September 2026
Most small-business AI pilots stall in the same three places
None of the three are technical. What we see in offices of thirty to two hundred people, and what the ones that work did differently.
By Ryan Sanders, Partnership Director
Almost every business we support has now tried something with AI. A licence bought in a renewal, a free tool somebody found, a pilot that got a lot of attention in February. A year on, a small number of those have quietly become part of how the work gets done, and most of the rest are a browser tab nobody has opened since spring. The difference is rarely the tool.
One: nobody owned it
A pilot with no name against it belongs to everybody, which means it belongs to nobody. The successful ones we have seen all had a person whose job it was to make it work, usually not a technical person, usually someone senior enough to change how a task is done. That person does not need to understand the model. They need to be allowed to say "from now on we do the first draft this way".
Two: it was pointed at the wrong work
These tools are good at drafting, summarising, restructuring text and finding things in documents you already have. They are not good at numbers that have to be right, and they are confidently wrong in a way that is hard to catch. Pilots aimed at reporting, reconciliation or anything with a total at the bottom tend to fail, and they take the office's enthusiasm with them when they go. Aim the first one at a task where being 90 per cent right and edited by a human is a genuine improvement.
Three: the training never happened
A licence handed out with a link to the vendor's tour is not a rollout. Twenty minutes of real training, on your actual work, with two or three examples from your own business, changes adoption more than anything else in the budget. It is also the cheapest part, and the one most often skipped because it does not feel like progress.
What the ones that worked had in common
A named owner. A small, unglamorous first task. A written rule about what could be pasted in, agreed before anyone started. The enterprise edition rather than the free one, on a company account. And a decision date: six or eight weeks out, someone says whether it stays or goes, and the answer is allowed to be no.
And a fair warning
If your backups have never been tested, or your staff are not on multi-factor authentication, an AI pilot is not the most valuable thing you could do this quarter. We will tell you that in the first meeting. It is not the answer anyone is hoping for, but it is the one that saves businesses.
From people who chose local
Order the internet circuit the week the lease is signed. Carriers take six to twelve weeks. Nothing else on a move fails as often.
The relocation checklist Did you knowA backup that has never been restored is a hope, not a plan. Ask for the date of the last tested restore.
What to actually test
"Well done!"
Kevin, manufacturerMore from the survey wall- A technician who knows your setup takes every request. Not a dispatcher, not a queue.
- Every closed ticket is surveyed and a partner reads every response.
- 99.5% of surveys come back positive.
- Family-owned since 2005. 21 years in the same corner of Wisconsin, not going anywhere.
"Tyler got to my request very quickly and resolved the issue within minutes."
Nichole, insurance brokerageTalk to a partner, not a sales script.
Tell us what is going on. A partner will call you back the same business day, look at what you have, and tell you honestly whether we are the right fit. No pricing pressure, no pitch deck.