Switching Costs, Lock-In, and Who Owns the Standard

Key takeaways

• Lock-in shows up as duration, not affection. ADP's fiscal 2026 Form 10-K estimates client retention at about 13 years in Employer Services and about 6 years in PEO. A 13-year average life is the arithmetic of roughly 7.7% annual attrition, in a business that pays one in six US workers.
• The rent from a locked-in customer is bid away before you collect it. Farrell and Klemperer's survey of the literature finds that this shift of rivalry to the front of the relationship, through introductory offers and price wars for the market, is usually a poor substitute for ordinary competition and tends to strengthen incumbents.
• The toll on a standard is the most profitable line anyone runs. Qualcomm's licensing segment earned $4,043 million of pre-tax profit on $5,582 million of revenue in fiscal 2025, a 72% margin, while its far larger chip segment earned $11,670 million on $38,367 million, a 30% margin.
• Switching costs now have statutory expiry dates. Article 29 of the EU Data Act bans cloud switching charges outright from 12 January 2027, and the common charger directive made USB Type-C mandatory across 13 device categories from 28 December 2024.

A switching cost is a bill the customer pays to leave

A switching cost is whatever a customer must spend (in money, hours, risk, or lost work) to stop being your customer and start being someone else's. It is not loyalty and it is not satisfaction. It is a bill that only arrives at the exit, which is why it never appears on anyone's income statement and why most operators have never actually added it up.

It comes in four recognisable forms, and they behave differently. Learning costs are the retraining: the team that knows the keyboard shortcuts, the closing checklist, the report that only one person can build. Asset specificity is everything the customer has poured into your system that does not come out in usable shape: the configuration, the custom fields, seven years of history, the rules nobody documented. Contractual costs are the visible ones: termination fees, notice periods, minimum commitments, the annual plan billed up front. And complementary lock-in is the web of integrations, file formats, and downstream systems that all assume you are still there.

The fourth kind is the one legislators have started to attack by name, because it is the easiest to see. Recital 88 of the EU Data Act observes that unnecessarily high data egress charges and other unjustified charges unrelated to actual switching costs inhibit customers from switching and cause lock-in effects. It does not stop at observing. Article 29 says that from 12 January 2027 providers of data processing services shall not impose any switching charges at all, and that in the interim window from 11 January 2024 they may charge only reduced amounts not exceeding the costs directly linked to the switching process. Article 25 sets the operational shape: a maximum notice period that shall not exceed two months, and a mandatory maximum transitional period of 30 calendar days for the migration itself. The Regulation applies from 12 September 2025.

Read that as a warning about which of your four switching costs are durable. Contractual friction is the one a regulator can delete with a date. Learning costs and asset specificity are the ones you actually own, because no statute can make a customer's staff un-learn your product or make seven years of configuration portable in thirty days. If your retention rests on an exit fee, you are renting your moat from a legislature that has already shown it will take it back.

One distinction worth nailing down before going further, because the two get blurred constantly. A switching cost binds a customer to their own past purchases. A network effect binds them to other users' choices. Farrell and Klemperer treat them together for good reason, since both make compatibility valuable and both create lock-in, but the levers differ. This briefing is about the first one.

Measuring the lock-in you actually have

You cannot observe a switching cost directly, but you can observe what it does: customers stay longer than the product's merits alone would justify. Duration is the proxy, and the cleanest published examples come from businesses that are boring on purpose.

ADP's fiscal 2026 Form 10-K states that, based on its retention levels in fiscal 2026, client retention is estimated at approximately 13 years in Employer Services and approximately 6 years in PEO. That is the company's own estimate derived from its retention rate, not a measured cohort followed for thirteen years, and it should be read that way, but the arithmetic behind it is unambiguous. An average life of 13 years is what roughly 7.7% annual attrition produces. For scale, the same filing reports 1.1 million clients, 42 million workers on the platform, $21,947.4 million of fiscal 2026 revenue, and the claim that ADP pays one in six US workers and moved $3.5 trillion during the year. Payroll is not a beloved product. It is a product whose replacement project has a tax-filing deadline attached to it.

The second measurement most software businesses reach for is net revenue retention, and it needs handling with gloves. Snowflake's Form 10-K for the year ended 31 January 2026 reports net revenue retention of 125%, against 126% a year earlier and 133% the year before that, on 13,328 customers and $4,472.3 million of product revenue. A number above 100% means the surviving base bought more than the departing base took away. It does not mean nobody left. NRR nets expansion against churn inside one figure, so a healthy-looking 125% is fully compatible with losing small logos steadily, and the three-year slide from 133% is the more informative series, because it says the expansion inside the locked-in base is decelerating even as the base grows.

Illustrative only: the following figures are invented round numbers chosen to make the arithmetic legible. Take a vertical SaaS with 400 customers paying $1,000 a month, so $12,000 a year each and $4.8 million of ARR. At a 70% gross margin each customer-year throws off $8,400 of gross profit. Value a customer as gross profit divided by the sum of churn and a 10% discount rate. At 8% annual gross logo churn, that is $8,400 ÷ 0.18 = $46,667 per customer. At 18% churn, it is $8,400 ÷ 0.28 = $30,000. Ten points of retention is worth $16,667 per customer, and across 400 customers roughly $6.7 million of enterprise value. Against a $18,000 CAC, the same business is either a 2.6x LTV-to-CAC operation or a 1.7x one, and nothing about the product changed.

The lever, then, is not a loyalty programme. It is deliberately manufacturing the two switching costs a regulator cannot delete: get the customer's data into a shape only you hold, and get their staff fluent in your workflow. Onboarding depth, integrations built during month one, historical import, certification and training: these are investments in the customer's exit cost, and they are worth spending real money on, which the arithmetic above prices.

The risk is measuring the wrong thing and believing it. Track gross logo retention separately from net revenue retention, always, and segment retention by cohort tenure. Lock-in that only appears after month eighteen is a very different asset from lock-in that holds a customer through month three. And watch the annual-contract illusion: a 12-month prepaid term produces flattering in-year retention while telling you nothing about what happens at renewal, which is the only moment the switching cost is actually tested.

Bargain first, harvest later — and why the profit often disappears

Here is the trade-off almost nobody states out loud. Lock-in raises the value of a customer you already have. It does not raise your profit, because your competitors can do the same arithmetic you just did and will spend against it.

Farrell and Klemperer's Handbook of Industrial Organization survey is blunt about the mechanism. When products are incompatible and buyers get locked in, sellers gain real power after the sale. Competition does not vanish; it relocates. It moves to the front of the relationship, where firms fight with introductory pricing and price wars to capture a customer they can then harvest. The authors' conclusion is that this competition for the market is usually a poor substitute for ordinary competition, that it tends to result in weaker overall rivalry and to strengthen incumbents, and that firms therefore have an excessive private incentive to pursue incompatibility.

Apply that to the illustrative example. A locked-in seat in that business is worth $16,667 more than a loose one. A rational competitor facing the same customer will spend up to that $16,667, in free implementation, migration services, a year of waived fees, or a data-conversion project done at their cost, to take the seat, because the prize is exactly that big. Every dollar of future rent you identified is a dollar someone will bid to be the one who collects it. If the market is competitive at the point of acquisition, the ex-post rent is dissipated ex-ante and you are back where you started, having simply moved the money from the harvest years to the discount at signing.

The rent survives only under specific conditions, and they are worth naming because they are what you should actually be building. It survives when the incumbent knows which customers will stay and the challenger does not, so the challenger must discount to everyone to win a few. It survives when acquisition is capacity-constrained rather than price-constrained, when the binding limit is implementation engineers rather than budget. It survives when the installed base was acquired before the market understood its value, which is why the durable switching-cost businesses tend to be old. And it survives when the switching cost itself keeps growing after signature, because then the competitor's required bid rises every year while your acquisition cost was fixed in the past.

The risk on the other side is the harvest trap, and it is the most common way a lock-in business kills itself. Pricing power over a captive customer is real, and using it hard converts a passive customer into one who is actively shopping. Every price increase, every fee added because they cannot easily leave, funds the internal business case for the migration project. Blockbuster's late fees were the archetype: a revenue line that worked precisely because customers could not avoid it, and that made switching feel like justice rather than merely a decision. The discipline is to price to the value you deliver, not to the cost of leaving, because the moment the second number sets your price, you have started an eviction countdown you cannot see.

A standard is a switching cost nobody owns

A standard is what happens when an industry agrees to make one category of switching cost disappear for everyone at once. It is the opposite move to the one the previous section described, and firms adopt it for a reason that is easy to state and hard to accept: growing the pie beats owning a small one.

The mechanics are worth being precise about. Before a standard, every seller's connector, file format, or protocol is a private lock-in: valuable to that seller, and a tax on the buyer, who must bet on which one survives. That bet is itself a cost. It suppresses the whole category, because buyers delay, hedge, and buy less than they otherwise would. A standard removes the bet. Complements proliferate because a complement supplier now builds once rather than five times. The category grows, and every participant's addressable market grows with it, but no participant can any longer charge for compatibility, because compatibility is now free.

The scale a settled standard reaches is the argument for adopting one. The Bluetooth SIG reports more than 79,000 new products qualified during 2025, and cites ABI Research for 5.9 billion Bluetooth-enabled products shipping in 2026. No single company's proprietary short-range radio was ever going to reach a number like that, and the reason is not engineering. It is that a proprietary radio requires every accessory maker to bet on you, and 5.9 billion units of annual demand is not the kind of thing that gets built on a bet.

Coordination is the hard part, which is why standards need either a body with real convening power or, failing that, a regulator. The clearest recent case of the second is the EU's common charger rule. Directive (EU) 2022/2380 mandates a USB Type-C receptacle, specifically the connector defined in EN IEC 62680-1-3:2021, across 13 categories of radio equipment capable of wired charging, from phones and tablets to headphones, e-readers, keyboards, mice and portable speakers. Member States apply the measures from 28 December 2024 for the first twelve categories and from 28 April 2026 for laptops. The European Parliament's own framing of the benefit is a consumer switching cost stated in cash and rubbish: consumers save up to €250 million a year on unnecessary charger purchases, and disposed-of and unused chargers account for about 11,000 tonnes of e-waste annually in the EU.

That is a legislature finishing a standards war the market had declined to finish, and it is the risk every proprietary-interface strategy now carries. If your differentiation is a connector, a cable, a file format, or a protocol, something a customer experiences purely as friction and never as value, you are holding an asset with political exposure. The defensible version is the reverse: adopt the standard at the interface, and compete on the thing above it that a directive cannot specify.

Who gets paid when everybody adopts

If a standard makes compatibility free, who collects? The answer is whoever owns patents that the standard cannot be implemented without, and the margins on that position are unlike anything else in the same building.

Qualcomm's fiscal 2025 Form 10-K, for the year ended 28 September 2025, splits the company cleanly. QCT, the semiconductor business, reported $38,367 million of revenue and $11,670 million of earnings before taxes: a 30% margin, on the back of physical products, fabs, supply chains and design cycles. QTL, which grants licences to a patent portfolio including rights essential to cellular standards, reported $5,582 million of revenue and $4,043 million of EBT: a 72% margin. Put the two together and QTL is 13% of the combined revenue and 26% of the combined pre-tax profit. Nothing about the licensing business ships, breaks, or needs a wafer.

Nokia's 2025 report shows the same shape at a different scale. Nokia Technologies posted full-year net sales of EUR 1,501 million on a reported gross margin of 100.0% and an operating profit of EUR 1,059 million, a 70.6% operating margin. Against group net sales of EUR 19,889 million and group comparable operating profit of EUR 2,024 million, that means the licensing segment was 7.5% of what Nokia sold and 52% of what Nokia earned. The economics are the whole point: the R&D was spent once, years ago, into a standard that the rest of the industry then built its products around, and the marginal cost of licensing one more handset maker is a signature.

The price of that position is a promise, and the promise is the reason the position exists at all. Contributing patented technology to a standard normally comes with an undertaking to license it on fair, reasonable and non-discriminatory terms. Without that undertaking the standards body would not adopt the technology, because no implementer would build on a foundation someone could withdraw. So the holder trades the right to refuse, and to charge whatever an exclusive position would bear, for guaranteed universal adoption at a rate that must be defensible. It is a smaller per-unit number applied to essentially the entire market.

The fragility is easy to miss behind those margins, and both filings show it in the same year. Nokia Technologies' net sales fell 22% from EUR 1,928 million, which the company attributes to lower catch-up net sales than the year-ago period, while stating that the annual net sales run-rate remained approximately EUR 1.4 billion. Qualcomm's QTL revenues were roughly flat, and the filing notes that from the second quarter of fiscal 2025 they no longer included royalties from Huawei, whose licence agreement had expired, while new long-term agreements with two Chinese OEMs and with Transsion were signed. Royalty income is a portfolio of bilateral negotiations that expire on staggered clocks. Reported revenue swings on renewals and back payments rather than on end-market demand, which is why both companies steer readers toward a run-rate rather than the printed line.

For a small operator none of this is copyable, and pretending otherwise wastes time. What is readable is the principle underneath it: the durable position in a standardised world is the one levied on the standard, not on the product. The equivalents at ordinary scale are the certification everyone in your trade must hold, the data set every competitor's tool has to reference, the compliance template an entire sector files against. Owning the thing that must be used is a different business from selling the thing that uses it.

What actually breaks lock-in

Switching costs do not decay gently. They hold, and then a specific event dissolves them, usually one of five. Knowing which one you are exposed to is more useful than a general sense that customers might leave.

The first is a regulator putting a date on it, which is now a live and dated risk rather than a hypothetical. The EU Data Act's Article 29 deletes cloud switching charges from 12 January 2027; Article 25 caps the notice period at two months and the migration itself at 30 calendar days. Directive (EU) 2022/2380 deleted the proprietary charging connector across 13 device categories. In both cases a commercial asset that took years to build stopped existing on a calendar date published in advance.

The second is a migration tool. A switching cost is mostly a project plan, and project plans are automatable. The moment a competitor ships an importer that reads your export format and reconstructs the configuration, a six-month migration becomes a weekend, and everything you built into onboarding depth converts from a moat into a one-time inconvenience. Watch for this specifically: the leading indicator is competitors publishing switching guides that name you.

The third is a platform shift that resets everyone's accumulated investment at once. When the substrate changes (a new operating environment, a new interface paradigm, a new regulatory regime that forces a rebuild anyway) the customer's sunk configuration stops being an asset they are protecting and becomes work they are redoing regardless. If they must rebuild either way, your incumbency is worth roughly nothing during that window, and challengers know it.

The fourth is your own price increase, and it is the only one entirely within your control. Every raise funds the business case for the migration project. There is a threshold at which the annual saving from leaving exceeds the one-time cost of leaving, and past it the customer's finance team does the arithmetic without being asked. The practical test is the one to run on yourself: if the total cost of leaving you is less than a year of your fees, you do not have lock-in, you have inertia, and inertia does not survive a budget review.

The fifth is consolidation on the customer's side. An acquisition, a merger, a new group CIO standardising three subsidiaries onto one stack: the decision to switch gets made above the person who likes your product, on grounds that have nothing to do with your product. Concentration risk and switching-cost risk are the same risk viewed from two angles, and a base of large customers is a base where a single reorganisation moves several years of revenue.

The honest summary is that lock-in is a decaying asset that has to be topped up. Every year the customer's staff turns over, their integrations get rewritten, the migration tools improve, and the statutory clock ticks. The operators who hold it are the ones who keep adding new reasons to stay, whether new data accumulating inside the system, new workflows learned, or new integrations built, rather than the ones defending an exit fee written into a contract nobody has read since signing.

Put it to work

Write down what a customer would actually have to do to leave you: hours, data, integrations, retraining, contract. Price the total. Then compare it to your annual price. If leaving costs less than a year's fees, you have no lock-in, only inertia. Track gross logo retention separately from net revenue retention, and audit your egress terms before a regulator or a competitor does.

Sources & references

Linked entries open the named source directly. Entries without a link say exactly what kind of reference they are — and how to check them yourself.

• Qualcomm Incorporated — Form 10-K for fiscal year ended 28 September 2025 (QCT and QTL segment revenue and EBT, licensing terms, Huawei and Transsion agreements)
• Nokia Corporation — Report for Q4 and full year 2025 (Nokia Technologies net sales, operating profit and margin; group net sales and comparable operating profit; run-rate)
• Automatic Data Processing, Inc. — Form 10-K for fiscal year ended 30 June 2026 (client retention estimated at ~13 years in Employer Services and ~6 years in PEO; revenue, clients, one in six US workers)
• Snowflake Inc. — Form 10-K for fiscal year ended 31 January 2026 (net revenue retention 125% / 126% / 133%; customer count; product revenue)
• Regulation (EU) 2023/2854 (Data Act) — Article 25 (switching contract terms, two-month notice, 30-day transitional period), Article 29 (gradual withdrawal of switching charges), Recital 88 (egress charges and lock-in), Article 50 (application from 12 September 2025)
• Directive (EU) 2022/2380 — the common charger directive (USB Type-C per EN IEC 62680-1-3:2021, 13 device categories, application from 28 December 2024 and 28 April 2026 for laptops)
• European Parliament — press release on the common charger rules (up to €250 million a year of consumer savings; about 11,000 tonnes of annual EU e-waste from disposed and unused chargers)
• Farrell, J. & Klemperer, P., "Coordination and Lock-In: Competition with Switching Costs and Network Effects," Handbook of Industrial Organization, vol. 3, ch. 31, pp. 1967–2072, Elsevier (2007) — citation and abstract record
• Bluetooth SIG — About us (79,000+ new products qualified in 2025; 5.9 billion Bluetooth-enabled products shipping in 2026, sourced by the SIG to ABI Research)
• The worked example is illustrative — The 400 customers, $1,000 monthly price, 70% gross margin, 10% discount rate, 8% and 18% churn rates and $18,000 CAC are invented round numbers chosen to make the arithmetic legible. They are not benchmarks. Replace every one of them with your own measured figures before drawing a conclusion.