Three systems, three truths
ISF trains candidates in Geneva for a state certification. Three systems, the payment tool, the sales tracker, the management spreadsheet, each held part of the truth about the same company. Before building anything, every number was given a written definition, a formula and a source.
§ 01The context
ISF, Institut Suisse de Formation, trains candidates in Geneva for a state certification and supports them through to employment. The institute recruits its own candidates through advertising, across several acquisition channels, and works them by phone with one team booking the appointments and another closing them. It’s a fast machine, with money coming in every day, and two founders who decide together.
§ 02The problem, as it was lived
A company growing at that speed always ends up in the same place: several systems that each work perfectly well on their own, and nobody whose job is to make them agree.
At ISF, the payment system said one thing, the sales tracker said another, and the spreadsheet kept by management said a third. Each one was right from where it stood, because each was measuring something different:
Money received
Deal closed
Commission owed
None of the three could answer a question as simple as “where is the month” on its own. The consolidation existed, it happened in someone’s head, and it had to happen again every time the question was asked.
Their ask was therefore simple to state and brutal to honor: a dashboard that answers six questions about their own company in thirty seconds.
§ 03How we worked
This is the part that matters in this case study, so it’s told in order, the way it happened.
Before day one of the work
Contract signed on a TuesdayThe intake questionnaire goes out immediately after: the team and who does what, the rituals, the access, the numbers they believe they have, their priorities, the boundary with their existing vendor. ISF sends it back completed on Friday morning, four hours before the kickoff call. That isn’t an administrative detail, it’s the structural choice of the whole engagement: the first call isn’t there to discover the company, it’s there to make decisions.
The kickoff call
53 minutesNo slides, no pitch: the sale is done, this call sells the calendar. We go back through their written answers in their own words, and we ask the five truth questions that reading the questionnaire surfaced. Two of the numbers in the questionnaire couldn’t both be true at once, not because either was badly filled in, but because they weren’t measuring the same thing: one counted what was signed, the other what was collected. We say it out loud, in week two of the relationship, rather than discovering it at delivery. We leave the call with dates on both sides, a firm deadline on access, and the team interviews scheduled inside the call itself, not promised for later.
The week of interviews
5 interviews, Monday to ThursdayThe technical vendor, the head of sales, two sellers, and a fifth person who holds both roles at once. Thirty minutes each, separately, recorded, transcribed, and summarized into one page per person. Separately is the rule, and that’s where the real value sits: five people from the same team never quite describe the same company, and the gap between their descriptions is information in itself. We don’t pick between them to make things tidy. We record the gap, we go verify it in the data, and we hand both versions back to the founders along with what the data says about it.
The numbers taken back to source
Nothing declarativeWe don’t ask for the numbers, we go get them: several thousand sales opportunities read one by one, every payment since inception, every appointment booked over six months, more than four months of call recordings inventoried, the advertising campaigns channel by channel, the application funnel from the ad through to the payment. The first pull of appointments was thrown out and redone by a different route; the second found more than half again as many. We published the second, and wrote in the report that the first was wrong and why.
The definitions document
Submitted for sign-off before the first line of codeEach of the questions the dashboard has to answer gets its own page: the client’s request quoted word for word, the definition adopted, the calculation formula, the data source, the known limits, and a separate section for the points that aren’t settled. That last section states in plain writing what we don’t know yet. These questions have no right answer in the abstract, they have the answer management chooses. The document goes to both founders, and a written approval from both is requested before construction begins. A yes on a call does not close that door, and neither does silence.
The cadence, every week
A fixed weekly call, a written reportAlways in the same four-block format: what moved, what’s in progress, the next step with its dates, and what we need from you. That last block is the only place in the project where we hand them work, and it runs to three lines. Certain stages don’t open while a written answer is missing, and that’s stated right there, in the open.
§ 04What it has already changed
The engagement is two weeks old. What follows is what has changed, not what is going to.
Trade-offs became possible. The cheapest acquisition channel was running capped by its own budget ceiling, and that can now be demonstrated with numbers rather than argued from intuition. Inbound conversations weren’t assigned to anyone by default, and the ones left without an owner rarely got a reply, including appointment cancellations that turned into avoidable no-shows. That’s a setting, not negligence, and it’s now identified and quantified.
Weeks before ISF came back on its own with new projects to build, unprompted. The signal that beats any promise.
§ 05Exactly where this case study stands
Nothing is in production yet, and the engagement runs four months.
What’s being built now, on the definitions once they’re signed: the operational dashboard that answers the original six questions plus the seventh the client added at kickoff, readable from a phone; the automation of commission calculation and of matching every payment to the seller who earned it, two tasks done by hand at every month’s end today; and a weekly loop that listens to sales calls, scores each one against a rubric, and produces feedback per person. The engagement is ramping up, not winding down.
This case study is in the portfolio because it shows what happens before a system exists. At this stage, what the client holds is the definitions document, baseline measurements across the whole business, five team interviews written up, the written list of what isn’t settled, and one report a week.
You’ll find out how it ends: this case study will be updated with what’s running, or with what didn’t work.
If you’ve already paid for a dashboard that showed accurate numbers on definitions nobody had written down, you know what that costs: a decision made on a poorly defined number costs more than no number at all. The audit starts exactly there, with what your data actually says, before proposing anything to build.