When an AI project goes sideways in a small company, the technology is rarely the reason. Usually the system works. The team just stops using it, bit by bit, and a few months later nobody can say why.
In a company with 30 or 50 people, everyone knows everyone. If the office manager decides the new system is a threat, it’s dead, whatever the owner thinks. So the people side comes first. It decides whether the build was worth it.
Say what’s changing, early
People hear “AI” and assume the worst. If you don’t tell them what you’re planning, they’ll fill the gap themselves, and usually with the worst version.
Say which work is changing, why, and what it means for their job. If the plan is not to replace someone when they leave, say that too, and say it early. People deal much better with a clear answer than with rumours in the break room.
Start with the work nobody wants
Pick the first workflow from the work your team likes least. Chasing invoices, typing the same data into two systems, digging for the right version of a document. Nobody defends that work. When it disappears, people notice the relief before they notice the change.
Give the sceptic a job
Every office has one person who is most wary of AI. Often it’s someone who has been there for years. They know which customer gets a call instead of a letter, and which number in a quote always looks wrong.
Make that person the one who checks the drafts. They’ll catch a bad draft faster than anyone else, and it’s their judgment on the line, which is exactly what you want.
Let their corrections count
When someone changes a draft before it goes out, keep both versions. If the same correction keeps coming up, it becomes a rule the system follows.
Every few weeks, show the team what changed because of them. That way the system becomes something they shape. People are much harder on a tool that was put in front of them than on one they helped build.
Decide whose name goes on it
Emails and letters the system drafts still go out in someone’s name. Agree early who sends what, which signature goes on it, and which customers always get a person.
Some customers want to hear a voice they know. That’s worth keeping, and your team knows better than anyone who those customers are.
Don’t change everything at once
A team can take maybe two changes a month. More than that, and people stop learning the new way and start working around it. Run one workflow until it feels normal, then start the next one.
Let the team say when a step can run alone
In the first weeks, a person reads every draft. Later, only the ones the system is unsure about, plus a sample. Eventually only the exceptions.
The person who does the work should be the one who says when a step is ready for more freedom. Not the owner, and not us. And if they want to move it back, one switch does it.
Decide what happens with the time
This is the part owners skip most often. If a system takes ten hours a week off someone’s desk, what happens with those ten hours?
If nobody decides, the person either feels watched or fills the time with other routine work, and nothing changes for the business. Say what you’d like them to spend it on. The customers who need more attention, the quotes that never got a follow-up, the projects that have been waiting for months.
And look at what’s left of the job. If every routine task goes, what remains can be only the hardest calls, all day. For some people that’s a promotion. For others it’s exhausting. Talk about it with them before it happens.
What it takes
None of this needs a big change program. It needs an owner who says clearly what’s happening, a first workflow nobody will miss, and a team that gets to shape the system they work with.
