There is a sentence in almost every SaaS strategy deck: increase switching costs. It is presented as straightforwardly good — a moat, defensibility, pricing power. It is also, stated plainly, a plan to make it expensive for your customers to stop paying you for reasons unrelated to whether your product is any good.
This article is about building without that, and about the two failures that travel with it: metrics that quietly take over the roadmap, and supply chains arranged so that nobody decided anything.
Lock-in: retention you did not earn
The register that governs this publication names this centralisation and enclosure: network effects convert a commons into a landlordship; exit is priced out. The word landlordship is precise. A landlord's income does not depend on continuing to deserve it — it depends on the cost of leaving.
The problem is not that lock-in is unkind. It is that it destroys your instruments. Retention is your primary signal of whether the product is working. Raise switching costs and retention measures two things at once — value delivered and captivity — with no way to separate them. You have blinded yourself and called it a moat.
What follows is predictable and well documented: the product stops improving in the dimensions customers care about, because the feedback that would have driven those improvements has been suppressed. Then a competitor arrives with a migration tool, and the accumulated dissatisfaction leaves at once. The moat did not prevent the loss; it deferred and concentrated it.
The testIf every customer could leave costlessly tomorrow, how many would? If you do not know, your retention number is not telling you what you think. If you do know and the answer frightens you, that is the real state of the business — the switching costs were only postponing your knowledge of it.
The exit duty, as a design principle
The governing instrument states a duty it calls exit: complete export in a documented, non-proprietary format, available without asking and without negotiation; no technical bar to interoperation; stated notice before adverse term changes.
Read that as a product spec rather than a compliance chore. Each clause closes a specific evasion:
| Clause | The evasion it closes |
|---|---|
| Complete | Exporting the records but not the relationships, history, attachments or configuration — technically an export, practically useless. |
| Documented | A JSON dump with no schema. The data is present and unusable without reverse engineering. |
| Non-proprietary | An export that only your own product can read. |
| Without asking | Export by support ticket, where retention staff intercept the request. This is the most common evasion and the least defensible. |
| Without negotiation | Export as a bargaining chip in a renewal conversation. |
| No technical bar to interoperation | Rate limits, API restrictions and terms of service that prohibit migration tooling. |
| Notice before adverse changes | Repricing or removing a feature customers built on, effective immediately. |
Our position is that shipping this well is a commercial advantage rather than a cost. It is a credible, verifiable signal that you intend to keep earning the renewal — the kind of claim competitors cannot make cheaply, because making it requires actually doing it. It also forces internal honesty: a team that knows customers can leave at any moment prioritises differently, and better.
Metric colonisation: when the scoreboard starts coaching
The second harm: metric colonisation — the measure replaces the thing measured, then governs it. Every growth organisation runs on ARR, net revenue retention, activation rate and churn. Each is a reasonable proxy. Collectively they become the definition of the product's purpose, and the product bends toward them.
The bend is visible in patterns every user recognises:
- Onboarding that optimises activation events rather than the user reaching competence.
- Notifications that raise engagement and lower the product's usefulness per session.
- Features shipped because they move a number in a quarter, deprecated when attention moves.
- Cancellation flows measured on saves, which converts a support surface into an obstacle course.
The governing instrument's response is a duty it calls metric honesty: state what the metric does not capture, and do not couple a single metric to an automatic material consequence without human review.
The first half is a cheap, high-value practice. Next to every headline number in your internal reporting, write the sentence describing what it cannot see. "NRR does not distinguish expansion from customers who could not leave." "Activation does not measure whether the user got the outcome they came for." The sentence does not fix the metric. It keeps the gap between the measure and the purpose visible to the people acting on it, which is the thing that otherwise disappears.
Responsibility diffusion: the harm with no author
The third: responsibility diffusion — the supply chain is arranged, by accident or design, so that every party can truthfully say the decision was not theirs. Note "truthfully." This is not about lying. It is about a configuration in which everyone's account is accurate and nobody is answerable.
A modern product is assembled from an identity provider, a payments processor, an email platform, an analytics vendor, a support tool, a fraud service and a model API. When a customer is wrongly locked out, the honest answers are: the fraud vendor flagged it, the identity provider enforced it, the support tool routed it, and the team that could override it did not see it.
The governing instrument responds with a chain-of-responsibility duty. The operative test is one question, and it is the most useful governance question we know:
The question to be able to answerFor any behaviour of material consequence in your product, can you name the party who could change it? Not the team that owns the ticket — the party with the practical capacity to make it stop. If you cannot name them, responsibility has already drained out, and it will not reappear during an incident.
Two corollaries in the same duty are worth stating for anyone writing contracts:
- Where two parties could each say the decision was the other's, they are jointly responsible until one publicly accepts it. Ambiguity does not distribute duty; it multiplies it.
- A contract term allocating a duty to a party who lacks the practical capacity to discharge it is void for these purposes. You cannot contract a duty onto a supplier who has no ability to perform it.
Labour that is disassembled until it is deniable
The fourth harm in this category is labour fragmentation: human labour is disassembled until it is deniable, then denied. It reaches ordinary software businesses through contractor classification, outsourced moderation, offshore support and human-in-the-loop review presented as automation.
The duty attaching to it is disclosure-shaped: where human labour is intermediated, or output presented as automated is materially human-produced, disclose the human involvement, state the pricing rule, and make deactivation appealable.
Whether presenting human-assisted output as "AI-powered" is deceptive is genuinely disputed — the industry's position is that the composite system is the product and its internals are implementation detail. Our position is that it matters when a buyer would price the product differently knowing the truth, which is frequently. We hold that position while acknowledging the counterargument is not frivolous.
The counterproductivity test for a product
The single most useful governance practice, and the cheapest: state the condition under which your product would be frustrating its own purpose. Not failing commercially — succeeding commercially while defeating the reason it exists.
- State the purpose in a sentence with no metric in it"Teams ship correct code faster." Not "increase deploys per week."
- State the measurable condition of counterproductivity"Deploy frequency rises while incidents rise faster" — the product is now producing speed at the cost of the correctness it promised.
- Measure it on a schedule and write down the answerAnnually is enough. The value is the record, because it shows whether your threshold moved.
- Publish it if you canA public counterproductivity condition is the strongest available signal that the claim is not marketing. Very few competitors will match it.
Honest limits
This is harder to fund. Investors price defensibility, and switching costs are the most legible form of it. A business built for exit tells a harder story in a fundraise. That is a real cost and pretending otherwise would be dishonest.
Some lock-in is unavoidable. Data gravity is a property of holding data, not a strategy. The distinction is between switching costs that arise from doing the job and switching costs manufactured to raise the price of leaving — and the boundary is genuinely fuzzy in places.
Metric honesty does not prevent metric colonisation. Writing down what a number misses helps and does not stop the number governing behaviour, because incentives are stronger than annotations. Compensation tied to a single metric will defeat any amount of thoughtful commentary next to it.
And the strongest counterargument: a business that fails helps nobody. If exit-friendliness kills the company, customers lose the product entirely — a worse outcome than mild captivity. We think this is usually overstated, and that businesses collapse from unearned retention hiding a defect far more often than from making departure easy. But it is the honest case against, and anyone adopting this should hold it in view.
Revision log. Sep 2026: first publication. Written as the mastery track for the business desk's structural issues, referenced from the "More Details" section of Fiscora's business guides. Reviewed annually.