Guides
Automation opportunity assessment
An automation opportunity assessment is how you decide what to automate first, and the useful version is a worksheet you can actually fill in, not an article about one. The whole thing is on this page, free.
Step 1: inventory the processes
List the repetitive work your teams do by hand, one row per process. Keep it concrete: not "finance admin" but "match supplier invoices to purchase orders". For each one, capture five facts. You are describing the process as it actually runs, not as the manual says it runs.
| Process | Volume | Rules or judgment | Systems touched | Cost of a failure |
|---|---|---|---|---|
| Match supplier invoices to purchase orders | ~600 / month | Mostly rules, some judgment on mismatches | Email inbox, ERP, PO system | Overpayment or a strained supplier |
The first row is a worked example we score below. The blank rows are yours. The page prints cleanly from your browser if you want it on paper or as a PDF; we deliberately do not gate it behind an email form.
Step 2: score the value
For each process, answer these six questions, scoring 0, 1, or 2. Add them up for a value score out of 12. A high score means automation is likely to pay; a low score means the process will fight you.
| Question | 0 | 1 | 2 |
|---|---|---|---|
| Volume | A few times a month. | Most days. | Many times a day, high volume. |
| Rules over judgment | Mostly human judgment. | A mix of rules and judgment. | Clear rules a person could write down. |
| Stability | The steps change often. | Changes a few times a year. | Stable for a year or more. |
| Digital inputs | Paper, phone calls, or scattered notes. | Digital but unstructured: PDFs, emails, scans. | Structured data already in a system. |
| System access | Only reachable by a human clicking screens. | Exports and uploads exist. | APIs or a real integration path. |
| A clear success test | You would know a bad result when you saw it, but cannot define it. | Partly measurable. | A rule can tell a right result from a wrong one. |
Step 3: score the risk
Value is only half the picture. A process can be worth automating and still need a person kept firmly in the loop. Score these three the same way, 0 to 2 each, for a risk score out of 6. Risk does not stop you automating; it decides how much guardrail the automation needs.
| Question | 0 | 1 | 2 |
|---|---|---|---|
| Cost of a wrong output | Someone notices and fixes it cheaply. | It costs real money or annoys a customer. | Regulatory, financial, or safety consequences. |
| Reversibility | Easily undone. | Undone with effort. | Irreversible: money moved, message sent, filing made. |
| Regulated data or decision | Neither. | Personal or confidential data. | A regulated decision: credit, claims, hiring, health. |
Step 4: read the verdict
Put the value score and the risk score together and each process lands in one of four places:
- Value 8 to 12, risk 0 to 2: automate now. High return, low blast radius. This is where a build pays back fastest, so it is where to start.
- Value 8 to 12, risk 3 to 6: automate with guardrails. Worth doing, but a person stays on the decisions, every action is logged, and the automation prepares work rather than finishing it unsupervised.
- Value 5 to 7: fix the process first. Something is holding it back, usually messy inputs, unreachable systems, or steps that change too often. Simplify those, then score it again. Automate only the stable, rule-based sub-steps for now.
- Value 0 to 4: leave it to people. Automation will cost more than it saves here. Revisit only if the volume grows or the process settles down.
One override: any process that scores the full 6 on risk (irreversible, a regulated decision, on regulated data) never runs fully hands-off, no matter how high its value. Automate the preparation and keep a human on the call.
The worked example
Take the invoice-matching row from the inventory, at a mid-size distributor. This is a labelled hypothetical, not a client. Volume is high (2), it is mostly rules with judgment only on mismatches (1), the steps are stable (2), inputs are digital but unstructured PDFs (1), the ERP has an API but the inbox does not (1), and success is largely testable against the purchase order (2). Value score: 9. On risk: a wrong match costs money (1), a payment is hard to reverse once sent (1), and there is no regulated decision (0). Risk score: 2.
Value 9, risk 2 sits right on the line between "automate now" and "automate with guardrails". The honest reading is guardrails: automate the reading and matching, flag the mismatches to a person, and never let it release a payment on its own. That is a real, buildable scope, and it is the kind of answer the worksheet is meant to produce.
When not to automate at all
The most valuable output of an honest assessment is often a "no". These are the rules we apply to our own work, and they are why we sometimes tell a client a process is not worth automating:
- Do not automate a broken process. Automation scales whatever you give it, including the mess. Fix or delete the process first, then automate the version that works.
- Do not automate a process nobody can describe. If the steps live only in one person's head, you are not ready. Write them down first; the act of documenting usually reveals the real work.
- Do not automate a process that changes every month. You will spend more maintaining the automation than you ever saved. Wait for it to settle, or automate only the stable parts.
- Do not fully automate an irreversible, regulated decision. Keep a person on the final call for anything a regulator can ask you to explain: a credit denial, a claim rejection, a hiring cut. Automate the preparation, not the decision.
- Do not automate low-volume work just because you can. Ten minutes a week is not worth a build and its ongoing maintenance. The maths only works when the volume is real.
- Do not automate to dodge a staffing conversation. If the real problem is how the team is organised, automation just hides it and makes the eventual reorganisation harder.
What this shares with an AI readiness check
This worksheet works at the level of a single process: is this specific piece of work worth automating. That is a different question from whether your organisation as a whole can carry AI, which is what our AI readiness checklist covers. Use this one to pick the work; use that one to check the foundations under it. They are deliberately separate.
Or have us do it with you
This is genuinely our method, not a lite version of it. The difference in the paid engagement is who does the verifying and how deep it goes: the two of us inside your actual systems for 3–6 weeks, scoring your real processes against evidence rather than your best guess, and handing back a written recommendation with the numbers behind it. It is a fixed scope at $20,000 to $80,000, and if the answer is that little of it is worth automating, that is what we will tell you. The build itself, when there is one, is on the AI automation service page. When you are ready, start with an assessment.
Questions people ask
- How do you decide what to automate first?
- Score every candidate process on two things: how much automating it is worth (volume, how rule-based it is, how stable it is, how clean the inputs are, whether the systems can be reached, and whether success is measurable) and how much could go wrong (the cost of a mistake, whether it can be undone, and whether it touches regulated data). High value with low risk goes first. That is exactly what the worksheet above produces.
- What is the 80/20 rule for automation?
- The idea that roughly 80 percent of the benefit comes from automating about 20 percent of the work, usually the high-volume, rule-based steps, while the last stretch of edge cases costs far more than it returns. In practice it means automate the routine core, keep people on the exceptions, and stop before you chase every rare case. The value score in the worksheet is built to surface that 20 percent.
- What is the 30% rule for AI?
- A rule of thumb that AI should handle roughly 30 percent of a process, the routine part, while people keep the judgment calls. It is a decent guard against over-automating, though the right split varies by process, which is what the scoring here is for.
- What are the four types of automation?
- Usually named as fixed (hard-coded for one repetitive task), programmable (reconfigurable for different jobs), flexible (adapts within set limits), and intelligent (uses AI to handle variation and judgment). The label matters less than the fit: this worksheet tells you which kind, if any, a given process actually needs.
- How much does it cost to automate a process?
- It depends entirely on the process, which is why we assess before we quote. Our fixed-scope assessment runs $20,000 to $80,000; the ranges for the build itself are broken down on the AI automation service page. The worksheet above is the free version of the first step: deciding whether a build is worth pricing at all.
Tell us about the work.
A few lines is enough. We read every enquiry ourselves and reply within one business day.