How do I find AI use cases in my company?

Map a process that runs at least weekly, break it into steps, and score each step on frequency, clarity of a good output, availability of input, risk and hours consumed. The highest scoring low-risk steps are your use cases. Starting from tools instead of steps is the most common reason AI programmes stall.

Which marketing tasks are best suited to AI?

Brief structuring, adapting assets across formats and markets, transcreation, first-draft copy from an approved brief, research and competitor summaries, performance reporting, and tagging and routing. They repeat often, have a recognisable good result and carry low risk when reviewed.

What should never be handed to AI?

Positioning and prioritisation, final creative and commercial sign-off, regulated or legally sensitive claims, negotiation, and anything involving personal or confidential data outside sanctioned tools. AI can prepare the material for these; a person decides.

How many AI pilots should we run at once?

Two or three at most. Every pilot draws on the same limited attention, and a small number of measured pilots produces better evidence than a broad programme nobody has time to evaluate.

How do we measure whether an AI opportunity paid off?

Record baseline hours, elapsed time, review rounds and error rate before the pilot, then compare the same measures after four to six weeks. Without a baseline you cannot tell improvement from enthusiasm.

Guide · 6 min read

How to Identify AI Opportunities in Your Organisation

By Fredrika Frenkiel, Head of Studio & AI Creative Operations at Lunar, founder of Code of Alfred. About my work

Short answer

You identify AI opportunities by looking at workflow steps, not at tools or job titles. An AI readiness assessment that starts from the org chart finds nothing useful. Break a real process into its individual steps, then score each one on frequency, clarity of a good output, availability of input data, risk if it goes wrong, and hours consumed. Rank the results, and start with the two or three steps that are high frequency, low risk and expensive in time. Anything that fails the risk or clarity test belongs to a person, with AI preparing the material.

1. Start from work, not from tools

The most common way an AI adoption strategy fails is to begin with a tool demonstration and then hunt for somewhere to use it. Start instead from a process the organisation runs every week. List its steps in the order they happen and note, honestly, how long each one takes and how often it is redone. You cannot evaluate an AI opportunity you have not described as a step with an input and an output.

2. The five-question score

For each step ask: Frequency, does it happen weekly or more? Clarity, would a competent colleague recognise a good result immediately? Input, does the material already exist as text, data or an asset? Risk, what happens if the output is wrong, and would review catch it? Cost, how many hours per month does it consume? Score each from one to five. Steps with high frequency, high clarity, available input, low risk and high cost are your shortlist. This takes an afternoon, not a consulting engagement.

3. What usually scores high in marketing

In marketing operations, the same steps score high almost everywhere. Structuring and completing incoming briefs. Adapting an approved asset across formats, channels and markets. Transcreation and localisation with a human language check. First-draft copy from an approved brief and tone guide. Summarising research, category and competitor material. Turning performance data into a written status. Tagging, naming and routing assets. None of these replace judgment, and together they often account for a large share of elapsed production time.

4. What should stay human

Positioning and prioritisation, final creative judgment, regulated or legally sensitive claims, anything involving personal or confidential client data outside sanctioned tools, negotiation and stakeholder management, and any final sign-off with commercial consequence. Being explicit about this is not caution for its own sake; it is what makes the rest of the programme credible to legal, to clients and to the people doing the work.

5. Rank by value, not by enthusiasm

Rank AI use cases in marketing by value: multiply hours saved per month by how confident you are that the step can be automated well, then subtract the effort to build and the risk to manage. The result is an ordered list rather than a wish list. Two or three genuinely good candidates beat twenty pilots, because every pilot consumes attention from the same small group of people who are already busy.

6. Validate with a measured pilot

Marketing workflow automation earns its place through measurement. Take the top candidate, record the baseline (hours, review rounds, elapsed time, error rate), run it for four to six weeks with a named owner, and compare. Keep the manual path available so nobody is blocked. Then decide with numbers: scale it, adjust the workflow around it, or stop. A stopped pilot with a clear reason is a good outcome, not a failure.

How to do it, step by step

01
Pick a process that runs weekly
Choose a real, recurring process rather than an aspirational one.
02
Break it into steps with inputs and outputs
Write each step as a discrete action with what goes in and what comes out.
03
Score frequency, clarity, input, risk and cost
Rate each step from one to five on all five dimensions.
04
Shortlist high-frequency, low-risk, high-cost steps
Keep only the steps that are repeated, safe to review and expensive in hours.
05
Mark what stays human and why
Name the steps that require judgment, sensitive data or legal sign-off.
06
Rank by hours saved minus effort and risk
Turn the shortlist into an ordered list you can defend to leadership.
07
Pilot the top candidate with baseline numbers
Run four to six weeks with a named owner and compare against the recorded baseline.

The opportunity is rarely where the loudest tool marketing points. It is in the repeated, unglamorous steps that quietly consume the week, and finding them takes a map of the work rather than a licence agreement. See how I approach this in practice.

Common questions

How do I find AI use cases in my company?: Map a process that runs at least weekly, break it into steps, and score each step on frequency, clarity of a good output, availability of input, risk and hours consumed. The highest scoring low-risk steps are your use cases. Starting from tools instead of steps is the most common reason AI programmes stall.
Which marketing tasks are best suited to AI?: Brief structuring, adapting assets across formats and markets, transcreation, first-draft copy from an approved brief, research and competitor summaries, performance reporting, and tagging and routing. They repeat often, have a recognisable good result and carry low risk when reviewed.
What should never be handed to AI?: Positioning and prioritisation, final creative and commercial sign-off, regulated or legally sensitive claims, negotiation, and anything involving personal or confidential data outside sanctioned tools. AI can prepare the material for these; a person decides.
How many AI pilots should we run at once?: Two or three at most. Every pilot draws on the same limited attention, and a small number of measured pilots produces better evidence than a broad programme nobody has time to evaluate.
How do we measure whether an AI opportunity paid off?: Record baseline hours, elapsed time, review rounds and error rate before the pilot, then compare the same measures after four to six weeks. Without a baseline you cannot tell improvement from enthusiasm.

Last updated: 2026-09-09