned for its easing are positioned. The firms that did not discover they have a feature-parity problem with a competitor they did not previously consider a competitor.
Every product CEO in this position — feeling the pressure of a possibly-commoditizing core, watching retention become the most important metric the board quietly cares about — has to make a capital-allocation decision the field commonly miscategorizes as "roadmap prioritization."
It is not a roadmap decision. It is a strategic posture. Investing in new product means building things that grow the revenue line by either attracting new customers or upselling existing ones. Investing in stickiness means building things that make the cost of leaving higher than the cost of staying, which grows the revenue line by making the existing base harder to displace. In a calm market, a well-run product company does both. In a market where the floor is moving, the company that tries to do both often does neither well.
The hard truth is that most existing product roadmaps skew aggressively toward the new-features side, because new features are what sales asks for, what marketing can announce, what the board can be shown, and what the product team finds motivating. Stickiness work is quieter, slower, and less visible. It does not produce launch events. It produces, six quarters later, a retention number that did not crater when a competitor did something the incumbent did not expect.
There are three sources of real stickiness — the kind that survives a CFO review. New feature development that does not contribute to at least one of them is decoration.
A useful discipline for any product CEO right now is to run a periodic audit asking: what was a bullet point on my competitive teardown twelve to eighteen months ago that is now — quietly, without fanfare — the minimum expectation for any product in my category? Three stand out across most of B2B SaaS.
Buyers increasingly expect to give instructions to an AI assistant and have the product act on them, not navigate a UI manually. This is not the "AI feature" marketing slide. It is a structural change in how buyers interact with software. Products without an agent-callable interface — MCP server, structured function-calling API, well-documented action surface — are being filtered out of evaluations by buyers who run their workflows through LLM assistants. The filter is often silent. The buyer does not email to say they were excluded.
Every feature must be one step away from the user's primary workflow, because cognitive load is the new churn. In the era when users compared products feature-for-feature, breadth of capability was the win. In the era when users compare products moment-by-moment against the alternative of "just ask the LLM," every UI step is a micro-tax. Products that make users click through four screens to do what the LLM would do in one prompt are quietly losing users to the LLM — or to a competitor who built the same feature in one click.
Customers are increasingly running their own intelligence layer on top of your product — their own analytics, their own LLM workflows, their own retrieval systems. Products that make data export hostile, or that produce exports in formats unfit for LLM consumption, are being routed around. The customer still uses you for operational workflows, but the strategic analytical layer — the work the CFO and the executive team care about — happens outside your product. That is how you go from being the system of record to being a data entry tool for the customer's actual system of intelligence. That is a terrible position to be in.
None of these three individually kill a product. All three together quietly disqualify products from an increasing share of evaluations. The compounding effect is what makes this dangerous — and what makes the honest audit worth running, even if the answer is uncomfortable.
The pillar piece of this series argued that speed with precision replaces information asymmetry as the source of B2B advantage. For product companies, that abstract thesis shows up as a very specific operational capability: knowing which of your accounts are expanding, which are churning, and which integrations are generating the stickiness that protects the subscription — before the account becomes a retention problem on the quarterly report.
Most product companies do not have this. They have usage dashboards. They have quarterly customer-success reviews. They have a CSM team that manages accounts reactively, often in response to a support ticket volume spike or a declining login count that has already signaled the customer is halfway out the door. By the time the signal reaches the CSM, the intervention window has closed.
The firms that have instrumented their go-to-market motion around speed with precision catch the signals early enough to act. A key stakeholder changes role inside an account — the new stakeholder's priorities are often different, and the product may no longer fit them. A heavily-used integration gets silently deprecated by the customer's IT team — the product's stickiness just lost a load-bearing pillar. A new AI-powered lighter alternative gets mentioned on a customer's internal Slack — the evaluation you do not know is happening has started. Each of these is a signal. Each is actionable if caught in days. Each is a retention crisis if caught in quarters.
Retention is no longer a customer success metric. It is the primary product metric.
The product is not what you shipped last quarter. It is the compounding depth of stickiness your customers are choosing not to unwind, quarter after quarter. New features that do not deepen stickiness are decoration. New products that do not deepen stickiness are distraction. And the single highest-leverage capability a product company can build right now is the operational ability to see, in real time, what is happening in the accounts already paying them — because the retention fight of the next three years will be won by the firms that caught the signals before the customer started evaluating alternatives.
Not a demo with a sales rep. A direct conversation about whether your product company has the kind of account-level instrumentation this piece describes — and whether the timing is right to close that gap. If it isn't, we'll tell you that.
All five parts at the series hub.