Skip to main content
Category: Effects
Type: Remaining with a product, supplier, or technical standard because the cost of moving outweighs the gain from the alternative
Origin: W. Brian Arthur, text dated June 1987 and revised May 1988. That copy states publication in the Economic Journal 99, pages 116–131 (1989), and an earlier version as IIASA WP-83-90 (September 1983). Buyer-level bill: Paul Klemperer, abstracts only, Quarterly Journal of Economics 102(2), May 1987, pages 375–394, and RAND Journal of Economics 18(1), Spring 1987, pages 138–150. Complementary assets: Carl Shapiro and Hal R. Varian, slide deck “Recognizing Lock-In,” for Information Rules
Also known as: customer lock-in; platform lock-in. Switching costs are the moving bill, not a second title for this page
Quick Answer — The Lock-in Effect is remaining with a product, a supplier, or a technical standard because the cost of moving outweighs the gain from the alternative. Arthur’s model locks a market onto one technology path when further adoption makes that path more attractive. Klemperer’s 1987 abstracts use the same words for a buyer who is only partly locked in by learning, by transaction costs, or by a firm-made cost such as a lost repeat discount. A UK account-switch count shows completed moves after one of those costs is cut. It does not show that every user is stuck.

​
What is Lock-in Effect?

The Lock-in Effect is remaining with a product, a supplier, or a technical standard because the cost of moving outweighs the gain from the alternative.
It shows that the economy, over time, can become locked-in by “random” historical events to a technological path that is not necessarily efficient, not possible to predict from usual knowledge of supply and demand functions, and not easy to change by standard tax or subsidy policies.
That sentence is the abstract of the Arthur text read here. His lock-in is a market share. The buyer’s lock-in is a different object that uses the same words. You stay with a phone, a bank, or a chat app because the next step costs more than the other option is worth. That step may be relearning the controls, moving the files, rebuilding the payments, or leaving people who did not move. An everyday picture is a key ring. The new lock may be better. The cost is recutting every key that still has to open something. Two familiar pictures use different stocks, and this page prints no retention rate for either. Apple’s devices are the usual picture of hardware, of purchases and files that do not all travel, and of gestures you have learned. WeChat is the usual picture of coordination. The household and the coworkers are on one chat app, so exporting your own thread does not move the group. Status Quo Bias is a preference for the current option when the move itself is cheap. The Lock-in Effect needs a real moving bill. Sunk Cost Fallacy is letting money already spent decide the next step. A switching cost is still ahead of you. Network Effects raise the value of a system because other people are on it. That can hold you even after your own file would export.

​
Lock-in Effect in 3 Depths

  • Beginner: A salary and a set of household bills land in one bank account. Leaving means teaching every payer the new number. The teaching is the lock. The logo is not.
  • Practitioner: Before you adopt a tool, write three lines. What you would have to relearn. What file would not come out. What the new supplier would have to rebuild. Shapiro and Varian’s slides say the data stock and the brand-specific training rise with time. A durable machine’s hold often falls as the machine ages. The three lines will not all move the same way.
  • Advanced: Arthur’s model can absorb a market into one of two unsponsored technologies, because an early chance in the order of adopters is magnified when returns rise with adoption. A standard tax or subsidy need not unlock it. In his setup the adjustment required grows without a bound. Klemperer’s RAND abstract says buyers are partially locked in, second-period demand is less elastic, and prices are lower in the first period than later, because the later share is valuable. “May be higher in both periods” is his wording. It is not a law that every introductory price is a trap.

​
Origin

The market model and the buyer’s bill were written in the 1980s. They do not share one sample. Neither paper is a study of a phone or a chat app. The copy read for the market model is W. Brian Arthur, “Competing Technologies, Increasing Returns, and Lock-In by Historical Events.” The text is dated June 1987 and revised May 1988. Its first page says an earlier version appeared as IIASA Paper WP-83-90 in September 1983, and that the paper was published in the Economic Journal, 99, pages 116–131, in 1989. Arthur was Morrison Professor of Population Studies and Economics at Stanford University. The journal printing was not re-read. Quotations are from the 1988 text. He studies two open technologies, A and B, that no firm prices. He calls those unsponsored. A note says sponsored technologies are proprietary and can be priced. Agents are of two types, R and S, in equal numbers. An R agent naturally prefers A. An S agent naturally prefers B. Each agent picks the version available at that moment and uses that version from then on. The payoff depends on how many people have already chosen each technology. When returns rise with adoption, an early run of the type that prefers an initially attractive but slow-to-improve technology can lock the market into that inferior option. Equal development of the excluded technology would, in his setup, pay both types better in the long run. He calls the process flexible when a subsidy or tax can always influence later choices. Under increasing returns that flexibility fails. The adjustment needed to shift the barriers increases without a bound, and a lead that has begun to emerge becomes progressively more locked in. Expecting the lock-in hastens it. He points to the QWERTY keyboard, citing Paul A. David (1985), as a case where the increasing returns come mainly from coordination rather than from learning by using. This page does not re-read David, so it carries no typing-speed gap. The buyer’s wording is Paul Klemperer. Two abstracts were read. The articles were not. “Markets with Consumer Switching Costs” is in the Quarterly Journal of Economics, volume 102, number 2, May 1987, pages 375–394. The Oxford record places him at St. Catherine’s College and the Institute of Economics and Statistics, Oxford. Products that look the same before the first purchase can be set apart afterward by switching costs. Those costs include learning costs, transaction costs, or artificial costs imposed by firms, such as repeat-purchase discounts. The noncooperative outcome may be the same as the collusive outcome of an otherwise identical market without those costs. The prospect of the later profits leads to vigorous competition for market share at the start. “The Competitiveness of Markets with Switching Costs” is in the RAND Journal of Economics, volume 18, number 1, Spring 1987, pages 138–150. Consumers are partially locked in by switching costs they face in the second period. Demand is more inelastic then, and also in the first period, because a firm with a higher share charges more later and is a worse firm to be attached to. Prices are lower in the first period than later. Prices may be higher in both periods than they would be without switching costs. Carl Shapiro and Hal R. Varian, in the slide deck “Recognizing Lock-In” for Information Rules, treat the gap between staying and leaving as durable investments in complementary assets. Their three labels are hardware, software, and wetware, the skill in the user’s head. Total switching costs are the customer’s costs plus the new supplier’s costs of putting that customer back in the same position. Their seven groupings are durable purchases, brand-specific training, information and data, specialized suppliers, search costs, loyalty programs, and contractual commitments. Training and data rise with time. A durable purchase usually falls with time. The slides’ dollar lines are classroom arithmetic. They are not on this page.

​
Key Points

Four cuts keep a personal moving bill, a cost that grows with use, a cost the new supplier pays, and a market path from inheriting one another.
1

The bill is paid only if you leave

Klemperer’s May 1987 abstract lists learning costs, transaction costs, and firm-made costs such as a lost repeat-purchase discount. None of those is the price you already paid. A student who has learned one editor, a household that has taught every payer one account number, and a team that loses a volume discount if it splits the order are three bills of that kind. The Lock-in Effect is the comparison of that bill with the gain from the other option. If the gain is larger, the abstracts do not say you must stay.
2

Training and files tend to grow; a machine's hold often shrinks

Shapiro and Varian’s slides do not give every cost the same clock. Brand-specific training rises with time. Information and data rise with time. A durable purchase usually falls with time, because the hold depends on what the machine is still worth. Search costs, loyalty programs that make you rebuild a points balance, contractual commitments, and specialized suppliers are separate lines. A phone bought three years ago and a chat history that grew every month are not one stock. Write the direction next to each line before you call the whole product locked.
3

Add what the new supplier must spend

The same slides define total switching costs as the customer’s costs plus the new supplier’s costs. A rival bank that cannot redirect a salary has not made a cheap offer just because its monthly fee is lower. A rival app that cannot open the old files asks you to rebuild the library. The slides’ principle is to look ahead and reason back. The later bill is part of today’s choice. That is also why Klemperer’s first period can be the cheaper one.
4

A market path is not one person's export button

Arthur’s agents do not pay a personal fee to the owner of a standard. In the rising-returns case the market can still be absorbed into one technology, including one he calls inferior for the long run. A standard tax or subsidy need not release it. He is explicit that the simple model is unsponsored. A priced ecosystem is the sponsored case, which that model sets aside. Use his result for a shared standard that becomes harder to leave as more people adopt it. Use Klemperer’s abstracts for a buyer who faces a bill to change supplier. Do not paste either result onto a retention rate the papers do not print.

​
Applications

Use the moving bill, its direction over time, and the new supplier’s share of it. A famous logo is not a measurement.

Write the three stocks before you join

Hardware, software, and wetware are the slides’ names for the machine, the files and programs, and the skill in your head. A first laptop, a household printer, and a team’s editor each have a different mix. If the wetware line is the long one, a short trial on the real task tells you more than a spec sheet.

Mark which lines grow while you stay

The slides say training and data rise with time, and a durable machine’s hold usually falls. A family chat log and a student’s course files are rising lines. A phone handset is often a falling one. Export, or leave, before the rising line is the one you cannot face. Do not wait for the handset to feel old if the archive is the lock.

Price the rival's rebuild, not only the fee

Total cost includes what the new supplier spends to restore your position. Ask whether payees, files, and learned steps survive the move. A lower monthly price that strands the salary is a different offer from a lower price that redirects it. Klemperer’s early competition for share is not proof that the later bill is zero.

Separate a ticked box from a migration

Default Effect is staying with an option that was already selected, often when changing the tick is cheap. Endowment Effect is asking more to give up what you hold than you would pay to acquire it. Neither one is a file that will not export. If the export is one click and you still stay, you have not measured the Lock-in Effect.

​
Case Study

The counted case is a payment trail between UK banks. It is not a handset, and it is not a circle of friends. A current account collects incoming payments and a set of outgoing ones. Pay.UK’s Current Account Switch Service, which Issue 51 of its dashboard dates to a launch in 2013, is the system used to move direct debits, standing orders, and bill payments. A full switch closes the old account. It adds the Switch Guarantee and payment redirection. The clock printed for that work, in this issue, is seven working days. Someone who leaves the old account open is still counted, and does not receive the guarantee or the redirection. Issue 51 covers 1 April 2026 to 30 June 2026. The central engine had completed 12.9 million switches since launch. The quarter itself was 288,999 switches, across 53 participating brands. In the twelve months to 30 June 2026 the count was 1,216,271. In that quarter, 99.3 percent of switches were completed inside seven working days. Awareness came from a monthly online omnibus of people aged 18 and over, with a sample of over 2,000 in Great Britain and Northern Ireland. It was 75 percent. The question was whether the person had heard of the Service. The dashboard says an awareness target of 75 percent was set in the middle of 2015. This quarter meets that target. It is not an estimate of how many people feel stuck. Satisfaction among people who had used the Service in the last three years was 92 percent. Of those who switched in the quarter, 74 percent said they preferred the new account, and 2 percent said it was worse. Online or mobile banking was named by 46 percent, interest earned by 33 percent, customer service by 29 percent, and spending benefits by 25 percent. Those are separate answers. They are not shares of one pie. The lesson is narrow. Once that payment trail can be moved, the Service records millions of completed switches, and 74 percent of this quarter’s switchers prefer the account they moved to. The lock being cut is the trail. The 74 percent is a preference after the move. It is not a measured switching cost. The dashboard does not print the number of current accounts in the country, so 12.9 million is not a share of all accounts. Gains and losses by brand on the same PDF are for January to March 2026, three months in arrears. They are not the quarter’s 288,999. A note on the first page says a temporary fault at one participating bank made the count of redirected payments abnormally large this quarter. That payment count is not used here. One boundary belongs in the story. A chat circle and a library of bought apps are not direct debits. Redirecting a salary does not move your friends, and it does not open a file format the new tool cannot read.

​
Boundaries and Failure Modes

The Lock-in Effect fails when the moving bill is imaginary, when the market model is pasted onto one customer, or when one mechanism is asked to explain a different stock. Arthur’s absorption result is for unsponsored technologies whose payoff rises with the number of earlier adopters. A firm that sets a price, bundles a store, or can change the format is the sponsored case he set aside. Do not treat a handset’s price as a test of his unbounded tax. Klemperer’s “partially” is doing work. A bill can be real and still smaller than the gain. The UK switches are the illustration. The trail was costly enough to justify a national redirection service, and people still left once that service ran. A second boundary is the clock. Training and data that rise with time do not describe a machine whose remaining value is falling, or a contract that ends on a date. Loyalty points that you can rebuild are a constructed bill. They are not a file format. Search costs fall once you already know the rival. Calling all seven of the slides’ lines a sealed door erases the directions they printed. The common misuse is to read a large installed base as proof that each person faces a high personal bill. Network Effects can make staying attractive because other people are there. That is the coordination case. Your own export can still be easy. A hard export can hold you when your friends have already left. Headcount is not the bill.

​
Common Misconceptions

Three retellings turn a partial bill into a sealed door, a past purchase into the lock, or a crowd into a file format.
Klemperer’s RAND abstract says partially locked in. The UK service had completed 12.9 million switches since its 2013 launch, and 99.3 percent of the second-quarter 2026 switches finished inside seven working days. Arthur’s model does say that the tax needed to shift a locked technology path increases without a bound. That sentence is about his unsponsored adoption process. It is not a claim that a household cannot close a bank account.
The price of the laptop, the years of subscription, and the time already used on the first screens are spent whether you leave or stay. Sunk Cost Fallacy is the mistake of letting that spent money decide. Klemperer’s switching costs are learning still to do, transactions still to run, or a discount you lose only if you move. Shapiro and Varian add the new supplier’s rebuild. A past receipt does not appear in that sum.
Arthur separates coordination, his reading of the keyboard case, from learning by using. Friends who did not move can hold you after your own history would export. Files and gestures can hold you after your friends already use several phones. The UK case is a third stock: payments tied to an account number. Redirecting that trail does not export a chat. Exporting a chat does not redirect a salary. One story does not measure the other.
The Lock-in Effect sits next to a preference for the current option, a past spend, a crowd, a ticked box, and a higher price for giving something up. Each of those pages has its own mechanism.

Status Quo Bias

The current option wins even when moving is cheap. The Lock-in Effect requires a moving bill that outweighs the gain.

Sunk Cost Fallacy

Money already spent is allowed to decide the next step. A switching cost is a cost you pay only if you leave.

Network Effects

The system is worth more because other people use it. That coordination can hold you when your own file would already export.

Default Effect

An option that was pre-selected keeps its place. Changing a tick can be cheap. A migration is a different task.

Endowment Effect

People ask more to give up what they hold than they would pay to acquire it. A file that will not open elsewhere is not that gap.

​
One-Line Takeaway

Before you stay, write the cost you would pay only if you left, and add what the new supplier must rebuild. A past receipt and a large crowd are different sums.