The honest number for the jobs-created box counts new capability, not payroll preserved or removed. If an AI product saves labour but creates no new work, customer capacity, or path into higher-value roles, entering a flattering figure only hides the product decision the form has exposed.
At 8:12 a.m., the ECOWAS award form is still open on my screen. One field asks how many jobs the product has created.
I can count the people on payroll. I can count contractors. What I cannot do honestly is treat every hour automated as a job created. The product handles work someone previously did, and the form has forced a harder question: did we use that saved time to build more capability, or did we merely reduce the labour required?
What bank tellers reveal about automation
When automated teller machines spread through US banking in the 1980s and 1990s, the obvious prediction was fewer bank tellers. The machine could dispense cash, accept deposits, and handle routine transactions without a person behind the counter.
Yet US teller employment grew for years.
Economist James Bessen examined this in his research on technology and work, including his 2015 book Learning by Doing. ATMs reduced the cost of operating a branch. Banks opened more branches, and those branches still required tellers. The teller’s work also changed. Routine cash handling became less central, while customer service and product-related tasks became more important.
That outcome was not guaranteed by the ATM itself. Banks could have installed machines, reduced headcount, and stopped there. Employment grew because lower operating costs supported expansion and because the surrounding role acquired different responsibilities.
This is the useful distinction for an AI founder completing a jobs-created field. Automation creates the possibility of removing work. The company decides whether the released capacity becomes lower payroll, greater output, a new service, or a more capable role.
Count the change your product made possible
A founder can automate invoice matching for a business in Accra and report that the finance team saves time each week. That is a valid product result. It is not automatically job creation.
The stronger question is what happens next.
Can the same team process more suppliers without missing records? Can one employee move from copying fields into checking exceptions and resolving disputed invoices? Can the customer serve a second market without building another administrative layer? Can a junior operator learn to supervise the workflow, investigate failures, and improve the underlying process?
Those are capability changes. They can be observed and, with care, measured.
The distinction also affects product design. An AI system built only to replace a task tends to optimise for completion: take the input, produce the output, remove the person. A system built to increase capability must expose uncertainty, preserve review points, and help someone make a better decision.
That concern appeared in what an incomplete supplier record taught Ama about trust. Completing a workflow is less valuable when the system conceals the missing information a person needed to judge the result.
Three numbers belong beside “jobs created”
I would not answer the award form with a single impressive figure unless I could trace it to a real hiring decision. I would keep three separate counts.
First, direct jobs: people hired because the product or company exists. This includes employees and recurring paid roles, provided the definition is stated clearly.
Second, capacity created: additional transactions, customers, projects, or markets a team can handle with the same headcount. This is usually where an early AI product has its clearest evidence.
Third, role progression: people who moved into work requiring more judgement, customer contact, technical skill, or operational ownership because routine steps were automated.
These measures should never be collapsed into one number. Ten people saving time does not equal ten jobs. A customer processing twice as much work does not prove that anyone gained responsibility or income. A new title does not prove that the role improved.
The evidence sits in before-and-after workflow records, customer interviews, changed responsibilities, new hires, and actual expansion decisions. Where that evidence is absent, write zero or “not yet measured.” An award entry can survive a smaller number. A product strategy built on an invented one cannot.
Design for the second decision
The first decision is which task the AI should perform. The second is what the customer can do once that task requires less labour.
That second decision belongs in product discovery. Before building another automation, ask the customer what currently waits behind the bottleneck. More orders? Better exception handling? A delayed market launch? Time for someone to speak with customers instead of moving data between screens?
Then design and measure against that next action.
The ATM story matters because the machine’s labour effect came from what banks did around it. The cheaper transaction supported more branches, and the teller role shifted. The technology opened an option; operating choices determined the employment outcome.
At 8:12 a.m., the safest answer in the jobs-created box may be zero. Beside it, I would keep the harder evidence: which customer can now do more, whose role has changed, and what new work exists because the product shipped.
Comments
No comments yet.