AI doesn't replace people on the team, but teams that use it well outpace those that ignore it. Here's what works in small companies.
Start with pain, not the model
The worst rollout is "we have ChatGPT, use it". What works: pick two or three concrete pains, meeting notes, client replies, LinkedIn drafts, and wire one workflow to each. Without pain, no adoption; without adoption, AI is just another line in the budget.
1. Start with notes and summaries
Meeting recordings → notes → action items in Asana. Saves 3–5 hours a week.
2. A shared prompt library
Without one, everyone asks differently and results are inconsistent. Build 15–20 prompts for common situations: client brief, proposal draft, 1:1 notes, interview prep, competitor research. Each prompt has a name, context, input, output and example. Keep the library in Notion or Coda, not Slack, Slack loses things, libraries must be searchable. Give prompts an owner and review them quarterly.
3. Rules for what AI must not do
Boring and critical. Three minimum rules: no client PII into public models; no financial documents without an enterprise contract; human review before anything goes out. Put them on one page and have everyone sign, even by email. Without an explicit rule, something will leak eventually, and under GDPR and B2B contracts that's a problem that doesn't unbreak.
4. Weave AI into the weekly rhythm
If you run an operating rhythm, add a 5-minute monthly-close item: "what AI saved us and where it failed us". Without it, problems stay unnamed, hallucinated dates in notes, mistranslated offers, made-up research. With it, the prompt library improves on its own.
5. Tools, what to actually use
For small teams, two or three tools are enough. In 2026: one general model for writing and analysis (ChatGPT Team or Claude Pro), one document tool (NotebookLM or Claude Projects), one meeting tool (Granola / Fireflies). More tools means more contexts, and context disappears faster than anything else.
6. Measure impact, not activity
Hours saved, tasks delivered on time, team satisfaction. Without numbers, AI is a toy. Track time from meeting to published notes (hours to minutes), share of drafts shipped after under 10 minutes of review, prompts used from the library per week. The first two move fast; the third is the early signal of whether AI is sticking.
7. Onboarding new people
Week one: prompt library, rules, two end-to-end "done with AI" workflows, a 30-minute sit-down with someone who already uses the tool. Without onboarding, new hires won't use AI, they don't know how or aren't sure about the rules. The team splits into "pre-AI" and "post-AI" and culture fractures.
8. Common mistakes
Shipping AI output without review (a hallucinated client email is expensive). Ten tools instead of three (context dies). Hiding AI from clients (transparency is now an advantage, not a risk). Counting prompts instead of impact. Trusting free tiers with customer data (free tier = training data).
9. A 90-day plan
Month 1: prompt library, rules, meeting notes. Month 2: integration into the weekly rhythm, measurement. Month 3: advanced workflows (custom GPTs, research agents, handoff automation). One new workflow a week is realistic. Faster breaks the team; slower changes nothing.
Want to roll this out? See our stepface PRO service, the AI Prompt Kit (80+ tested prompts), or just email hello@stepface.com.