How to decide what to automate with AI
Every guide on this question says the same thing in prose: automate the repetitive work, keep a person on the judgement calls. The problem is that prose does not sort your tasks — and the one free tool that tries, a8gent's Workflow Automation Mapper, matches each step against twelve keyword patterns and prints fixed percentages (sending email always 95, research always 60), with a plain average as the headline and only two labels, Auto or Human. It says so itself: its percentages are editorial constants, not measurements, and keyword matching is not process analysis. There is no middle band for "AI drafts, a person decides", no way to say how often a task happens or how long it takes, and no estimate of what automating it is actually worth.
This tool does the sorting differently. When you list your tasks here, your browser:
- Scores each task on four ratings — repetitiveness, judgement needed, error cost and data sensitivity — either the values you type or, if you leave them out, ratings suggested from the wording and clearly marked as suggestions.
- Places it on the map — Automate (70+), AI assists (40–69) or Stays human (under 40) — with every factor's points shown, so the score can be checked by hand.
- Works out what it is worth — weekly minutes from your own frequency and duration, recoverable time at 100% for Automate and 50% for AI assists, then hours and money per year at your hourly rate and working weeks.
- Tells you where to start — the biggest Automate task first, with the AI-assists task to draft next and the human tasks to leave alone.
What makes this mapper different
The pages that top the results for "which tasks to automate with AI" — montekristo, worksbuddy, theqdacademy, dijbi — are guides, not tools: they explain the decision and leave the sorting to you. The only interactive rival, a8gent's mapper, uses fixed keyword percentages, a two-label verdict and a plain average, and recommends specific vendors for each step. This page is the only one we found that scores each task on four visible ratings with the reasons shown, gives a three-band decision with a middle "AI assists" band, plots every task on a visual map against the time it takes, and works out the recoverable time and money from your own numbers — free, fully client-side, with no account and no vendor recommendations, plus a live network-request counter so you can verify the privacy claim yourself. It runs entirely in your browser — there is no server to upload to.
When is this the right tool?
Whenever "we should use AI" has been said in a meeting and nobody knows where to start: before buying any automation tool, when a team is drowning in admin, or when you want evidence for a conversation about what the team should stop doing by hand. List the recurring work, read the map, and take the plain-text report to the conversation — the scores do the arguing for you.
Honest limitations
The score is a heuristic, not a business case
Four ratings sort the work; they cannot judge a task's real value, know your tools, or price a project. Treat the map as a starting order, not a return-on-investment calculation.
Keyword ratings are guesses
Ratings suggested from the task wording are marked "ratings suggested" in the table. They can be wrong — a "review" can be a five-minute check or a judgement call. Type explicit R/J/E/S values to override them.
Savings need your numbers
Time saved is computed only from the frequency, duration, rate and weeks you enter. Without your numbers there is no saving — the tool never invents one, and the 50% figure for AI-assists tasks is an assumption, printed on the page.
A high score is not approval
A routine task that moves money or personal data still needs a person on the exception path. The score sorts the work; it does not approve the automation or remove the need for checks.
It does not recommend vendors
The map tells you where to start, not what to buy. Pick the automation tool that fits the software you already use — the decision comes first, the vendor second.
One list at a time
The map covers the tasks you list. It does not track changes over weeks — paste the updated list to re-map, and keep the text report as the record.