A managing director showed me, via a screen recording, how his sales team spends forty-five minutes typing in figures before every quote. I automated the repetitive clicking and left the three key decisions to human judgement. Why this approach works almost everywhere.
May 2026 · Noted by · reads in ~3 minutes
The managing director shared his screen to show me how his sales team puts together a quote. It took forty-five minutes. None of it was difficult. That was precisely the problem.
The company is based in Hamburg and has around ninety staff, most of whom work in development. It makes buildings energy-efficient. Sales are based on an energy-saving analysis: the customer sends in a questionnaire, the company calculates how much can be saved using its technology, and compiles the results into a presentation.
The managing director clicked his way through six steps. Typing out figures from the questionnaire. Transferring them to a spreadsheet with twenty-three worksheets. Standardising units, as some figures were in euros and others in cents. Calculating the payback period. Copying everything into a presentation with seventeen slides. Saving, sending off, and starting all over again with the next customer.
It wasn’t rocket science. It was hard graft, week in, week out, always the same.
The obvious solution would have been a single button: questionnaire in, presentation out. I deliberately designed it differently, and the reason applies to almost every organisation.
The value in this process isn’t the typing. It’s the three decisions made by an experienced person. What level of savings is realistic for this client? Buy or rent? Which payback period target suits this discussion? Added to that is an intuition that no script can capture. If a consumption figure looks ten times too high, a human being will pause to think. A programme carries on calculating and produces a neat, but incorrect, slide.
Anyone who automates these three decisions saves a few minutes but sacrifices the quality that drives the sale.
So I built an assistant that takes care of the tedious work and stops at precisely these three points.
The sales team opens the questionnaire. The system reads the figures, suggests a result and flags anything that looks suspicious. The human checks, corrects and decides. Then a pre-filled Excel file opens. This file performs the calculations accurately; the sales team saves it, and only then are the presentation and PDF generated. The verified Excel file has the final say – there’s no hidden model.
Forty-five minutes of typing has been reduced to a few seconds. The time saved is channelled into more client meetings. It took less than three weeks from the first meeting to the working prototype.
It would be a white lie to tell a ‘happy ending’ story without mentioning the dead ends.
One version of the input form suddenly spat out nonsense. A payback period of 0.1 years. Savings running into the millions. The cause was tiny: two different notations had swapped the thousands separator and the decimal point, turning 0.0827 into 827. You don’t spot a mistake like that from your desk. I calculated the same scenario once by hand and once using the agent, and compared the results side by side until they matched to the nearest cent.
The second stumbling block was Excel itself. A formula broke down when I recalculated it on the Mac. The solution was unspectacular: enter a fixed, verified value at that point, rather than relying on a formula that behaves differently depending on the version. Details like this determine whether a tool is used or ends up in a drawer after two weeks.
Almost every organisation has its own version of these forty-five minutes. Someone regularly pieces together a document from scattered data. Quotations, impact reports, grant applications, monthly reports.
Before you automate something like this, ask yourself three questions. Which task does someone have to do from scratch every week, even though it’s almost always the same? Where exactly does the judgement lie that a person has to make? And what happens if the data isn’t quite right?
Anyone who takes the last two questions seriously builds a tool that makes people smarter. Anyone who skips them builds one that they secretly circumvent.
45 minutes, one screen shared, an honest assessment. Not a sales pitch, I show you where I see leverage and where I do not.