The most common reason AI projects fail is starting too big. Instead of giant projects that drag on for months, inflate everyone's expectations, and end in disappointment, we propose a small, focused sprint: a working prototype in six weeks. The goal is not a perfect product but concrete proof you can test with real data and use to make a decision.
Why six weeks?
Six weeks is long enough to take something seriously but short enough not to lose focus. In that time you target a single clear problem, work with real data, and gather enough information to decide "go or stop." If you are heading the wrong way, your loss is limited to six weeks.
The week-by-week roadmap
Week 1: Pin down the problem
Clarify the single problem you will solve. Not "let's put AI everywhere" but a concrete goal like "automatically classify incoming invoices." Write what success means in measurable terms, for example "cut manual work by 50%." By the end of this week everyone should be able to repeat the same one sentence.
Week 2: Gather and clean the data
A prototype is only as good as the data you give it. This week, collect real examples, strip out sensitive fields, and make the data usable. In most projects the largest share of time goes here; accept that from the start.
Week 3: Build the first working version
Use existing ready-made models to create the simplest working version. Do not try to train a model from scratch. The aim is not beauty but an end-to-end skeleton: data goes in, a result comes out.
Week 4: Test with a real user
Give the prototype to the person who actually does the work. Their feedback is worth more than any technical metric. Note where it makes mistakes and where it saves time.
Week 5: Improve and define the limits
Fix the most critical errors based on feedback. Equally important is clearly marking where the system is unreliable. A good prototype also knows when a human should be consulted.
Week 6: Measure and decide
Return to your original success metric. Did you reach the goal? Combine results, cost, and user reaction into a clear decision: scale up, fix-and-retry, or stop. All three are valid and valuable outcomes.
A failed prototype is also a success; it shows you the wrong path in six weeks instead of after an expensive investment.
Common mistakes
- Scope creep: Adding a second problem derails the sprint. Stay loyal to one goal.
- Perfectionism: The prototype can be ugly; what matters is that it works.
- Leaving the user for last: If you wait until week four for a real user, you are too late.
- Skipping the decision moment: If you do not make a clear decision in week six, the project drifts in uncertainty.
After the sprint
If the decision is "scale up," do not move the prototype to production as is. A prototype is proof, not a production system. Use what you learned to plan a solid version for security, scalability, and maintenance. But now you have proof, and that puts all your investment decisions on much firmer ground.
The six-week sprint is the safest way to turn AI from a slogan into a real tool in your business. Start small, learn fast, decide on evidence.
