For the AI Strategy Lab, our team built a prototype for DataVance, a small data consultancy that builds dashboards for SMEs without their own data team. The app takes a sales-call transcript and the consultant’s notes and turns them into a one-page proposal, a sample dashboard on illustrative data and a cover email, with the consultant checking every step. One of my teammates works at DataVance and built the tool. My part was mainly the slide deck and the economics behind it. A lot of that work happened after the tool was already built, and the input figures came from DataVance itself, through my teammate’s conversations with the owners.
Working from the outside
At the start we all came up with several ideas, but we could only pick one. DataVance made sense: a real company, real problems and a teammate with direct access. Looking back, that choice also shaped what the rest of the project looked like for me. Because only one of us had contact with DataVance, he had information the rest of us didn’t, and that difference grew once he started building. Testing the tool and getting feedback on it mostly happened at DataVance. For me that made it hard to give usefull feedback myself
So I ended up working mostly on the part I could do without that access: turning what we had into a deck and checking whether the business case actually added up. That meant spending a lot of time with the numbers.
The numbers
The co-founder estimated that a consultant spends four to six unbilled hours on a proposal, and that about 70% of prospects don’t convert. With the tool, we assumed preparation would take one hour instead of four. That is conservative compared to the 36.5 minutes in our evaluation demo, but those were synthetic leads, so we didn’t want to rely on them. At €95 an hour, the preparation cost per lead drops from €380 to €96, about 75% less. On its own that looks like a strong result.
It looks less strong once you calculate per client won instead of per lead. At 30% conversion, DataVance prepares roughly three proposals for every client it actually wins, so the preparation cost per sale is €96 ÷ 30% = €320. Put that next to a €2,500 starter project with 20 hours of delivery (€1,900) and only €280 is left, about 11% of the price. I expected the results to be a bit more significant than that, although it wasn’t really a surprise either. The tool reduces the cost before the sale, but the package itself is priced too tightly.
This changed our recommendation. Before the calculation, the €2,500 package was simply part of the offer. Afterwards, our final slide said DataVance should change the offer before scaling: charge at least €2,960 or limit delivery to around 16 hours, to reach a 25% contribution.
The numbers also made something clear about scale. According to the owners, DataVance gets around three or four leads a month. The saving per lead is real, but at that volume it doesn’t add up to much in total, especially once you include the costs our calculation left out, like building and maintaining the app, hosting and licences. A firm with many more leads would get much more out of the same tool.
Limited leads
This led me to a thought that sits a bit uncomfortably next to our own deck. If you only have three or four prospects a month, every one of them matters. Handing the proposal over to AI then feels a bit lazy to me. Not because the output would necessarily be bad, but because a prospect at that stage would deserves the consultant’s full attention. Our design does keep the consultant in control: they answer the follow-up questions, edit the proposal, approve scope and price and send the email themselves. I think that is the right setup. But for a firm this size, I’m not sure the time saving alone is a strong argument for the tool.
What changed in how I see AI
What made the biggest impression on me was how the tool was made. It was built mostly with AI, in a programming language most of us had never used. Not long ago that would have required an actual developer. Now someone who knows what they want, and understands what the code is supposed to do, can get there. I think that is one of the real strengths of AI: you no longer need to write every line of code yourself, as long as you know what you are doing and how to use what it produces.
That raises a question for a company like DataVance. If building things becomes this easy, why would an SME without a data team still hire a consultancy? Thinking it through, I don’t think it makes them less needed. Building a dashboard is becoming easy, but understanding the data behind it and making the right decisions with it is not. Many companies still need someone who knows what the numbers mean and which ones matter. Our deck made a similar point in strategic terms: when proposals and generic dashboards become abundant, the advantage moves to what stays scarce, like knowledge of the client’s systems and numbers you can trust
