Answer
How should businesses prioritise AI automation opportunities?
By frequency first, then whether the output can be checked, then who would own it. Estimated value discriminates least.
Frequency first, because return scales with occurrences. Then checkability, because an unverifiable output cannot be trusted. Then ownership, because an unowned automation stops. Estimated value comes last and rarely separates the candidates.
The usual approach ranks candidates by expected value, which fails for a specific reason: every candidate looks valuable, because they were put on the list by someone who found the task annoying. The estimates are all optimistic, all uncertain and all similar, so the ranking is effectively arbitrary and the item that gets built is the one with the most enthusiastic advocate.
Frequency is the first sort and it is the most discriminating because it is a fact rather than an estimate. How many times a month does this happen? That number is usually knowable from records and it varies by orders of magnitude across a candidate list, which means it separates items in a way estimated value does not. Return accrues per occurrence, so a daily task with a small saving beats a monthly task with a large one on most realistic timelines.
Checkability is the second. Can you tell, quickly, whether the output was right? Where the answer is yes, the automation can be trusted incrementally and its errors are caught. Where it is no, you are choosing to act on unverified output, which is a much bigger commitment than the build. This sort moves several candidates down the list and occasionally removes them, which is a useful thing for a prioritisation exercise to do.
Ownership is the third and is the one omitted from every framework. Who will notice when this stops, and who will change it when the business changes? An automation without a named owner will run until something moves and then quietly stop, which is worse than never having built it because people are relying on it. If no name can be attached, the candidate is not ready regardless of its position on the other two.
After those three, most lists are short enough that value estimation is unnecessary. Where two candidates remain comparable, the tiebreaker worth using is what the business learns: an automation that produces data it does not currently have — how many enquiries by type, how long each stage takes — earns more than its direct saving, because it improves the next decision.
One structural recommendation: build the one that makes the others easier. In most businesses that is capturing the work as structured records rather than automating a step. Once arrival, category and timing are being recorded, several later automations become straightforward and the business can measure whether any of them worked, which it currently cannot.
Everything on the list looks valuable, which is exactly why value is the wrong thing to sort by.
Siddharth Sharma, Context Theory
Related questions
Should you start with the most painful task?
Only if it is also frequent, and painful tasks are often rare — the annual return, the quarterly reconciliation, the occasional awkward customer. The pain is real and the return is not, because the build cost is the same and the occurrences are few. Frequency is the less satisfying criterion and it is the one that pays.
How many candidates should be worked on at once?
One, until it has run for a month. The first automation teaches what the recurring cost actually is in your business, and that figure should govern the next decision. Starting three in parallel means learning it three times simultaneously and having three unfinished things when it turns out to be higher than expected.
METHOD
Every figure below carries its source and the date it was verified. Nothing on this page is asserted.
The numbers on this page.
| What | Value | Specific to |
|---|---|---|
| Sub-15-minute compliance — automated routing vs manual only | 62.5% vs 39.1% | Category-wide |
| Odds of qualifying a lead — replying within the first hour vs after it | 7× | Category-wide |
2026 speed-to-lead benchmark · verified
Oldroyd, McElheran & Elkington, "The Short Life of Online Sales Leads", Harvard Business Review (March 2011) · 1.25M inbound leads across 2,241 US firms · verified
What is specific to this page.
| Kind | Claim | Check it against |
|---|---|---|
| Buying behaviour | Ranking by estimated value fails because every candidate on the list was proposed by someone who found the task annoying, making the estimates uniformly optimistic and similar, so the ranking becomes a function of advocacy. | Comparing the estimated values on an existing candidate list for spread. |
| Workflow | Frequency is knowable from records rather than estimated and varies by orders of magnitude across a candidate list, which makes it the most discriminating sort available. | Counting monthly occurrences for each candidate from existing business records. |
| Constraint | An automation without a named owner runs until something upstream moves and then stops quietly, which is worse than not building it because people have come to rely on it. | Naming the person who would notice each existing automation stopping. |
| Response | Capturing work as structured records is the change that makes subsequent automations straightforward and makes their effect measurable, which is why it outranks automating a step in most businesses. | Checking whether the business can currently report enquiry counts by type and stage durations. |
Each row would be wrong on another industry's page. Where a sourced figure exists it is in the table above instead; these are the constraints that shape the work and do not happen to be numbers.
Start with the measurement.
Reading about a benchmark is not the same as knowing your own number. The audit produces yours, measured rather than estimated.
$497 · delivered in 5 business days · credited against month one