Switching Costs and Lock-In

Why do switching costs matter to founders?#

They shape retention, but not the way the dashboard implies#

High switching costs can lower logo churn, because a dissatisfied customer needs a bigger improvement before moving. They can also delay churn: an account reduces seats, usage, new workloads, references, and expansion while it quietly prepares a migration. Renewal rate looks healthy while preference has already collapsed.

Track four retention states separately:

• Economic — the customer actively chooses the product and would renew if it evaluated today.
• Relationship — trust and accumulated operating knowledge keep creating value.
• Contractual — the customer still owes payment or cannot exit yet.
• Operational — the customer stays because migration is not ready.

Only the first two are evidence that the company is improving the customer's position.

They change pricing power — and tempt you to misuse it#

An attached customer will tolerate a price increase smaller than its effective switching cost. That does not mean charging up to it. The customer compares the present value of a recurring price difference against a mostly one-time migration burden: a small annual increase compounds, and each year makes the alternative's case easier.

Pricing near the theoretical threshold also creates adverse selection. Customers with the best alternatives and lowest migration costs leave first. The remaining base looks more inelastic but is concentrated in legacy use cases and trapped accounts. Near-term ARR rises while the product gets harder to sell to informed new buyers. See willingness to pay for the distinction between value-based price acceptance and tolerance caused by captivity.

They are a tax on entrants, including you#

A startup displacing an incumbent must finance or remove enough of the migration burden to make the customer's net switching value positive — through migration tooling, implementation services, contract buyouts, dual-running support, compatibility, or outcome-based milestones.

This belongs in the go-to-market model. Competitive displacement has a different sales cycle, win rate, implementation cost, and payback than a greenfield deal; conflating them in funnel metrics makes a transition-economics problem look like a positioning problem.

They are set by architecture#

Architecture allocates future switching cost. Proprietary schemas, tightly coupled APIs, irreversible model state, hidden metadata, and provider-specific infrastructure speed the first release and make everyone harder to move later — including you. A company can embed deeply in customer workflows while depending on a single cloud, model vendor, app store, or payment provider, building customer lock-in on one side while accepting supplier lock-in on the other. API-as-a-product design decisions are switching-cost decisions.

They get overclaimed in fundraising#

Investors look for switching costs because durable retention lengthens customer life and makes cash flows more predictable. But NRR, GRR, renewal rate, integrations per account, and contract duration are evidence to investigate, not proof of a mechanism — as Snowflake's cohort-defined 125% NRR illustrates. The metric is real; what causes it is a separate question.

They are an active policy subject#

The US Merger Guidelines treat switching costs as a potential barrier to competition, noting that control over a complement, interoperability layer, or service that helps customers use multiple providers can entrench a dominant position. In the EU, the Data Act removes switching charges entirely from January 2027. In the UK, the CMA's cloud investigation extracted interoperability and egress commitments from the two largest providers.

The operational lesson is not legalistic: portability and interoperability are not stable variables. A model that depends on blocking them can be undone by standards, customer architecture, competitor tooling, procurement requirements, or regulation.

How do you calculate net switching value?#

Use the customer's decision, not the vendor's retention narrative, as the unit of analysis. For a horizon of T periods and discount rate r:

PV recurring advantage = sum over t=1..T of [(value_alt,t - price_alt,t) - (value_inc,t - price_inc,t)] / (1 + r)^t

S_eff = search + migration + implementation + training + parallel run + expected disruption + contractual cost + lost functionality + lost relationship or network value - migration credits - recoverable asset value

net switching value (NSV) = PV recurring advantage - S_eff

If NSV is positive, switching creates value over the chosen horizon. If it is negative, staying does. If it is near zero, timing, capital, risk tolerance, and reversibility decide.

Converting components to customer economics#

Cash invoices are the easiest costs to see and usually the smallest:

employee-time cost = hours x fully loaded hourly cost

expected disruption cost = probability of adverse event x economic loss if it occurs

temporary productivity cost = affected users x hours lost per user x loaded hourly cost

Use low/base/high ranges. A transparent range beats a false-precision point estimate — and this is where economic value estimation technique carries over directly.

The break-even incumbent price#

When the recurring difference is roughly level, use the present-value annuity factor PVAF(r,T) = [1 - (1 + r)^(-T)] / r. With P_A the alternative's annual price, Q the incumbent's annual value advantage (negative if the alternative is better), and S_eff the effective one-time cost, the incumbent price at which the customer is indifferent is:

P_I* = P_A + Q + S_eff / PVAF(r,T)

This is a decision threshold, not a pricing recommendation. Charging just below it assumes you know the customer's alternatives, costs, horizon, risk tolerance, and reaction — and ignores trust, referrals, negotiation, and the option value of making migration easy.

Worked example: a renewal under pressure#

A 100-person company is renewing a workflow SaaS product. The incumbent proposes $72,000 per year. A credible alternative costs $44,000 per year, but users would initially lose about $4,000 per year of productivity because one incumbent workflow is better. The buyer uses a three-year horizon and a 10% discount rate.

Step 1 — Recurring advantage of switching#

Annual price saving: $72,000 - $44,000 = $28,000. Net of the productivity gap: $28,000 - $4,000 = $24,000 per year.

PVAF(10%,3) = [1 - 1.10^-3] / 0.10 = 2.4869

PV recurring advantage = $24,000 x 2.4869 = $59,684

Step 2 — Effective switching cost#

Step 3 — Net switching value#

NSV = $59,684 - $52,000 = $7,684

The alternative wins by roughly $7,700 in present value — if the estimate is credible and the customer can fund a one-time migration out of a budget that is usually not sized for it.

Step 4 — The incumbent's price threshold#

P_I* = $44,000 + $4,000 + ($52,000 / 2.4869) = $48,000 + $20,910 = $68,910

Below about $68,900, staying wins. At $72,000, switching wins. If the incumbent renews at $66,000 instead, the annual advantage falls to ($66,000 - $44,000) - $4,000 = $18,000, so PV = $44,763 and NSV = -$7,237. Same product, same customer, same alternative — opposite decision, from an 8% price move.

Step 5 — The entrant's migration offer#

Suppose the alternative offers a $15,000 migration credit tied to successful data validation and cutover. Effective switching cost falls to $37,000, and at the incumbent's $66,000 price NSV = $44,763 - $37,000 = $7,763. The entrant has converted an uneconomic displacement into a viable one.

But the credit is acquisition and implementation cost. At an 80% gross margin on $44,000 of annual revenue, first-year gross profit before migration support is $35,200, and a $15,000 credit consumes 42.6% of it. Sales leadership should test payback, retention, and implementation capacity rather than celebrate the logo.

Step 6 — Stress-test before believing it#

The $7,684 base case is small relative to the uncertainty in its inputs. Raise cutover risk from a 20% to a 40% chance of a $40,000 loss and effective cost rises $8,000 — NSV turns negative. Close the productivity gap and NSV rises by $4,000 x 2.4869 = $9,947.

So the honest conclusion is not "switching costs are $52,000." It is:

Under a three-year, 10% base case the current proposal is about $7,700 worse than switching, and cutover risk and the productivity gap are the two decision-sensitive assumptions.

That gives the buyer, the incumbent, and the entrant something they can each act on. Note what the arithmetic hides: a single discount rate, a horizon chosen by convention, and a cutover probability that nobody can observe directly. Treat the model as a way to locate the two assumptions that matter, not as a number to quote.

How do you measure switching costs in your own business?#

1. Define the switching event. Name the incumbent, alternative, segment, workflow, scope, decision date, and horizon. Replacing one module is not replacing the system of record, and a buyer may multi-home or migrate only new workloads. Without this scope, teams mix partial migration costs with full-platform retention.

2. Map assets, dependencies, and owners. Inventory data, metadata, permissions and history; integrations, scripts, reports and identity; workflows, approvals and controls; user and admin knowledge; reputation, collaborators and partner access; contracts, credits, equipment and certifications. Assign an owner and an evidence source to each. "Deeply embedded" is not a measurement.

3. Separate sunk cost, ongoing value, and switching cost. For each asset ask: has the cost already been incurred regardless of today's choice (sunk)? Does it create value only with the incumbent (lost recurring value)? Must it be recreated to use the alternative (switching cost)? An integration built last year is not a switching cost equal to its original invoice — if a standard connector replaces it in two days, only that burden counts; if it embodies undocumented logic, the current cost may exceed the original spend.

4. Estimate ranges and avoid double counting. Use low/base/high for hours, cutover loss, feature gaps, and export completeness. Correlate risks where they are correlated — a poor export raises both migration time and outage probability. Do not add the full value of a lost function and all the labour of the workaround for it.

5. Observe behaviour, not opinions. Customers do not know the work until they try it. Stronger evidence: completed exports and restore tests; migration plans and actual hours; competitor-displacement postmortems; usage before and after a cancellation request; greenfield versus displacement win rates; the discounts or credits needed to win a switcher; and how many customers chose multi-homing over replacement.

Run an exit rehearsal on yourself: export a realistic account, rebuild it in a neutral environment, validate data and permissions, and record what is missing. You should know your own portability burden before a customer, regulator, acquirer, or incident discovers it.

6. Link the mechanism to unit economics. Connect switching evidence per segment to logo and revenue retention, pre-churn contraction, renewal price change, expansion, support and implementation cost, displacement CAC and payback, and contribution margin. If accounts with ten integrations retain better only because they are larger and better supported, integration count is not causal.

7. Record what would falsify the claim. For example: configuration-deep customers churn at the same rate as shallow ones after controlling for size and tenure; a new migration tool cuts cutover from eight weeks to two without moving win or churn rates; renewals fall after price increases even among supposedly locked-in accounts; retention disappears when contract terms expire.

How do founders, operators, and investors use this?#

Product — create portable value. The best attachment is a product that gets more useful as the customer configures, integrates, and trusts it while staying intelligible and exportable: documented APIs and common export formats; exports that include relationships, identifiers, permissions and metadata; configuration-as-code; reversible pilots; reliable import tools for competitive migrations. JFrog's decision to market multi-environment deployment as helping customers avoid vendor lock-in is the strategic alternative to captivity — sell flexibility as part of the value proposition.

Go-to-market — sell the migration, not just the destination. In a displacement the buyer purchases two things: the future operating state, and the transition. Build a migration offer with named scope, data completeness, responsibilities, timeline, validation, rollback, and risk allocation — and price it. Free services with no capacity model destroy gross margin and trust. Then segment pipeline by motion:

If displacement win rate is low but post-pilot preference is high, the constraint is transition economics, not positioning.

Pricing — separate value capture from captivity. Measure price realisation by segment, tenure, adoption depth, and alternative maturity; track contraction, delayed renewal, escalation, and advocacy after price changes. Offer longer terms in exchange for real consideration — price certainty, implementation funding, reserved capacity, service levels — rather than as a trap. Avoid renewal, cancellation, or export practices a customer cannot explain to its own procurement team. In usage-based and credit-drawdown models, commitments and expiring credits are switching costs; price them knowingly.

Customer success — distinguish risk from resentment. Keep an account-level record of outcomes still valued, workflows actively used, unresolved export or billing concerns, renewal intent separate from obligation, credible alternatives and migration readiness, and any shadow-system adoption. "We can't leave this year" is not a success signal; the question is whether the constraint is continuing value, a planned transition, or a punitive barrier.

Finance — model retention states, not one churn rate. Split cohorts into preferred and expanding, preferred but budget-constrained, contractually retained, operationally trapped but preparing to migrate, multi-homing, and at risk because an alternative reduced migration cost. Apply different renewal, expansion, and support assumptions to each. Valuing a trapped customer like an advocate is how forecasts break.

Investors — audit the causal chain. Ask which mechanism is claimed and what evidence supports it: matched cohorts by workflow depth (not integrations available); restore-tested export gaps and rebuild time (not records stored); time-to-proficiency and switching postmortems (not user tenure); renewal after the initial term (not remaining contracted ARR). Also ask who bears the cost — a workflow that is hard for the customer to move may be equally hard for the vendor to support, producing services-heavy revenue and thin contribution margin.

What are the common mistakes?#

Treating retention as proof of lock-in#

Customers stay because the product is good, alternatives are poor, budgets are frozen, contracts have not expired, or nobody has revisited the decision. Use switching evidence, not the outcome alone. This is the same discipline that separates a real moat from a favourable moment.

Counting sunk implementation spend#

The original implementation invoice is not today's switching cost. Estimate only what must be paid or lost from this point forward.

Calling an unusable export a moat#

Missing fields, proprietary formats, artificial egress charges, and obscure cancellation paths delay exit. They also raise procurement friction, weaken trust, and — as both the EU Data Act and the CMA cloud commitments show — attract the kind of attention that removes them on someone else's timetable.

Assuming higher switching costs always soften competition#

Vendors discount aggressively before attachment because the installed base has future value, and buyers demand a lower entry price because they anticipate later lock-in. Viard's toll-free-number evidence runs the other way: removing switching costs lowered prices. Both results are market-specific, which is the point — the sign is an empirical question, not a rule.

Equating a long contract with a durable customer#

A three-year term shifts the timing of reported churn while usage and preference migrate elsewhere. Review adoption, consumption, expansion, and renewal intent throughout the term, not at the end of it.

Ignoring your own lock-in#

A founder who celebrates proprietary customer dependencies while depending on one cloud, one model vendor, one channel, or one payment provider may have a larger switching problem than the customer base does.

When does the switching-cost story break?#

Portability and interoperability improve#

Standard formats, documented APIs, direct transfers, identity standards, and migration tools turn a multi-month project into routine work. Switching cost is a depreciating asset whenever the ecosystem is investing in compatibility — and where regulation is doing the investing, the depreciation has a published date.

Portability rights have limits worth understanding precisely, though. UK data-protection guidance gives individuals a right to receive certain personal data in a structured, commonly used, machine-readable format, but only under specific conditions and generally for data they provided — not for profiles the controller inferred. A legal right to some data is not functional equivalence of a whole product.

A new architecture bypasses the migration#

An entrant need not recreate the incumbent. It can take new workloads, sit above the existing system, use a neutral data layer, or replace one high-value job — letting the customer switch incrementally instead of absorbing a full cutover.

The customer multi-homes#

Running two providers preserves option value, builds transferable skills, and moves new workloads gradually. It costs more in the short term while sharply reducing the incumbent's future pricing power.

The product fails a trust threshold#

An outage, security incident, compliance failure, billing dispute, or strategic betrayal can make the risk of staying exceed the risk of moving. Switching costs do not retain a customer whose expected value from the incumbent has gone negative.

The buyer, sponsor, or budget changes#

A new executive treats legacy configuration and relationships as liabilities rather than assets. A merger, reorganisation, or new compliance rule shortens the horizon and changes who bears the transition cost.

The attachment costs more to maintain than it earns#

Custom integrations, old versions, bespoke workflows, and relationship-specific service retain customers while raising support cost and slowing the roadmap. If customer-specific complexity grows faster than gross profit, the switching cost is not an attractive moat — it is a tax you are paying to keep revenue you cannot serve profitably.

High switching costs suppress the market#

Buyers who fear future lock-in avoid adoption, reduce scope, demand shorter terms, insist on self-hosting, or stay on a spreadsheet. A retention advantage can shrink conversion and expansion; measure the full lifecycle, not the back half.

Founder checklist#

Answer each with a number, a test, or an explicit hypothesis:

1. Which exact customer, workflow, incumbent, and alternative are being compared, and over what horizon?
2. Which costs are incremental from today, and which are sunk?
3. What data, metadata, permissions, history, and model state must actually move?
4. Which integrations and workflows are active rather than merely installed?
5. What is the low, base, and high estimate for migration and cutover — and what is the net switching value at each?
6. Which two assumptions can reverse the decision?
7. Is retention economic, relationship-based, contractual, or operational — and in what proportion?
8. What happens to usage and expansion before logo churn appears?
9. How do greenfield and displacement win rates, CAC, and payback differ?
10. Which costs create continuing customer value, and which merely obstruct exit?
11. Where is the startup itself locked into a supplier, cloud, model, or channel?
12. What product improvement would make customers stay even if switching were free?

If the last question has no answer, the company is defending an installed base, not building a durable product.