AI Workflow Decision Mapper: which tasks should AI do?

List your recurring tasks and see which to automate, which need AI assistance and which stay human — each one scored on your own numbers and placed on a visual map. Scored entirely on your device.

  • Nothing leaves your device
  • Automate / assist / human map
  • Time + money saved, your numbers
  • Transparent scoring, reasons shown

Map your workflow

Format: Task name | 20x/week | 5 min | R5 J2 E2 S1 — name only works too (ratings are suggested from the wording). R = repetitiveness, J = judgement needed, E = error cost, S = data sensitivity, each 1–5. Defaults: 5×/week, 15 min each.

Blended rate, incl. benefits and overhead.
48 allows for holidays and time off.
No tasks yet. Everything stays on this device. Mapping is instant.

Privacy you can verify, not just trust

Most workflow tools ask you to describe your business to their servers first — and then they keep a copy of how you work. This one can't: the scoring is plain JavaScript running in your browser. Your task list never leaves the device — the only third-party requests are the site's own analytics, which count page views — and it uploads nothing, ever. There is no file fetched after load for the scoring itself. Don't take our word for it:

New cross-origin requests during use: 0
  1. Press F12 to open developer tools, then the Network tab.
  2. Map any tasks above. Watch: apart from the site's analytics, nothing new is requested — no uploads, and no task data ever leaves your device.
  3. Extra credit: disconnect your internet after this page loads — mapping still works.

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:

  1. 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.
  2. 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.
  3. 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.
  4. 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.

Frequently asked questions

Do my tasks upload anywhere?

No. Your task list is scored entirely in your browser with plain JavaScript — your tasks never leave the device, and the only third-party requests are the site's own analytics, which count page views. You can verify it yourself with the network-request counter in the privacy section below the mapper.

How do I list my tasks?

One task per line. Just the name works — for example 'Sort the daily inbox' — and the tool suggests ratings from the wording. For exact control, add the details after a | character: 'Sort the daily inbox | 20x/week | 5 min | R5 J2 E2 S1'. Frequency can be '20x/week', 'daily', 'weekly' or 'monthly'; duration can be '5 min' or '2h'; R is repetitiveness, J is judgement needed, E is error cost and S is data sensitivity, each 1 to 5. Leave anything out and the defaults apply (5 times a week, 15 minutes each).

How is the suitability score calculated?

Every task scores 0–100 from four ratings you can see and change: score = 50 + 12×(repetitiveness−3) − 10×(judgement−3) − 8×(error cost−3) − 6×(sensitivity−3), clamped to 0–100. Repetitive, rule-based work scores high; work that needs judgement, where mistakes are expensive, or that touches sensitive data scores low. The table shows each factor's contribution in points, so you can check the arithmetic for any task.

What do the three bands mean?

70–100 is Automate: routine, rule-based work where a mistake is cheap — hand it to software or AI and check a sample of the output. 40–69 is AI assists: AI drafts or prepares, and a person decides — the usual home for writing, summarizing and first-pass research. Under 40 is Stays human: judgement calls, sensitive conversations and expensive decisions where a person should handle the work and AI can only prepare for it.

How is the time saved worked out?

Only from the numbers you type: frequency × minutes per occurrence gives each task's weekly minutes. Automate tasks count at 100% of that time, AI-assists tasks at 50% (the assumption is printed on the page), and human tasks at 0%. The yearly figure multiplies the weekly saving by the working weeks you set (48 by default) and by your hourly rate. Change any number and the saving changes with it — the tool never invents a saving.

What if I disagree with a rating?

Type explicit values on the line and they win: 'Handle a customer complaint | 4x/week | 25 min | R2 J5 E4 S3' scores from your ratings, not from a guess. Ratings the tool suggested from the task wording are marked 'ratings suggested' in the table — they are guesses from keywords, clearly labelled, and adding your own R/J/E/S values overrides them.

Does it tell me which AI tool to use for each task?

No — deliberately. This tool maps the decision (automate, assist, or keep human) and the reasons behind it; it does not recommend vendors, and it is not paid to. Once a task lands in Automate, pick the tool that fits your existing software — the map tells you where to start, not what to buy.

Can I map a whole team's workflow?

Yes. List every recurring task the team does, one per line, with each person's frequency and duration — up to 100 tasks. The map sorts the whole team's work the same way, and the per-task table shows who should keep what. For a fair comparison, use the same hourly rate for everyone or run the map once per role.

What are the honest limits of the map?

The score is a heuristic from four ratings, not a business case: it cannot judge a task's real value, it does not know your tools, and it does not estimate real savings without your own numbers. Keyword-suggested ratings are guesses from wording and can be wrong. And a high score never means 'automate without checking' — a routine task that touches money or personal data still needs a human on the exception path.