AutomateWisely Editorial Team · Last reviewed October 31, 2026 · How we research
Most companies have already tried AI. Very few have changed how they work because of it. The gap is rarely the technology: it is choosing the wrong first project, skipping the baseline and never deciding who owns the result. This guide gives you a 90 day plan built around those three problems, with a scorer to pick your first project.
The pattern is clear: using AI is easy, getting measurable value from it is not. The plan below is designed to get one project across that line in about a quarter, then repeat.
The 90 day plan
Each phase ends with a gate: a decision you make on evidence before spending more. Small companies can move faster; the order matters more than the exact weeks.
Start from a problem, then pick one project
Ask each team which tasks are repetitive, text heavy or involve searching for information. Write each as a business problem with a number, such as “support takes 6 hours to send a first reply”. Then score your three best candidates with the tool below and pick one.
Gate: one project, one owner with authority to change the process, one metric.Compare three possible first projects
Educational tool. Rename the projects and move the sliders. Nothing you enter is stored or sent anywhere.
Notice how the exciting project in the example loses. High value does not compensate for weak data and high risk in a first pilot; it belongs in a later quarter, once you have proved the basics.
Measure the baseline and check the data
- Record how the task performs today for two weeks: time per item, volume, errors and cost. Without this, you cannot prove anything later.
- List the systems and data the project needs, and confirm you can access them legally and technically.
- Decide buy or build. For a first project, an existing tool almost always wins.
- Write a one page AI use rule: what data may go into which tools, and what always needs human review.
Run a small pilot with real work
Give the tool to a small group who do the task every day, and keep a person reviewing every AI output at first. Meet weekly for 20 minutes: what worked, what failed, what to change in the instructions or the process. Track the same metrics as the baseline.
Gate: the metric moved, and the people using it want to keep it.Compare, decide and plan the next one
Compare pilot results with the baseline, including all costs: licences, setup time, review time and training. Then make one of three decisions: scale it to more people, fix and extend the pilot, or stop. Stopping a weak project is a good outcome; it frees budget for the next candidate on your list. Our guide on how to measure the ROI of AI shows the calculation.
Gate: a written decision with numbers, and the next project chosen.What to budget for
Licence prices change often and vary by plan, so check vendors’ current pages. What catches companies out is everything around the licence:
Per seat subscriptions or pay per use. Count only the people who will really use it.
Connecting the tool to your systems, plus automation platforms if needed.
Checking outputs is real work, especially in the first weeks.
Short, task specific training beats generic AI courses.
Cleaning documents and records so the AI has something reliable to use.
Policies, vendor checks and, in some cases, legal advice.
If your project involves automating a workflow, our hands-on test of n8n, Make and Zapier shows how differently those tools count usage. If it involves an agent that takes actions, read what AI agents are before you give it access to anything.
The mistakes that stall most projects
Starting with a tool instead of a problem
Buying licences and “seeing what people do” rarely changes a process. Start with a specific, measured task.
No baseline
If you did not measure the old process, you cannot show the improvement, and the project loses support at budget time.
No owner
Someone must have the authority to change how the work is done, not just to run the tool.
Picking the most exciting project first
Autonomous agents and customer facing AI carry more risk. Prove value on lower risk tasks, then move up.
Ignoring the rules
Personal data, confidential documents and, in the EU, the AI Act create obligations. See our EU AI Act checklist.
Questions we get about AI rollouts
How long does it take to implement AI in a small business?
A first project with an off the shelf tool can go from idea to pilot results in about a quarter, as in the plan above. Wider rollout across the business takes longer and happens project by project.
Do we need a data scientist?
Not for most first projects, which use existing tools. You need someone who understands the process and someone who can connect systems; a specialist helps once you build custom solutions.
Which department should start?
The one with a repetitive, measurable task and a manager willing to own the change. Customer support, finance operations and sales administration are common starting points.
What if the pilot fails?
Record why, keep the baseline and move to the next project on your list. A pilot that ends with a clear “no” in 90 days is cheaper than one that drifts for a year.
- Ask three teams for their most repetitive, measurable task.
- Score the best three in the tool above.
- Name one owner and start recording the baseline on Monday.
For more ideas, see AI for small business and AI agent use cases by department.
Sources
- McKinsey, The State of AI (November 2025)
- Silicon Canals, Summary of McKinsey’s 2025 global AI survey
- Gartner, Prediction on agentic AI projects (June 2025)
This guide is general information, not legal, financial or professional advice. AutomateWisely is independent and is not paid by any AI vendor.