An AI project usually starts with an idea - and many open questions: Do users really need it? Will the technology hold? What does it cost to get there? The reality-check sprint answers these questions before much budget is spent.
In short
- Duration: typically 2-4 weeks.
- Outcome: a clickable prototype or a first measured result - depending on the project - and an honest recommendation: build, rework or drop.
- Your effort: a workshop of around 4 hours, a few hours for questions and feedback - and 3-5 users for interviews or tests, about an hour each.
- Price: a fixed price we agree up front - usually possible to commission without a tender. If we continue, it is credited to the follow-up project.
The workshop: finding the right question
We start together with the people who know the topic - business unit, product owners and IT where needed. From our side, the team your topic needs joins - experts from design, consulting, engineering and operations, depending on the question.
- Sharpen the starting question - which problem are we solving, for whom, and how will we know it is solved?
- Collect assumptions - what has to be true for the idea to hold? For users, data, technology and operations.
- Pick the riskiest assumption - we test that one first.
- Choose the users - who from your organisation or your customers we talk to and test the prototype with.
Design variants - and drop them
After the workshop we design several approaches, not just one. Often it turns out that the first idea is not the best: a chat instead of a form, an assistant in the background instead of a new app, or no AI at all because a simple rule is enough.
What does not hold, we drop deliberately and early. That is not failure but the real value of the sprint. More in Why good AI products start with a rejected idea.
The prototype: experience instead of describe
We make the strongest variant tangible:
- For product ideas a clickable prototype users can try - with real data where possible.
- For engineering topics a first measured result, for example on a real slice of your code or processes - code access only after an NDA and your approval.
We test the prototype with real users - in interviews or short tests. Recordings only with consent, stored in your or our EU environment and deleted when the sprint ends. Because we come from delivery, we build it to show what will really be possible later - not just what looks good in a presentation.
The recommendation: build, rework or drop
At the end you get a recommendation you can use internally:
- What we learned - which assumptions were confirmed and which were not.
- What we recommend - keep building, rework or drop. A well-founded “drop” is a good outcome too: it saves budget for what holds.
- How to continue - scope, next steps and which billing model fits (billing models in detail).
An example: what the result can look like
An invented case, not a client project - it shows how a recommendation at the end of a sprint is structured.
An insurer wants a chatbot through which customers report claims.
- Starting question: How do customers report a claim faster and more completely - so that claims handlers do not have to follow up?
- Riskiest assumption: Customers prefer reporting a claim in a chat rather than in a form.
- Three variants: (A) a chatbot for the whole report, (B) the existing form with an AI assistant that spots missing details and reads photos, (C) a call-back whose conversation the AI summarises for the claims handler.
- Interviews and tests with 5 customers: In the chat many lose track of what is still missing. The real bottleneck is photos and documents - not the form.
- Dropped: A, because customers want more guidance, not less. C, because it would be expensive to run and does not make reports more complete.
- Recommendation: rework. Variant B, first for a single type of claim - with the prototype as the basis for the committee decision.
The original idea “chatbot” would not have been the best solution here. That is exactly what a sprint should show early.
Who it suits
Enterprises that want to validate an idea before it goes through committees - with a prototype and a recommendation instead of just a slide. And mid-sized companies that want to find out, with a manageable effort, whether AI pays off for a specific topic.
The first step is a free 30-minute first conversation. In it we clarify whether a reality-check sprint fits your project - and if not, we say so.
