Consultants, project managers and analysts live on three deliverables: understand fast, organize well and communicate clearly. Language models speed up all three, but only when the request is well made. This guide brings together the techniques that make the biggest difference day to day, plus ready-to-copy templates.

The structure of a good prompt

Weak prompts ask “write a summary”. Strong prompts give the model the same context you would give a new colleague on the project. Use six blocks:

  1. Role: who the model should be (“senior BI analyst”).
  2. Context: company, audience, goal and constraints.
  3. Task: what to do, with a clear verb.
  4. Material: the data or text, separated by delimiters.
  5. Format: table, bullet points, JSON, word count, language.
  6. Criteria: what makes the answer good and what to avoid.
Tip: wrap the material in tags such as <minutes>...</minutes> or <data>...</data>. It stops the model from confusing instructions with content and reduces made-up answers.

Advanced techniques that work

1. Examples (few-shot)

Showing two or three examples of the expected output is worth more than a paragraph of explanation. It works very well to classify requests, standardize text and turn requirements into user stories.

2. Step-by-step reasoning

For multi-step analysis, ask the model to list hypotheses, evaluate each one and only then conclude. The latest reasoning models already do this internally; with them, ask only for the conclusion and a short justification.

3. Structured output

When the answer will feed a spreadsheet, Power BI or an automation, ask for JSON or a table with fixed columns. Name each field and say what to do when information is missing (for example, null).

4. Prompt chaining

Instead of one giant prompt, split it: extract the facts first, then analyze, then write. Each step is easier to review and fix.

5. Self-critique

After the first version, ask: “review the answer above as a skeptical director: point out three weaknesses and rewrite it”. The second version is almost always better.

6. A test set

If the prompt will be used by a team or inside an agent, keep 10 to 20 real cases with the expected answer and run them every time you change the text. That is the difference between a prompt that “looked good” and a reliable one.

Ready-to-use templates for consultants

Copy, replace what is in brackets and paste the material inside the tags.

Process diagnosis

You are a senior process and automation consultant.
Context: [company, industry, size]. Goal: reduce time and errors in the process below.
Task: analyze the description and deliver:
1) the current flow in numbered steps;
2) the 3 biggest bottlenecks, with evidence for each;
3) what can be automated with simple rules and what needs AI;
4) an estimate of hours saved per week, with explicit assumptions.
If information is missing, list the questions I should ask the client.
<process>[paste here]</process>

Meeting minutes and action plan

You are an experienced project manager.
From the transcript below, produce:
- a 5-bullet summary;
- decisions made;
- an action table with columns: action | owner | due date | status;
- risks and open points.
Do not invent owners or dates: if they are not in the transcript, write "TBD".
<transcript>[paste here]</transcript>

Requirements to user stories

You are a Product Owner. Turn the requirements below into user stories
in the format "As a [persona], I want [action], so that [benefit]", each with
3 acceptance criteria in Gherkin (Given/When/Then).
Group by epic and flag any ambiguous requirement with [?].
<requirements>[paste here]</requirements>

Explain a DAX measure to the client

You are a Power BI expert who explains things to non-technical managers.
Explain the DAX measure below in up to 120 words: what it calculates,
which filters change it and a simple numeric example.
Then, in a "For the technical team" section, point out performance risks.
<dax>[paste here]</dax>

Proposal draft

You are a consultant who writes concise proposals.
Based on the diagnosis below, write a proposal with:
problem (3 lines), phased solution, deliverables, timeline per phase,
assumptions and what is out of scope. Tone: direct, no jargon.
Do not include prices; leave the [investment] field for me to fill in.
<diagnosis>[paste here]</diagnosis>

Project risk analysis

Act as a PMO. List the top 8 risks of the project below in a table:
risk | cause | probability (1-5) | impact (1-5) | response | warning trigger.
Sort by probability × impact.
<project>[paste here]</project>

Handling client data with care

  • Use business accounts or APIs where data is not used to train models.
  • Remove names, IDs and sensitive figures when the task does not need them.
  • Always check numbers, dates and names before sending anything to a client: AI speeds up the draft, the responsibility is still yours.
Next step: when the team uses a prompt every week, it is worth turning it into an agent or an automation, with data arriving on its own and the answer going straight to the right system.
AI consulting All articles