How to Use AI as a Strategy Controller: Audit Your Task Tracker Against the Plan
Connect an AI model to Jira or another tracker, give it your yearly plan and a team map, and it finds delays, duplicate work and the strategy vs BAU split.
You can use an AI model as a strategy controller. Give it your yearly plan, a plain description of what each team owns, and read access to your task tracker. It will compare what was planned with what is actually happening: which initiatives are late, where two teams are doing the same work, and how much of everyone's time goes to strategy versus keeping the lights on. The analysis is slow and usually needs one manual correction, but it surfaces things no status meeting will.
What problem does this solve?
Most companies have a strategic plan in one place (a slide deck, a spreadsheet) and the actual work in another (Jira, Asana, Linear, ClickUp). Nobody reconciles the two in detail, because reading hundreds of tickets against a 40-row plan is a week of work for a controller you probably don't employ.
An AI model with a connector to your tracker can do that reading. It doesn't get tired at ticket 300, and you can rerun it every month.
What do you need?
- The plan. A yearly plan as a deck plus a spreadsheet broken down by quarter is ideal. Any format the model can read works.
- A team map in plain language. One paragraph per department: what it owns, what its goal is this year, and which project or workspace in the tracker belongs to it.
- A connector to the tracker. ChatGPT and Claude both connect to Jira (Atlassian's Rovo exposes Jira to AI tools), and most trackers now have an MCP server.
- Patience. A full analysis across a year of tickets on a top model can take hours. Start it in the evening.
How do you run the first analysis?
- Upload the plan and the team map into one project or conversation.
- Map teams to tracker keys explicitly. This is the correction almost everyone needs. Without it, the model guesses which project belongs to which team and gets some of them wrong. Write it out: "Expansion is project EXP."
- Ask for specific outputs, not "analyse this":
- every plan item with its matching tickets, status and delay;
- estimated versus actual duration per team;
- tickets that look like duplicates across teams;
- the share of work that maps to the strategic plan versus business as usual.
- Define business as usual yourself. BAU is the recurring work without which the company stops running (payroll, support, maintenance). The model can't know your line between BAU and strategy. Write it down and rerun.
- Ask for a PDF report. It's worth printing and reading with a pen.
What does it typically find?
The useful findings are the ones a dashboard doesn't show:
- Duplicate work. Two teams working on the same thing without knowing it.
- The strategy versus BAU split. It shows how much capacity the plan really gets once the daily work is done.
- Estimate drift. It shows which teams consistently underestimate, and by how much.
Don't treat the report as a verdict. Take each finding to the team lead and ask them to explain it. The model can be wrong about context it doesn't have, and you don't want to confront a team over a misread ticket.
How do you make it a weekly habit for team leads?
The monthly plan review is the heavy layer. The light layer is a Monday-morning digest for every lead with five or more reports: a scheduled task over the tracker connector that summarises the previous week. It covers what was finished, what slipped, and actual versus estimated time. It takes minutes to generate, instead of reading a week of notifications.
Let the first few leads build their own scheduled prompt. Once three of them have a version they like, turn it into a shared template.
What are the limits?
- Garbage in, garbage out. If tickets are vague or nobody closes them, the report will be vague too. Fix tracker hygiene first.
- Verify before you act. Back-check any finding before it reaches a team.
- Cost and time. A multi-hour run on a top model isn't free. Monthly is enough for the full review. The weekly digest is the cheap, frequent layer.
Where should you start?
Write the one-paragraph team map this week. It costs nothing, it's the input the model needs most, and it usually exposes ownership gaps before any AI runs.
If you'd rather have someone set up the connector and the first report with you, that's the kind of thing I help companies with.
