Where does a small business lose information when work moves between its AI tools?

Most small businesses that use AI tools describe their problems as tool problems. The chat assistant summarised something badly, the spreadsheet formula was wrong, the quote looked off. Dr. Jonah Tebaa, an AI strategist based in Lebanon who advises operators across the MENA region, argues that this diagnosis is usually aimed at the wrong place. In his work, the information a business loses rarely disappears inside a tool. It disappears in the moment a person moves it from one tool to the next.

He calls those moments carries, and he has built a short method around them, the Carry Map.

A composite scene about an expiry date

Dr. Tebaa explains the idea through an illustrative composite rather than a case study. A Beirut tile importer receives a supplier price update on WhatsApp that ends with a validity line: the prices hold for seven days. The operations manager pastes the message into a chat assistant, which summarises it correctly, validity included. She then copies the prices into a spreadsheet that has no column for validity. When a contractor asks for a quote, the quote template prints its default wording, which says thirty days.

Nobody notices for two weeks. When the contractor orders at the quoted price, the supplier's price has already moved. In Dr. Tebaa's telling, nobody in that office made a careless error, and no tool malfunctioned. A line of context was simply left behind at one hand-over, and a default filled its place. He is explicit that the scene is a composite, assembled from the shape of situations he has encountered, and that its figures are illustrations rather than research results.

What counts as a carry

For Dr. Tebaa, a carry is any step where a person connects two tools by hand. Selecting and copying, retyping, forwarding a message, or exporting a file and uploading it elsewhere all qualify. His observation is that owners undercount them because the action is so small. A copy and paste takes seconds, appears in no budget and leaves no trace, so a job that seems to involve two tools often involves five.

He adds a structural point. Each tool has a login and an owner, but the gap between two tools has neither. Responsibility for what crosses it is, in practice, nobody's.

The Carry Map in three parts

The Carry Map is Dr. Tebaa's name for a three-step exercise that can be done on paper in about an hour.

The first step is to list the carries. The business takes one real job it repeats, follows it from the first input to the final output, and writes each hand-move between tools on its own line.

The second step is to rate what each carry drops. He uses three letters. F stands for Format: columns, currency, line breaks, and in a bilingual market the choice between Arabic-Indic and Latin digits. C stands for Context: the reason behind the content, such as a validity date, an exception or the identity of the customer. S stands for Sign-off: who last looked at the work, and what they compared it with. A carry earns a mark for each letter that has gone missing in the past month. The carry with the most marks is the leaky one, and any C mark on a price or a date overrides the count.

The third step is to fix one seam, in a fixed order. The cheapest repair that closes the worst carry is the correct one.

Why Context is the one he watches

Format losses, in Dr. Tebaa's experience, announce themselves. A number that will not add up, or two lines fused into one, looks wrong and someone fixes it. Context loss does not announce itself. After the carry the content is still tidy and plausible, and nothing signals that a clause has gone. The importer's quote looked complete. That is why he treats a missing C on money or dates as the first thing to repair, whatever the mark count elsewhere.

Sign-off losses are quieter still. In the composite, one person checked the quote against the spreadsheet, and the sender assumed it had been checked against the supplier. Each did what was asked. The check simply compared the output with something that was already incomplete.

The order of repairs, and the case against buying first

Dr. Tebaa's fix hierarchy has three rungs. The first is a paste template: a short block of fields that travels with the content, for example Source, Valid until, Customer and Checked by. The second is a single source file that both tools read, so that nothing is copied at all. The third is a connector or an agent, and he reserves it for cases where the first two still leave marks after two weeks of real use.

The ordering reflects a view about where money should go. A faster connection does not decide what matters. If a hand-carried step drops an expiry date, an automated step drops it at higher speed. Defining the fields first means that any later automation has something concrete to preserve. He acknowledges the pull towards buying a bigger tool or an agent that promises to do everything, and he does not dismiss it. His argument is that the cheapest repair should be tried while the problem is still small enough to see on a single sheet of paper.

Once the chosen seam is repaired, he suggests walking the same job once more to recount the marks, and then stopping. The aim is one demonstrably better carry, not a map of the whole company.

Where to start, and what to leave

He recommends beginning with work that is repeated weekly, finished often enough to have recent examples, and tied to money or dates: quotes, invoices, meeting follow-ups, product listings. Rarer, one-off tasks are poor candidates, because there is no repeated path to draw. He also advises against mapping everything at once, since a long map is seldom finished and a finished map for one job beats an abandoned one for ten.

The practical lesson he draws is modest and concrete. Before replacing a tool, an owner can ask where a price or a date leaves one tool and enters the next, and look at that single place first.

Frequently asked questions

Why does Dr. Jonah Tebaa treat the space between tools as a separate thing from the tools themselves?

Dr. Jonah Tebaa argues that each tool has an owner and a login, but the hand-carried step between two tools has neither, so what crosses it is nobody's responsibility. In his composite example, the assistant summarised a supplier's seven-day price correctly; the expiry date was lost when a person copied the result into a spreadsheet. He therefore examines the carry separately from the tool that produced or received the content.

How does the Carry Map's Context category differ from Format in his example?

In Dr. Jonah Tebaa's Carry Map, Format covers the visible shape of content: columns, currency, line breaks, Arabic or Latin digits. Context covers the reason behind it, such as a validity date, an exception or who the customer is. Format losses usually look wrong and get fixed. A Context loss leaves tidy, plausible content, so in his example a lost seven-day limit went unnoticed for two weeks.

In what order does Dr. Jonah Tebaa suggest fixing a leaky carry, and why is the cheapest fix first?

He suggests a paste template first, a single source file that both tools read second, and a connector or agent only if marks remain after two weeks. The cheapest fix comes first because a faster connection does not decide what matters. An automated step would drop an expiry date as readily as a hand-carried one, so defining the fields comes before any automation.

Which kinds of small-business jobs does he suggest mapping first, and which would he leave alone?

Dr. Jonah Tebaa suggests starting with jobs done at least weekly that involve money or dates, such as quotes, invoices, meeting follow-ups and product listings. He would leave one-off tasks alone, because they have no repeated path to draw, and he advises against mapping every job in the company at once. One repaired carry on one job is the target.