Most development organizations feel the same pressure right now: we have to do something with artificial intelligence. A donor is asking, another organization has launched a pilot, and someone on the team tried a tool that seems to solve everything.
The pressure is understandable, but it makes for poor decisions. A rushed purchase often ends as a license few people use, or a pilot that never leaves the slide deck. These five questions lead to a better decision, and none of them requires technical knowledge.
1. Which decision or process do we want to improve?
"Using AI" isn't a goal. "Cutting the time the team spends consolidating quarterly reports" is. If you can't name the process, the people involved and what goes wrong today, you don't have a case for investing yet.
Mapping the process often shows that the bottleneck isn't technological at all: an approval nobody is clear about, a different format in every office, or a figure that gets requested twice.
2. What data does it need, and who is responsible for it?
AI works with what you give it. Before investing, check what information exists, where it lives, how current it is and who can correct it. If your organization's knowledge sits in scattered files and in the heads of three people, that's the first job.
Ask who that data belongs to as well. In projects with communities, farmers or Indigenous peoples, having access to information doesn't mean you may use it however you like.
3. What happens when it gets something wrong?
Errors don't all cost the same. A mistake in a draft report gets caught in review. A wrong recommendation to a farmer can cost them a harvest.
Rank each use by the harm it could cause, and decide how much human oversight it needs. The higher the risk, the closer a person should be to the final decision.
4. Who will use it, and who will keep it running?
A solution the team doesn't adopt solves nothing. Talk to the people who will use it, and make sure it fits their daily work and the channels and devices they already have.
Think about the day after the project ends, too. Who pays the running costs? Who keeps the knowledge up to date? Who checks quality? If the answer depends on funding that ends in twelve months, build that into the design now.
5. How will we know it worked?
Agree on indicators before building: hours of work freed, cycle time per process, cost per process or adoption by team. Measure where they stand today, because without a baseline you can't prove any improvement.
Ask vendors for evidence as well as a demo. They should be able to show what the system did, with which data and to what quality. We explain why in Why an AI agent needs an auditor.
When not to invest
An honest answer to these questions may be "not yet". If the process is poorly defined, the data isn't ready or nobody can sustain the solution, fix that first. That's a good result as well: it saves you from paying for a tool that wasn't going to work.
An AI readiness assessment exists to answer these questions with evidence before you commit a budget.