Software Growth

Switching costs

Switching costs are the money, time and risk of leaving one product for another. The six types, the workflow tax, what AI changes, and how to build or lower them.

Switching costs are everything a customer gives up to move from your product to a competitor's: money, time, effort and risk. The one-line rule: a customer switches only when the gain from the new tool pays back the one-time cost of moving fast enough to be worth the trouble.

For software, the price difference is rarely the big number. The big numbers are moving data, rewiring integrations, retraining people, rebuilding the rules and habits a team has set up, and the risk that something breaks during the move. High switching costs keep customers from leaving even when a rival is a little better or a little cheaper, which makes them one of the most reliable parts of a moat. This guide covers the six kinds of switching cost, the "workflow tax", whether AI is lowering switching costs, the line between healthy stickiness and hostile vendor lock-in, and how a small SaaS can build the first and lower a rival's for prospects who want to move to you.

How switching costs work

Every customer runs the same rough calculation, even if they never write it down. On one side is what they would gain each month from the new tool: a lower price, time saved, a feature they need. On the other is the one-time cost of getting there. Divide one by the other and you get the number of months the switch takes to pay off:

Say a customer pays you $50 a month and a rival charges $30. Moving would take 20 hours of staff time valued at $50 an hour, or $1,000.

That is more than four years. Few customers will bother unless the rival is far better. This is why a small price gap alone rarely moves customers, and why founders who undercut an incumbent by 20 percent are often disappointed. It also shows the two levers: an incumbent keeps customers by making the move expensive or the gain small, and a challenger wins them by making the gain big and the move cheap.

Six kinds of switching cost

Six kinds of switching cost shown as cards. Financial: new licenses, setup fees and paying for two tools during the move; grows with seats and plan size. Learning: retraining the team, rewriting how-to docs and slower work for weeks; grows with team size and habits. Data and migration: exporting, cleaning and importing records, history and files; grows with years of stored data. Integration: rewiring every tool that sends data in or takes data out; grows with each tool connected. Relational: losing a support team, account manager or partner you trust; grows with years of good service. Contractual: months left on an annual plan, exit fees and notice periods; grows with contract length.
Adapted from Matt Rickard (July 6, 2021) and Jason Cohen (May 21, 2023). Schematic.

Matt Rickard notes that economists traditionally sort switching costs into three groups: financial (buying the new product), procedural (learning it) and relational (building relationships with a new company). Software adds a few of its own. Combining his list with the contract terms that Jason Cohen files under coercion gives six kinds worth checking in your own product:

  • Financial. New licenses, setup or onboarding fees, and paying for both tools while the move happens. This is the cost founders think about most and customers often care about least.
  • Learning (procedural). Rickard's point about UI and workflow: software takes training, and people get used to specific workflows. Switching means retraining, rewriting internal docs and a few weeks of slower work.
  • Data and migration. Rickard calls it data gravity: data is expensive and hard to move. History, attachments, custom fields and the links between records are the hard part, which is why a good data export matters to both sides.
  • Integration. Rickard's point is that the larger and deeper an integration surface, the harder the product is to replace. Every tool that sends data into your product or reads data out of it has to be rewired. Identity (users, groups, permissions) is a special case with the same effect.
  • Relational. A support team that knows the account, a partner who set it up, an accountant or agency who recommends the tool. These take years to rebuild with a new vendor.
  • Contractual. Months left on an annual plan, exit fees and notice periods. Cohen lists "contract lock-in" among the coercive tactics: keeping a customer through paperwork rather than by choice.

Most products have a little of several. The useful question is which ones grow over time. Contract terms end. Data, integrations and habits keep piling up every month a customer stays.

The workflow tax

In a guest post for Lenny's Newsletter on October 13, 2026, Dan Hockenmaier names switching costs as one of only four compounding advantages that matter, alongside network effects, scale economies and data. He grants that AI makes the migration itself faster and cheaper, but argues that is not where most switching costs come from.

The bigger cost is what I’ll call the “workflow tax”

Dan Hockenmaier / How to tell if a startup has a moat, Lenny's Newsletter (October 13, 2026)

He defines it as all the time a team spends learning a product and building it into how they work, plus how disruptive it is to do that again. The workflow tax is why a customer can complain about a tool for years and still renew. Exporting the records takes an afternoon. Rebuilding the approvals, templates, automations, reports and habits around them takes months.

Rob Walling and Einar Vollset make the same case in episode 855 of Startups for the Rest of Us (October 20, 2026), where "you're the system of record" is the second of eight moats they think still work. Vollset's test is whether there is canonical state recorded only in your system. His example is Gymdesk, a gym management tool: who teaches the 6 AM class on Tuesday, who has paid this month and who has a key fob live there and nowhere else.

You're the system of record. The canonical state of the business lives in your app and other systems such as payments, door access, email and accounting are wired to it. Leaving means re-entering cards, reprogramming key fobs, rewiring integrations and retraining staff.
Adapted from Startups for the Rest of Us, episode 855 (Gymdesk example). Schematic.

The switching costs are still enormous.

Rob Walling / Startups for the Rest of Us, episode 855 (October 20, 2026), system of record segment from 28:28

Walling's list of what makes leaving hard is a good checklist: what can't you export, what breaks downstream in your integrations, what do you have to retrain people on, and what do they have to rebuild? Vollset adds the part that sits outside the customer's own team: the gym's members.

You’ve got to ask people to add their credit card again.

Einar Vollset / Startups for the Rest of Us, episode 855 (October 20, 2026)

When a switch needs the customer's own customers to do something, such as re-enter a card, accept a new invite or reconnect an account, the cost is no longer under the customer's control. That is often the largest single line in a switching budget, as the worked example below shows.

Does AI lower switching costs?

Partly. Walling says plainly in the same episode (from 24:09) that he expects high switching costs to go down for some products. He describes watching an AI agent click through software and figure it out on its own, and asks what happens to the "nobody wants to retrain 50 people" argument if agents do 40 or 50 percent of the work in the tool. Vollset adds that an integration with a well-documented API now takes a day or two. Those are real cuts to the learning, data and integration costs.

Bar chart of hours a fictional 10-person team spends leaving one tool, without and with an AI agent doing the mechanical work. Migration: export and import the data falls from 16 to 3 hours, reconnect the integrations from 20 to 8 hours, a total of 36 to 11 hours, down 69 percent. Workflow tax: learn the new screens falls from 30 to 15 hours, rebuild rules, templates and habits from 40 to 32 hours, and get customers to redo their part stays at 25 hours, a total of 95 to 72 hours, down 24 percent.
Fictional example. Framing adapted from Dan Hockenmaier (October 13, 2026) and Startups for the Rest of Us, episode 855 (October 20, 2026).

What AI barely touches is the rest of the workflow tax, and two of the moats from episode 855 explain why:

  • The system of record. An agent can copy the records. It cannot make the business trust a new source of truth overnight, and it cannot make 400 members re-enter their cards.
  • Being allowed to act. Vollset's question is whether your software does something in the world or only advises. A tool that recommends a payment is easy to replace. A tool that is authorized to make the payment, like Stripe Connect, sits in the money flow, and Walling calls it very hard to leave. Moving it is a risk decision, and a faster migration doesn't remove the risk.
You're allowed to act. An advisor tool produces a recommendation that a human must review and act on. An actor tool produces the result directly, for example by moving money.
Adapted from Startups for the Rest of Us, episode 855 (Stripe Connect example). Schematic.

Hockenmaier adds a point that cuts the other way: AI also helps teams build faster, so a product can embed itself across more of a customer's work than before. A practical rule follows. If your stickiness comes mainly from "our data is hard to move" or "our UI is hard to learn", expect it to erode. If it comes from being the record the business runs on, acting on the customer's behalf, or involving the customer's own customers, it should hold up much better.

Good lock-in versus hostile vendor lock-in

Hockenmaier points out that switching costs are often seen negatively, as the reason clunky legacy software never gets replaced. His counterpoint is that a product only gets that embedded by delivering a lot of value first. Jason Cohen draws the line in his essay on willingness to pay. He separates "utility" (integrations, being the system of record, a trained team, familiarity) from "coercion" (contract lock-in, holding data hostage, raising prices once switching is expensive). Utility leaves customers with no particular desire to leave. Coercion makes them allies of any competitor who shows up.

When your contract is up for renewal, you should be very afraid.

Jason Cohen / Willingness-to-pay: Creating permanent competitive advantage for the right reasons (May 21, 2023)

Cohen was describing companies whose customers are locked in against their will. Those customers go out of their way to build the case for switching, and they warn others off the product when asked. In practice the difference looks like this:

  • Healthy: the product becomes more useful the more it is used (history, reports, integrations, team settings, templates), and the customer can still take their data with them in a usable format.
  • Hostile: exports that are missing or deliberately crippled, auto-renewing contracts with long notice periods, fees to get your own data out, and price increases timed for when leaving is hardest.

Regulators are also pushing on the hostile side. The European Commission's Data Act explainer says the switching rules have applied since September 12, 2025, that providers of Platform and Software as a Service must offer open interfaces and, at a minimum, export data in a commonly used, machine-readable format, and that switching and data egress charges are removed entirely from January 12, 2027. If you sell to EU customers, check with a lawyer how the rules apply to your product. Either way, a clean export is a trust signal, and it costs you little if the product is good.

How a small SaaS builds healthy switching costs

Cohen's warning from his moats essay applies here: "we will create switching costs" is a hope, not a strategy. You need to say which product work will cause it. For a small SaaS, these are the moves that compound:

  1. Own one piece of canonical state. Pick the record your customer would be lost without, such as the schedule, the invoice ledger or the approved content, and make your product the place it lives.
  2. Integrate in both directions. Pull data in so you can act, and push data out so other tools depend on you. Each connected tool adds to the integration cost of leaving.
  3. Act, don't just advise. Where it is safe, let the product send the payment, file the form or publish the post rather than suggest it. Earn that permission with reliability.
  4. Let value grow with time. History, benchmarks against the customer's own past, saved templates and automations make month 24 more valuable than month 2.
  5. Spread across the team. One user can switch in an afternoon. A whole team, plus their clients or members, is a project. Shared workspaces and client portals raise the cost the healthy way.
  6. Keep the exit clean. Offer a complete export and an API. It removes the fear that stops cautious buyers from signing up, and it keeps your switching costs on the utility side of Cohen's line.

You can see the result in your numbers. Products that store important data, run daily workflows and connect to other tools tend to show lower customer churn and a higher net revenue retention than standalone tools. A simple check is to compare churn rate for customers with and without a connected integration, or with and without a second active user. If there's no difference, your product isn't embedded yet.

Lowering switching costs for prospects switching to you

If you are the challenger, the incumbent's switching costs are your biggest objection. The prospect may already agree your product is better and still not move. Work through the six kinds and remove as many as you can:

  • Importers. Build a direct import from the market leader. Plausible, a privacy-focused analytics tool, imports historical Google Analytics 4 stats so customers don't lose their history. Linear has importers for Jira, Asana, GitHub Issues and Shortcut.
  • Concierge migration. Do the import, the integration setup and the first configuration for them. For a small SaaS this is a few hours of support time per customer, which you earn back in the first months of the subscription.
  • Handle the customer's customers. If cards are stored with a payment processor, ask whether they can be moved rather than re-collected. Stripe's data migration docs say imports usually finish within 10 business days of receiving valid data, while the previous processor can take from a few days to several months to produce the export. Start early.
  • Remove the double payment. Offer the first month free, or credit the months left on the old contract, so the customer doesn't pay twice during a parallel run.
  • Time the offer. Contractual switching costs drop to zero at renewal. Ask prospects when their contract ends and follow up before the notice period.
  • Train the team live. A one-hour session cuts the learning cost and gives you a champion inside the account.

Worked example

A fictional three-location yoga studio pays $249 a month for its scheduling and billing tool. A challenger charges $179 a month and saves the front desk about 6 hours a month of admin, worth $150 at $25 an hour. The monthly gain from switching is $70 + $150 = $220.

Switching with no help costs the studio:

  • Export and import data: 6 hours of staff time × $25 = $150
  • Rebuild integrations with door access, accounting and email: 12 hours of the owner's time × $60 = $720
  • Members who never re-add a card: 10 members × one month at $120 = $1,200
  • Retrain 11 staff: 11 × 2 hours × $25 = $550
  • Pay for both tools during a one-month parallel run: $179

The total is $2,799, and the payback is $2,799 ÷ $220 = 12.7 months. Most studio owners would wait.

Now the challenger makes a migration offer: a concierge import (1 hour of studio time, $25), integrations set up by the challenger's team (2 owner hours, $120), card details moved through the payment processor so only 1 member drops off ($120), a one-hour live training (11 × $25 = $275), and the first month free ($0). The total falls to $540, and the payback to $540 ÷ $220 = 2.5 months.

Bar chart of a fictional yoga studio's one-time switching cost, with no help and with the challenger's migration offer. Export and import data: 150 dollars versus 25. Rebuild integrations: 720 versus 120. Members who never re-add a card: 1,200 versus 120. Retrain 11 staff: 550 versus 275. Pay for both tools for a month: 179 versus 0. Totals: 2,799 dollars, paid back in 12.7 months, versus 540 dollars, paid back in 2.5 months, at a gain of 220 dollars a month.
Fictional example. Payback = one-time cost ÷ monthly gain.

Two lessons. First, the data was the smallest line. The members and the integrations were the switching cost, which is why an importer alone often isn't enough. Second, the offer costs the challenger one month of revenue ($179) plus a few hours of support, against a customer worth $179 a month for as long as they stay.

Common mistakes

  • Undercutting an incumbent on price and expecting customers to move, when the price gap is small next to the cost of switching.
  • Assuming a hard-to-learn UI is a moat. AI agents and better onboarding are shrinking the learning cost fastest.
  • Locking in customers by crippling data export. It turns stickiness into resentment, and customers leave the moment an alternative appears.
  • Shipping an importer and calling the migration solved, while ignoring integrations, retraining and the customer's own customers.
  • Counting a long contract as a moat. It delays churn until renewal and does nothing to prevent it.

Sources

Back to the SaaS glossary