AI Guides › Playbooks

The Five-Question Asset Test

By Nigel Guy · 6 min read

Most advice about "building an asset with AI" shows you a task with extra steps. You write a prompt, get a result, sell or send the result, and next week you start again from nothing. It feels productive because something got made, and it fails because the thing you made does not work while you are asleep, on holiday or busy with a client.

The rule: if it stops working the day you stop touching it, it is a task. An asset is something you build once that keeps producing, can be handed to someone else, and gets better each time it is used.

A note on sources. The brief for this guide pointed to two worked examples of named people with dates and figures attached. We could not verify those people or numbers independently, so we have not repeated them. What follows is the durable part: a test you can run on anything, including examples you read online. If someone shows you an "asset" with a revenue figure but no way to check it, treat the figure as a claim, not evidence.

The Five-Question Asset Test

Run any candidate through these in order. One "no" in the first three means it is a task, however good the output.

# Question Passes if Fails if
1 Does it run without you? Someone else, or a scheduled process, can use it with no input from you beyond a short brief Every use needs you to prompt, edit and judge
2 Can you hand it over? It lives in a file, template, folder or system a stranger could open and follow It lives in your head or in one long chat
3 Does the second use cost less than the first? Reuse takes minutes because the hard decisions are written down Each use costs the same effort as the first
4 Do you own and control it? You hold the files, the audience list or the account It sits on a platform that can change terms, or on someone else's list
5 Can you see it working? There is a count you can check: uses, sales, enquiries, hours saved, measured before and after Your only evidence is that it feels useful

Questions 4 and 5 do not decide asset versus task, but they decide whether you can trust your own verdict.

How to run it, step by step

  1. Write down the thing in one sentence. "A reusable brief and checklist that lets a new client onboarding take 20 minutes instead of two hours" is a candidate. "A PDF" is not.
  2. Answer questions 1 to 3 honestly, in writing. Do not answer in your head. If you catch yourself writing "well, mostly", that is a no.
  3. Find the part that is still you. Usually it is judgement: deciding what good looks like. Write that judgement down as criteria, examples and exclusions. That written part is the asset. The AI output is the by-product.
  4. Store it where it can be reused. In Claude, the two natural homes are Projects and Skills. Per Anthropic's help centre at the time of checking, a Project is a self-contained workspace with its own chat history, uploaded documents and custom instructions, and free accounts can create up to five while paid plans get unlimited. Skills are instruction folders Claude loads for repeatable workflows, available on Free, Pro, Max, Team and Enterprise plans but requiring code execution to be switched on. Only Team and Enterprise plans let you share a Project with colleagues. Plans change, so check the current help pages before you rely on any of this.
  5. Do the second run and time it. If the second run is not clearly cheaper than the first, go back to step 3.
  6. Set a check date. Put a date in your diary four weeks out to look at your question 5 count. No count, no claim.

A hypothetical worked example

This is invented, to show the mechanics. Imagine a freelance bookkeeper in Leeds who uses AI each month to write plain-English summaries of client accounts.

Note the honest wobble. Most real assets pass some questions partly. The test is useful because it shows you which part to fix next.

A prompt to run the test with Claude

Paste this into a new chat. Fill in the square brackets.

You are a sceptical business advisor helping a UK sole trader or small team judge whether something they built with AI is a genuine reusable asset or just a one-off task.

Context: [DESCRIBE WHAT YOU BUILT AND HOW YOU USE IT IN 3-6 SENTENCES]
Who else could use it: [NAMES OR ROLES, OR "NO ONE YET"]
Where it is stored: [TOOL, FOLDER OR ACCOUNT]
What I have measured so far: [NUMBERS WITH DATES, OR "NOTHING"]

Goal: a clear verdict, not encouragement.

Work through these five questions in order, answering each as Pass, Partly or Fail with one or two sentences of reasoning drawn only from what I have told you:
1. Does it run without me?
2. Can I hand it over to someone else?
3. Does the second use cost less than the first?
4. Do I own and control it?
5. Can I see it working, with evidence?

Then give me:
- A one-line verdict: asset, partial asset or task.
- The single weakest answer and the smallest change that would improve it.
- A short list of what I should stop doing.

Rules: if any input is missing or too vague to judge, ask me for it before answering rather than guessing. Do not invent figures, benchmarks or examples. Where my evidence is only a feeling, say so.

Before replying, check that every Pass is backed by something I actually wrote.

What to skip

Guardrails

Sources

All 751 AI guides · JulieMango plans from £17/mo