An AI workflow is profitable only when the price covers the conditions required to deliver it. In Lagos, that means counting backup power, battery wear and outage delays alongside model fees, engineering time and cloud costs.
Consider Tunde, an invented composite of founders I have worked around in product teams. At 2:10 on a Thursday afternoon, he was in a small office in Yaba, watching the final batch of customer records move through an AI document workflow. His laptop sat beside a warm bottle of water. A client expected the processed files before the end of the day.
Then the power went out.
The margin disappeared in three hours
Tunde had priced the project carefully, or so he thought. He had estimated the hours needed to configure the workflow, review its output and correct exceptions. He had added API usage, hosting and a margin that made the contract worth accepting.
His spreadsheet assumed electricity would remain available.
The office battery carried the router and laptops for a while. As its charge fell, Tunde had to decide whether to pause processing or start the generator. Pausing put the delivery at risk. Running the generator meant buying fuel that had never appeared in the quote.
He chose the generator.
The workflow came back, but the outage had already changed the job. A batch stopped halfway through and needed checking before it could be restarted. The engineer responsible for validation had another commitment that evening. Tunde’s promised delivery window was closing, and the client had tied the files to work scheduled for the next morning.
For one uncomfortable hour, late delivery was the likely outcome. That meant an apology, a damaged first engagement and a smaller chance of winning the follow-on contract.
The files went out that evening. The invoice stayed the same.
The profit did not.
Infrastructure belongs inside the product cost
Founders often cost AI work around the visible system: model calls, cloud services, software subscriptions and human time. Those numbers are easy to retrieve, so they feel complete.
The operating environment adds another layer. Power interruptions can consume generator fuel, shorten battery life, interrupt uploads and force work into hours already allocated elsewhere. A three-hour outage rarely creates only three hours of cost. It can also create recovery work, rushed review and a delayed handoff.
Battery wear is especially easy to ignore because no receipt appears after each project. The replacement cost arrives months later, detached from the contract that helped consume it. By then, the founder may treat it as an office expense rather than a delivery cost.
That accounting choice makes weak margins look healthy.
The same principle applies outside Lagos. A founder in Berlin may need to count compliance review. A small team serving US clients may absorb late working hours because of time-zone overlap. The hidden cost changes by market, but the pricing question remains the same: what must stay available for this promise to be kept?
AI workflows depend on more than the model. Their cost includes the surrounding system that keeps data moving, people reviewing and delivery dates intact.
Price the promise before sending the quote
The next morning, Tunde reopened the spreadsheet. He did not add a vague contingency percentage and move on. He traced the delivery from the client’s file upload to the final handoff and marked every point where local conditions could create another expense.
Could processing pause safely during an outage? How much work would need to be repeated after an interruption? Which tasks required a live connection? How long could the battery support them? When would generator use become the cheaper option than a missed delivery?
Those questions changed the commercial model.
For the next proposal, he separated setup work from recurring processing. He added room for backup power and recovery time. He also changed the delivery promise so that one interruption would not turn an ordinary job into an emergency.
This is the same discipline behind making a hidden cash step visible before launch. In Youssef’s Monday deadline, the risk becomes manageable once the team names the step the plan had ignored. Tunde’s missing step was operational resilience.
A quote should survive the place where the work will happen. If the margin depends on uninterrupted power, immediate approvals or perfect model output, the margin exists only on the spreadsheet.
Build outage economics into the workflow
Tunde’s revised process began before the next invoice.
He recorded backup-power spending against each delivery instead of burying it in general overhead. He measured how much rework followed an interrupted run. He scheduled larger batches earlier, leaving time to recover without turning the client’s deadline into a crisis. Where possible, he added checkpoints so processing could resume without repeating the entire job.
These changes did not remove outages. They reduced the number of ways an outage could erase the project’s return.
They also improved the product itself. A workflow that can recover from interruption is easier to operate, easier to price and safer to promise. That matters when a small team has limited runway and every apparently profitable contract competes with product work.
On the next Thursday afternoon, Tunde still had the generator nearby. The difference was in the invoice, the schedule and the workflow state saved before processing began. If the lights went out again, he knew what the interruption would cost and where the job would restart.
Comments
No comments yet.