Which task should you automate first?

The scorer Rate one task, get a verdict

Finding the one task that pays first — and building it safely — is the practical spine of AI for Your Business: Basics →

How to read it A score is a shortlist, not a green light

The point of the scorer isn't to hand you a decision — it's to stop you automating the loudest task instead of the right one. Owners tend to reach for whatever annoyed them most this week. The score pulls you back to what actually pays: work that happens often, eats time each pass, and runs the same way every time. Those three, and only those three, make up the priority score.

Read the verdict band as a shortlist position, not a verdict on the whole idea. A high score means "this is a good candidate — worth a real build" — never "switch it on and walk away." A low score is just as useful: it tells you to leave a task with a person and spend your first effort elsewhere.

Value and risk are two different questions

Notice what the fourth input — cost of getting it wrong — does, and what it does not do. It does not raise the score. A task that would be expensive to get wrong is not a better automation candidate for being risky — it's a task that needs more care. So the error-cost is a separate human-review gate: when it's high (a 4 or 5), the verdict tells you to keep a person checking every result, no matter how high the value score climbs. Automating a high-stakes task blindly doesn't remove the risk — it amplifies the very mistake you were worried about, at machine speed. Value tells you whether to automate; error-cost tells you how much of a human to keep in the loop.

How the score works

Three value factors are each rated 1 to 5. Frequency and time count once; repetitiveness counts twice. So the formula is score = frequency + time + (2 × repetitiveness), which tops out at 5 + 5 + 10 = 20 and bottoms out at 1 + 1 + 2 = 4. Then the bands:

• 16 or more — Automate this first. Frequent, repetitive, time-hungry: exactly where AI earns its keep. (If a mistake is costly, the verdict becomes "automate — but keep a human reviewing every result.")
• 11–15 — Strong candidate. Clearly worth a build. (If a mistake is costly, keep a human reviewing every result.)
• 7–10 — Worth automating later. Real value, but something better is probably ahead of it.
• Under 7 — Leave it manual for now. Too rare, too varied, or too quick to be your first bet.

Why repetitiveness is weighted double

Because it's the factor that decides whether AI can do the job at all. Today's AI is at its best on work that repeats the same steps in the same shape — the same kind of email, the same intake form, the same lookup — because that's what it can learn and reproduce reliably. The opposite kind of work, the one-off that turns on judgement, context and reading the room, is the worst automation target there is, no matter how often it comes up or how long it takes. A task can be frequent and slow and still be a bad fit if every case is genuinely different. Weighting repetitiveness double keeps the score honest about that: it stops a busy-but-bespoke task from scoring its way to the front of the queue where it doesn't belong.

Fix the process before you automate it

One warning the score can't give you on its own: automating a broken process just makes it break faster. If the task is messy — inconsistent steps, missing information, a hand-off nobody owns — a tool will scale the mess, not clean it up. Straighten the process out by hand first, until a person can run it the same way every time. Then automate the version that already works. (You may notice that tidying the steps also raises the repetitiveness score — that's the same truth showing up twice.)

A worked example

Take answering routine booking calls: it happens many times a day (frequency 5), each call takes a few minutes (time 2), and it follows the same script almost every time (repetitiveness 5). The score is 5 + 2 + (2 × 5) = 17 — and because a fumbled routine call is easily recovered (error-cost 2), the verdict is a clean "automate this first."

Now change one thing. Keep the same frequency, time and repetitiveness — so the same score of 17 — but make it processing customer refunds, where a wrong amount costs money and trust (error-cost 5). The value is identical, so it's still worth automating. But the verdict changes to "automate — but keep a human reviewing every result," because a costly mistake here is never something you leave fully hands-off. Same score, different oversight — that's the whole point of separating value from risk.

Edge cases the scorer handles

• All three value factors at 5 score 5 + 5 + 10 = 20 — the clearest possible "automate this first," unless the cost of a mistake is high, in which case it's "automate, with a human reviewing every result."
• All three at 1 score 1 + 1 + 2 = 4 — under 7, so "leave it manual for now." Rare, varied and quick: there's nothing to gain.
• Slow but bespoke — an hours-long task (time 5) where every case differs (repetitiveness 1) can't climb far, and that's the point: length alone isn't a reason to automate.
• High score, high stakes — a task can score 20 and still not be hands-off. The error-cost gate holds a human on the result; the score never overrides it.

Common questions

Why doesn't "cost of getting it wrong" raise the score?

Because a costly mistake is a reason for caution, not enthusiasm. An earlier version of this tool added error-cost to the score — which meant a high-stakes, expensive-if-wrong task scored higher for automation. That's backwards and genuinely dangerous: it nudges you to hand your riskiest work to a machine first. So error-cost now does something safer — it sets a human-review gate. High stakes don't stop you automating; they stop you doing it hands-off.

Does a high score mean I should just switch it on?

No. A high score means the task is a good candidate — worth the effort of a real build. It still needs a proper build, a test on live-but-safe work, and a human checking the output before anything goes out the door — especially if the cost of a mistake is high. The score gets you to the starting line; it doesn't run the race.

Is the score a guarantee it'll work?

No. It's a shortlisting tool that reflects the numbers you entered. Whether a build actually pays depends on the tools, your data, and how well you set it up — none of which this box controls. Treat the verdict as a sensible place to start looking, not a promise of a result.

What about a task with a low repetitiveness score?

Keep it with a person. Low-repetition, high-judgement work — the calls that need context, discretion or a relationship — is where people are strongest and today's AI is weakest. A low score there isn't a failure to automate; it's the scorer correctly telling you to leave a human in charge.

Can one task score high on time but still not win?

Yes, and that's deliberate. Because repetitiveness counts double, a long task that's different every time will sit below a shorter task that runs the same way each pass. AI rewards sameness, so the score does too.