There is a particular kind of pressure on small institutions right now. The message everywhere is that you must adopt AI immediately or fall behind. Very little of that advice is written for an organization where a wrong decision costs real money and cannot be quietly absorbed.
A large company can run ten AI pilots, kill nine, and call the tenth a win. A twelve-person clinic, a lab living grant to grant, or a family business with thin margins does not have that luxury. For those organizations the useful question is never “how do we use AI.” It is narrower and more honest: where does a tool actually earn its place in the work, and where would it only add risk you cannot afford.
This is a guide to answering that question deliberately. It is how we advise the small institutions we work with, and it is the opposite of the move-fast-and-break-things posture that dominates the conversation.
Start with the job, not the technology
The first thing we do on any engagement is set the technology aside. We ask what work is slow, error-prone, or draining for your people today. Where does someone re-key the same data into three systems. Where does a decision wait two days because one person has to touch it. Where do good people spend their afternoons on something a computer should have done for them.
Those are the seams where a well-built tool pays for itself. If you cannot name the specific job an AI feature would do, that is not a reason to build a demo and hope. It is a reason to stop. Technology looking for a problem is the most expensive kind of software there is.
Naming the job also gives you something the hype cannot: a way to tell whether the thing worked. “We adopted AI” is not a result. “The intake that used to take our front desk two hours a day now takes twenty minutes, and the errors dropped” is a result. Start there.
Choose assistance over automation
For a small institution, the safe and usually correct pattern is assistance, not automation. Let the tool draft, sort, flag, summarize, and suggest. Then let a person decide.
The difference is not cosmetic. A model that suggests which grant records look incomplete is a gift to a busy administrator. A model that silently deletes the records it thinks are wrong is a liability waiting to happen. Same underlying capability, completely different risk. What separates them is whether a human stays accountable for the outcome, and that is a design choice you make on purpose, at the start, not a setting you discover later.
Keeping a person in the loop is not timidity. It is how you capture the upside of a fast, tireless assistant without handing your operation to a system nobody in the building fully understands. It also keeps your people sharp. A tool that does the thinking for your staff will, over a few years, leave you with staff who can no longer do the thinking. That is a bad trade for any organization that plans to be around in five years.
Own the parts that matter
Most AI advice quietly assumes you will rent everything: the model, the platform, the interface, and the pipes that connect them. For a small institution that assumption is the risk.
You do not need to own the model. Almost no one should train their own. But you should own the account it runs under, the code that calls it, the logic that makes it useful, and above all the data that flows through it. When those live in accounts with your name on them, a price change or a discontinued product is an inconvenience. When they live inside a vendor’s black box, the same event is an emergency. Deliberate adoption means keeping the keys to the parts that are genuinely yours.
Five questions to answer before you adopt anything
Before you commit to any AI tool, whether you build it or buy it, answer these plainly. If you cannot, you are not ready yet, and that is useful information.
- What specific job does this do. If the answer is vague, wait.
- Who stays accountable for every decision the tool influences.
- Where does your data go, who can see it, and how do you get it back.
- Can you turn it off and keep operating without it.
- Do you own the account and the code, so the tool cannot be taken from you.
Notice that only one of these is technical. The rest are questions about accountability, ownership, and continuity. That is not an accident. The failures we see in small institutions are almost never about the model being bad. They are about nobody deciding, up front, who is responsible when it is wrong.
What deliberate adoption actually buys you
Deliberate adoption is slower than the hype wants you to be. It is also how you end up with software that serves your people for years instead of a pilot you quietly abandon after the budget cycle ends.
The organizations that do this well are not the ones that adopted AI first. They are the ones that adopted it on purpose, in one clearly defined place, with a person accountable and a way to measure whether it helped. Then they did it again somewhere else. That is not a slogan. It is just how durable things get built.
If you are weighing where AI belongs in your work, that is exactly the kind of conversation we like to have. Including the honest version, where the answer for now is not yet.
Written by
Miles Bassett
Founder and principal craftsman at Prairie Code. He writes and speaks on deliberate AI adoption for small businesses and institutions.