How Can a Business Owner Test Whether an AI Tool Can Be Left?
A business owner can test whether an AI tool can be left by conducting Dr. Jonah Tebaa's timed, four-hour rehearsal across five steps: listing assets, attempting exports, rebuilding a minimal version on an alternative tool, tracing identity items, and scoring results. Using the Own, Borrow, Rent framework, owners classify four asset classes—records, knowledge, configuration, and identity—and calculate transfer days. In a composite case, testing the top ten questions revealed true vendor dependencies and switch timelines.
A furniture retailer in the Beirut area gets an email from the company that runs its WhatsApp quote assistant. The monthly fee, USD 180 for fourteen months, will become USD 420 at the next billing cycle. That is a rise of 133 percent, with 30 days to accept. Dr. Jonah Tebaa uses this scene, which he describes as a composite of situations he sees rather than a single client, to make a point about a question most owners never ask while a tool is working: could we leave?
His answer is that nobody knows until someone tries. A contract states what a vendor promises. It does not show what an owner can actually carry out of the door, and it says nothing about how long a phone number takes to move.
Four kinds of thing a tool holds
Dr. Tebaa starts by sorting what an AI tool holds into four classes. Records are the conversation history and customer data. Knowledge is the set of answers, price lists and FAQs the tool draws on. Configuration is the instructions, tone, rules and workflows. Identity is the phone number, sender address, domain, accounts and payment method.
In his composite, the retailer's setup breaks into seven assets. About 11,000 conversation threads export as a spreadsheet in twenty minutes. The 212 product and price answers live only in the vendor's dashboard with no export button. The 640-word instruction block copies in a minute, but nine of its 23 rules depend on features unique to that vendor. Customer tags cannot be exported at all. The WhatsApp Business number is registered under the vendor's partner account. The domain and billing are in the owner's name. And no one at the vendor is named as a handover contact.
A rehearsal, not a migration
The method Dr. Tebaa describes is a timed exercise that adds up to four hours. The owner lists the assets in thirty minutes, tries to pull each one in sixty, rebuilds a minimal version in ninety, traces who holds the identity items in thirty, and scores the result in the last thirty.
The rebuild is the part he stresses. On a second, inexpensive tool, the team sets up only the ten most frequent customer questions from the exported material. It is deliberately small. In the composite, those ten questions covered about 68 percent of a month of inbound messages, and the second tool handled eight of them acceptably. The two failures both depended on the vendor-specific rules, so the gap turned out to be two answers rather than a whole system.
The identity trace produced the larger surprise. Transferring the WhatsApp number takes five to ten working days, during which messaging pauses unless the switch is planned. The owner had assumed three days. The honest figure was fourteen.
Own, Borrow, Rent
Each asset then receives one of three labels and a count of days needed to leave. Own means the business holds it and can take it. Borrow means it can be recovered with effort, usually hours of manual copying or a dependence on the vendor's goodwill. Rent means it cannot be recovered, or only the vendor can release it.
The retailer scored three Own, two Borrow and two Rent. Dr. Tebaa notes that the useful part is what each label asks the owner to decide. An Own asset needs a current copy kept somewhere the business controls. A Borrow asset needs a named person and a schedule, because copies go stale. A Rent asset needs a price: either the terms are negotiated, or the cost of replacement is written down.
What changed after the test
The retailer did not leave. According to Dr. Tebaa, that is the usual outcome and not a disappointment. Armed with a credible alternative and a clear list of what was at stake, the owner negotiated the new fee down to USD 260 a month, asked for the phone number to be registered in the business's own name, and began copying the answer library into the company's own files every month. The test changed the price of staying.
He also recommends repeating the exercise twice a year, with the owner, the person who answers customers daily and the person who holds the logins and pays invoices. The setup drifts as new tags, rules and numbers accumulate, and each addition quietly moves an asset from Own toward Rent. A scheduled rehearsal catches that while it is still inexpensive to reverse.
For owners who want a place to start, Dr. Tebaa's advice is plain: run the five steps, keep the one-page scorecard with the contract, and put the next review date on the calendar. The full article, with the five steps and the scorecard template, is on jonahtebaa.com under the title "A Four-Hour Test of Whether You Could Leave Your AI Vendor."
Related evidence: The European Commission's guide to the EU Data Act says its switching rules exist so that customers can switch from one provider of data processing services to another quickly and smoothly, without losing data or application functionality, and names high data-egress charges and lengthy procedures as the barriers. (the European Commission's explainer on the Data Act's switching rules)
The UK Government Digital Service defines lock-in as the point where switching from one technology or provider to another is difficult, time consuming and disproportionately expensive, and tells organisations to retain ownership and access to any data held by third parties and to monitor where exit or switching costs could grow. (the UK Government Digital Service's guidance on managing lock-in)
