Choosing the right level: use what you have before you build

What it is

When people see a task AI could help with, they often jump to "we need to build something". Usually you do not. There are four levels of effort, and many needs are solved at the first or second.

Level 1: use a feature you already have. Your approved assistant, or the AI built into your email, documents or meeting tools, may already do the job.

Level 2: a reusable prompt or template. A well-written set of instructions, saved and shared, so anyone can get a good result in seconds.

Level 3: a small tool or connected assistant. A power user (someone confident with the tools, not a programmer) sets up a custom assistant, a saved command or a connection to a shared folder.

Level 4: a system that needs engineers. Software that connects to company systems, runs for many people, handles sensitive data and must be supported and secured.

Think of a leaking tap. First you check whether tightening it works. Then you try a new washer. Then you call a handyman. Only if the pipes are failing do you call in a plumbing firm. Each step costs more and takes longer.

Why it matters

• Lower levels are faster, cheaper and easier to change.
• You learn what the task really needs before spending money on building.
• Higher levels bring more risk: more data access, more upkeep, more to go wrong.
• Starting low means value arrives this week, not next quarter.

How to do it

1. Describe the task in one sentence: what goes in, what comes out, and who uses it.
2. Ask four questions: how often is it done, by how many people, what is the risk if it goes wrong, and what data does it touch.
3. Check Level 1 first: try the task with tools you already have, using made-up data.
4. Write a clear prompt or template if results are uneven, and test it on five real-looking examples (Level 2).
5. Move to Level 3 only if the template is used often by several people and needs files or tools to be connected.
6. Move to Level 4 only when the task involves many people, sensitive or regulated data, company systems, or a high cost of error.
7. Write down which level you chose and why, so others can follow your reasoning.

Example

A manager wants help turning meeting notes into action lists. The first idea is to "build an AI meeting tool".

Working through the levels: the team's approved meeting software already offers a summary (Level 1), but it misses owners and due dates. So the manager writes a template:

From these meeting notes, list every action.
For each: what, who owns it, due date (write "not stated" if missing).
Then list open decisions. Do not add anything not in the notes.
Notes: [paste notes]

Tested on five old meetings with names changed, it works well. It is shared with the team and saved as a reusable instruction. Nothing was built. If, later, the whole department wants actions sent into the task tracker, that becomes a Level 3 or 4 question.

Common mistakes

• Starting at Level 4. Building first wastes money on a problem nobody has tested. Try the lower levels first.
• Skipping the risk question. A task that touches customer or personal data may need a higher level, or may not be suitable at all. Check your organisation's rules.
• Judging on one try. One poor answer does not mean a level has failed. Test on several examples.
• Staying low when the need is real. If many people repeat the same manual steps daily, a small tool can be worth it.

Try it this week

1. Pick one task you wish AI could help with.
2. Answer the four questions: how often, how many people, what risk, what data.
3. Spend ten minutes trying it with a tool you already have, using made-up details.
4. Write down whether Level 1 solved it, and if not, what was missing.

Going further

• Many assistants, including Claude, ChatGPT, Microsoft Copilot and Gemini, offer saved instructions and projects (Level 2) and custom assistants (Level 3).
• Keep a short list of tasks and the level each sits at; it shows where a small build would help most.
• For Level 3, agree with IT which connections are approved before connecting anything.
• For Level 4, bring the tested Level 2 template to the engineers; it is a working specification.

Key words

• Level: how much effort and building a solution needs, from using what you have to an engineered system.
• Template: a saved prompt with blanks to fill in, reused for the same kind of task.
• Power user: someone confident setting up tools and assistants, without being a programmer.
• Connected assistant: an assistant given access to approved files or tools.
• Specification: a clear description of what a tool must do.

Examples and figures in this guide are illustrative. AI tools change often: check what your organisation's approved tools offer today. Use approved tools only, and a person checks and decides.

Prefer to print or share it? Download this guide as a PDF (241 KB), or the complete AI Field Guide (1.7 MB).