Everyone is using AI.
Almost nobody
is getting paid for it.
A half-day working session that ends with one agent built on your data, in your stack, with the measurement wired in — so in ninety days you can say what it actually returned.
66%
of US small businesses now use AI — up from 55% a year earlier
The eleven-point jump is the part worth reading. Whatever advantage existed in simply using these tools is closing, and what is left is the advantage of using them well.
95%
of GenAI pilots produced no measurable P&L impact
The finding that matters most here: adoption is not the bottleneck. Turning adoption into a number on the P&L is. Pilots stalled where tools could not hold context, take feedback, or fit an actual workflow.
70%
of small businesses say they need more training to use AI effectively
In the same survey, 57% said they were learning from YouTube and social media. Unstructured learning is the default, and it is the most likely reason so much adoption produces so little.
Read together: the tools have gone mainstream, the returns have not, and most people are teaching themselves off YouTube. The scarce thing is no longer access — it is knowing which process to point an agent at, and proving afterwards that it paid.
What we do in the room
- 01
Where the hours actually go
Before touching a tool, map the week. Which tasks are repetitive, structured, and consequential enough to be worth automating — and which only feel that way. Most teams automate the wrong task first because it is the easiest one to see.
You leave with A ranked list of your own processes by automation value.
- 02
What an agent is, and what it is not
The working distinction between a chatbot, a prompt, a workflow, and an agent — in operational terms rather than marketing ones. Where each one genuinely fits, and the failure mode of each.
You leave with Shared vocabulary, so the team stops talking past each other.
- 03
Build one, live
We take the top-ranked process from module one and build a working agent against it in the room, on your data, in your stack. Not a demo on someone else’s example.
You leave with One agent that works, that you keep.
- 04
The gates that keep it safe
Least-privilege access, secrets kept out of prompts, dry runs before writes, staged environments, and a named human approving anything consequential. This is the module that separates a system you can run from one you will quietly abandon.
You leave with A written governance page for your business.
- 05
Measure it or lose it
How to instrument the thing you just built so that in ninety days you can say what it returned — in hours and in dollars, with the denominator written down. The step almost every pilot skips, and the reason almost every pilot dies.
You leave with A baseline measurement and a 90-day review date.
Built for
- Owners and operators — of businesses roughly 2–200 people, who are paying for tools nobody has been trained to use.
- Teams already experimenting — where three people each found their own workflow and none of it is written down or repeatable.
- Anyone who has run a pilot that died — and has not been able to say precisely why.
Not built for
- Anyone wanting a keynote. This is a working session — you leave with something built, or it did not work.
- Teams looking to cut headcount. That is a legitimate goal, and it is not the one this is designed around.
- Businesses with no repeatable process yet. Automating an undefined process just makes the confusion faster.
Each figure is labelled by how strong the evidence behind it is. Measured means an outcome observed in data. Assessed means researchers judged it from interviews and document review. Reported means people said it happened to them. Estimated means people predicted it. They are not interchangeable, and a workshop that blurs them is selling you something.
- 15% measured
more issues resolved per hour
Brynjolfsson, Li & Raymond, “Generative AI at Work,” Quarterly Journal of Economics 140(2), 2025
Caveat — One firm, one job type. The size of the effect does not automatically transfer to other work.
- 95% assessed
of GenAI pilots produced no measurable P&L impact
Caveat — Enterprise-focused, and a modest base — 52 interviews, 153 survey responses, 300 public deployments. Directionally important, not a precise rate.
- 70% reported
of small businesses say they need more training to use AI effectively
- 66% reported
of US small businesses now use AI — up from 55% a year earlier
Thryv AI and Small Business Adoption Survey, April 2026 — 561 US SMB decision-makers
- 5.6 hrs reported
per week, average time saved per SMB employee
business.com SMB AI workplace study — 1,009 respondents at US businesses of 2–250 employees
Caveat — Self-reported, and the publisher does not disclose fieldwork dates. Treat as a directional figure.
- $500–$2,000 estimated
per month — what SMBs estimate AI saves them
Thryv AI and Small Business Adoption Survey, April 2026 — 561 US SMB decision-makers
Caveat — An estimate of savings, not an audited figure. Nobody reconciled this against a P&L.
- 91% reported
of SMBs using AI say it boosts revenue
Caveat — Correlation, and the causal arrow is genuinely unclear. Do not present this as proof AI caused revenue growth.