How to use it
- 01
Choose one task
Name a single, bounded piece of work, such as entering invoices or answering delivery questions. Assessing a whole department gives a vague answer, because the tasks inside it differ too much.
- 02
Answer as things are
Score the task as it runs today, not as it would run after a reorganization. If the rules live in one person's head, say so. The assessment is only useful when it is unflattering.
- 03
Read the result with the answers
The band is a summary. The answers that scored lowest are the real finding, because each of them names something specific to fix or to design around.
- 04
Compare several tasks
Clear the form and assess two or three other tasks. The purpose is to choose the best first use case among them, and the differences between their scores make that choice easy to explain.
What the eight questions measure
The questions come from what tends to decide whether an automation project succeeds. None of them is about the technology. They are about the task, its information and the consequences of an error.
| Part | What it settles | Why it matters |
|---|---|---|
| Volume | How often the task occurs | Rare tasks rarely repay the effort of automating them |
| Pattern | How alike the cases are | Consistency is what lets results be tested and trusted |
| Inputs | The form the work arrives in | Digital inputs can be processed, while phone calls must first be captured |
| Rules | Whether the method is written down | Unwritten judgment cannot be handed to a system |
| Data | Whether the needed information is reachable | Most pilots stall on access to data, not on the model |
| Mistakes | What a wrong result costs | Sets how much human review the design must include |
| Review | Whether a person can check results | A reviewable task can start safely and earn trust |
| Time | How much skilled effort it takes now | Shows whether the gain is worth the project |
A worked example
Supplier invoices at a garden supplies wholesaler
Load the example to see the assessment for an invented wholesaler's invoice entry. It lands in the top band, and the reasons are visible in the individual answers.
- Why it scores wellInvoices arrive many times a day, follow the same steps, start as documents and are matched against purchase orders that already exist in the system.
- Where care is neededThe inputs are PDFs, not structured data, so extraction will sometimes misread a figure. That is why review by the accounts clerk stays in the workflow.
- How success is measuredMinutes per invoice and the number corrected after entry. Both can be counted today, which gives the pilot a baseline to beat.
The same company's handling of supplier disputes would score far lower: few cases, each different, settled by judgment. Assessing both shows plainly which one to automate first.
What this tool does not do
A score from eight questions is a guide for choosing where to start. It should not be mistaken for a feasibility study.
- It does not test your data. Whether real documents can be read accurately is found out in a pilot.
- One answer can outweigh the total. A task that could harm a person without review should wait, whatever its score.
- It says nothing about cost. Running costs depend on volumes and on the models chosen.
- It does not replace legal or regulatory advice where decisions affect people.
What to do with the finished document
Take the highest-scoring task and write down today's figures for it before anything is built. If you would like help designing a pilot, send us the assessment and we will reply with how we would scope one. Send it through the contact page or book a 30-minute call and bring it along.
Common questions
What score do I need before starting a project?
The top band suggests a task worth piloting soon. The middle band usually means one or two conditions should be improved first, often data access or written rules. No score promises success or rules it out. The purpose is to compare tasks and to see, in the low answers, what would have to change.
Why is the cost of a mistake scored so heavily?
Because it decides how the system must be designed. Where errors are noticed and corrected easily, an automation can begin with light review and improve. Where an error harms a person or cannot be undone, every result needs checking by someone qualified, which changes the economics and sometimes removes the case for automating at all.
Does a low score mean AI cannot help with this task?
No. It means full automation is a poor first step. A person can still be assisted, for example with a drafted reply to edit or documents summarized before a decision. That kind of help has lower risk, because the person remains responsible. Our comparison of an AI agent and workflow automation explains the options.
How does this relate to planning a pilot?
The assessment chooses the task, and a pilot then tests it with real cases and agreed measures. Our guides on identifying workflows suitable for AI and planning an AI pilot take the next two steps in detail.