Another month of runway is justified when a working demo produces evidence that a defined buyer will commit, use it in a real workflow, and return under ordinary conditions. A convincing screen recording proves the team can build; it does not yet prove that the problem deserves the next month of payroll and attention.
At 4:47 PM on Friday, Ama watched the response appear in a browser window from a small coworking room in Accra. She had built an assistant for independent retailers to turn a week of WhatsApp orders into a stock list and a simple reorder recommendation. The model caught the messy abbreviations, grouped duplicate items, and explained one odd result in plain language.
Her cofounder filmed the last run on his phone. The demo had the clean rhythm founders hope for: upload, wait, result. A prospect who had agreed to watch nodded twice and asked whether it could handle her own order history.
Ama closed her laptop feeling the familiar temptation to call the week a win.
By Monday morning, the account balance made the question sharper. Extending the work would mean delaying a contract that could cover part of the team’s costs. If they spent another month improving a product nobody adopted, they could lose the contract and arrive at the next fundraising conversation with less cash and no clearer market signal. The demo had earned attention. It had not earned that trade.
Turn the demo into a buyer commitment
The first useful question on Monday was small: which evidence would make the decision easier by Friday?
Ama wrote down three conditions. One buyer had to send a real export or let the team work from a live order thread. That buyer had to name the person who would use the output each week. And they had to agree on what would count as a useful result before seeing it.
Those conditions changed the conversation. “Would you use this?” had produced polite interest. “Can we process last week’s orders together on Tuesday, then compare the stock list with the one you already made?” asked for a decision with weight.
A buyer may still say no. That is valuable. It tells the team whether the issue is trust, setup effort, a missing workflow step, or a problem that hurts less than the founder assumed.
The alternative is easy to recognise: more work on extraction accuracy, a cleaner interface, perhaps another prompt chain, all before anyone has put their own messy input on the table. That work can make the demo more persuasive while leaving the core risk untouched.
The same distinction matters when a prospect asks where your model came from. A credible answer about the technology supports a sales conversation. The decision to fund another month rests on whether the buyer will trust the product with work that matters.
Choose one workflow where failure has a cost
The evidence has to come from a job with consequences. “This would save time” is too loose to guide a runway decision.
Ama returned to the retailer who had watched the Friday demo. On Tuesday, the retailer shared a week of real order messages, including voice-note transcriptions and product names written three different ways. Her assistant produced a draft stock list. One quantity was wrong. Another recommendation was useful enough that the retailer checked it against what was already in the storeroom.
The bad outcome still sat on the table. If the retailer had to correct every output before using it, the product could become another task for an already busy person. A faster first draft would not be enough.
So Ama stopped measuring whether the model looked intelligent and started recording a narrower set of facts: how long setup took, where the output needed correction, whether the retailer used it to make a real ordering decision, and whether she asked to repeat the process the following week.
That last behaviour carries more weight than praise at the end of a call. A repeat request means the product has found a place in somebody’s routine.
This is also why demand evidence should come before a larger build. Mina’s strong candidacy touches the same tension from another direction: competence can be real while buyer evidence remains absent.
Set a decision rule before the week begins
Founders often extend runway in pieces. One more week becomes another because the team has already spent time on the prototype, or because the next improvement feels close.
A pre-set rule makes the choice less emotional. Ama’s rule was simple: continue only if two retailers completed a live run, at least one used the output in a real stock decision, and both could describe the recurring job without her explaining it back to them. If that evidence did not appear, the team would take the contract and keep the product alive as a smaller experiment.
The rule did not promise certainty. Early-stage evidence rarely does. It gave the team a way to distinguish a product risk worth funding from a demo that had received a generous audience.
Operating costs have a way of making this discipline urgent. When money is tight, AI spending can grow because capability has become cheaper to access, while the cost of pursuing the wrong workflow stays painfully high. The constraint is not whether the team can make another impressive output. It is whether the next month buys information they could not get any other way.
Monday’s evidence was narrower, and stronger
By the end of the week, Ama’s assistant had not become a full product. It had processed two real sets of orders. One retailer corrected parts of the output but asked to run the next week through it. The other liked the concept and declined to continue because the setup took too much effort for a task they handled informally.
That was enough to change the work. Ama did not spend the next month adding more recommendation types. She focused on the input step that had made the second retailer hesitate, then arranged another live run.
Friday’s demo gave her a reason to investigate. Monday’s decision came from what happened after someone placed their own work, and their own risk, on the line.
Comments
No comments yet.