Five questions to answer before your first AI project
Maangarad AI · 4 October 2026 · 3 min read
Before you pick a tool or sign with a supplier, answer five plain questions. They are cheaper than a failed pilot.
The first AI project in an organisation sets the tone for every one after it. If it fixes a real problem, people trust the next idea. If it ends up as an expensive demo nobody opens, colleagues will avoid the word AI for years. These five questions are the ones to settle before any money moves.
1. What problem are we solving, and for whom?
Start from the work, not the technology. Describe one task that is slow, repetitive or error-prone, and name the people who do it. “Our M&E team spends two days a month turning field reports into the donor summary” is a starting point. “We want to use AI” is not. If you cannot say who will be doing something differently at their desk when the project is done, it is not ready.
2. How will we know it worked?
Agree what success looks like before you start, and how you will check it. Hours saved on a task. Fewer corrections sent back by the finance office. Faster replies to the public. Measure the current state first, so there is an honest number to compare against. Pick two or three measures you can actually collect, not ten you will abandon.
3. Do we have the information the tool needs?
AI tools run on information: documents, records, forms, the questions people ask. Check whether it exists, where it lives, whether it is current and whether you are allowed to use it. Paper files in a cupboard, five versions of the same spreadsheet on five laptops, and data a donor has not cleared you to share are the usual blockers. Sometimes the most valuable first step is to put the records in order, and no AI is involved at all.
4. What could go wrong, and who gets hurt?
Be honest about the risks. Could the tool give a wrong answer that someone acts on? Could it leak private information? Could it work well in English and badly in Somali, so that one group gets a worse service? For each risk, decide how you will reduce it: a person reviewing the output, limits on what the tool is used for, testing with real users, written instructions. The closer a task sits to people’s lives, the more care it needs.
5. Who owns it after launch?
Most projects die after launch, not before. Someone has to own the tool: answer staff questions, watch for errors, update it when the forms change, and decide whether it is still earning its keep. Budget for training and support beyond the first few months. A tool with no owner is quietly abandoned within a year.
Start small, learn quickly
The best first projects are small enough to finish in a few months and visible enough that colleagues notice the difference. A focused pilot with a clear owner and honest measures teaches you more than a large strategy on paper. When it works, you have evidence, experience and support for the next step.
Maangarad’s AI consulting starts with these five questions, worked through with your team, so the first project begins on solid ground.
