Key takeaways
- Plan in quarters, with a test at the end of each before moving on.
- Quarter one is governed access for everyone. Do not skip it to get to the exciting parts.
- Quarters two and three are about workflows, one team at a time.
- Agents come in quarter four at the earliest, once you know what good looks like.
Most AI roadmaps written for smaller businesses are cut-down enterprise roadmaps: too many workstreams, too long a horizon, and no clear point at which you know you are finished with a phase. This one is built the other way round. Four quarters, one owner, and a test at the end of each quarter that tells you whether to move on.
Before you start: three decisions
- The owner. One person, named, with protected time.
- The provider. Usually decided by your productivity suite. See ChatGPT vs Claude vs Copilot vs Gemini.
- The budget ceiling. Seats for the year plus a capped usage line. See AI transformation cost.
Quarter 1: governed access
Goal: every person who will benefit has a company-controlled AI tool and knows how to use it for their job.
- Buy business-tier seats and configure admin, SSO, retention and spend caps.
- Publish a one-page usage policy.
- Run training built around each team's real tasks.
- Move people off personal accounts. See moving off personal ChatGPT accounts.
- Time five named tasks before and after.
Exit test: at least half of licensed staff use it weekly, and leadership can name time savings on specific tasks.
Quarter 2: first three workflows
Goal: three frequent, document-heavy workflows run the new way by default.
- Pick workflows that happen weekly or more, involve text or documents, and are already checked by a person.
- For each, write the old steps, the new steps, and who checks the output.
- Build reusable instructions or projects in the assistant so the new way is the easy way. See building a prompt library.
- Measure time per instance before and after.
Exit test: each workflow has a written new method, a named owner and a measured saving.
Quarter 3: spread and deepen
Goal: the pattern from quarter two is repeated across more teams, and the heaviest users get more capacity.
- Add three to five more workflows, ideally proposed by staff.
- Move heavy users to higher-usage seats where your provider offers them.
- Review seats and reclaim unused ones.
- Check connectors and integrations. Are the right ones on, and only those?
- Refresh training with what has been learnt.
Exit test: six to eight workflows changed, usage stable or rising, and spend inside the cap.
Quarter 4: first agents, carefully
Goal: one or two well-defined processes run with an agent doing the routine steps and a person approving the output.
- Choose a process you have already redesigned manually, so you know what good looks like.
- Give the agent the narrowest possible access. See AI agent permissions and connectors.
- Set a usage cap specific to the agent. See AI agent spending limits.
- Keep a human approval step until the error rate is known and acceptable.
- Write an agent policy. See the AI agent policy template.
Exit test: the agent saves measurable time, has not caused an incident, and someone owns it.
The one-page version
| Quarter | Focus | Measure | Exit test |
|---|---|---|---|
| 1 | Governed access | Weekly active users | Half of licensed staff weekly |
| 2 | Three workflows | Time per instance | Three workflows changed and measured |
| 3 | Spread and deepen | Workflows changed, spend vs cap | Six to eight workflows, spend in cap |
| 4 | First agents | Time saved, error rate | One agent owned and safe |
If you are not hitting a quarter's exit test, do not start the next quarter. Fix what is stuck first. The most common reasons are covered in why AI pilots stall.
Frequently asked questions
What should an AI roadmap include?
For a smaller business: the provider and plan, the governance settings, the workflows to change in priority order, an owner for each, the measures you will use, and a review date each quarter. One or two pages is enough.
How far ahead should an AI roadmap plan?
Plan the next quarter in detail and the following three at outline level. The tools change too quickly for a detailed multi-year plan to survive contact with reality.
Who should own AI transformation in a smaller business?
One named person with a few hours a week to a day a week, depending on size, who reports to the leadership team. It is more important that they are respected and organised than that they are technical.
When should we start building AI agents?
After staff are confidently using a governed assistant and you have redesigned at least a few workflows manually. Agents automate a process. You need to know what a good process looks like first.
Get a straight answer for your business
The Clarity Package gives you a written business case and a detailed implementation plan for 500 USD. The Implementation Package does the whole build, live in 30 days, for a fixed 5,000 USD.