At some point in the last year, someone has probably told you that your business needs to be doing something with AI. Maybe it was a vendor, a conference, a LinkedIn post, or a well-meaning employee. The pressure to act is real, even when the right action isn’t clear.
Here’s a more useful starting point: before buying anything, committing to anything, or putting AI on the agenda as a formal initiative — run through this process. It takes a few hours and it’ll tell you whether there’s a real opportunity worth pursuing, and if so, where.
Step 1: List the tasks your team actually repeats (30 minutes)
AI delivers value in two situations: where tasks are repetitive and follow a pattern, and where a useful starting point is faster than starting from scratch. Everything else is generally hype.
So start here. Gather your key people — or just think through a typical week yourself — and list the tasks that happen regularly, take meaningful time, and follow a reasonably consistent pattern.
Examples from businesses we work with:
- Writing similar emails or proposals repeatedly with slight variations
- Summarising long documents, reports, or email threads
- Answering the same customer questions across email, phone, and chat
- Manually entering data from one system into another
- Generating reports that pull the same information in the same format each week
- Screening or sorting inbound enquiries before routing them
- Creating meeting notes or action item lists after calls
Don’t edit the list yet. Just get it down.
Step 2: Score each task against three criteria (20 minutes)
For each item on your list, score it 1–3 on:
Volume: How often does this happen?
- 1 = occasionally (monthly or less)
- 2 = regularly (weekly)
- 3 = constantly (daily or multiple times a day)
Time cost: How long does it take per instance?
- 1 = a few minutes
- 2 = 15–30 minutes
- 3 = an hour or more
Consistency: How predictable is the pattern?
- 1 = highly variable, requires significant judgement each time
- 2 = somewhat consistent with variations
- 3 = follows a clear, repeatable pattern
Multiply the three scores. Tasks scoring 9–27 are your strongest AI candidates. Tasks scoring 1–4 are weak candidates regardless of how appealing they sound.
This isn’t a precise science — it’s a filter. It keeps you focused on the high-frequency, high-effort, high-consistency tasks where AI actually delivers, rather than the interesting-sounding use cases that look impressive in a demo and don’t change anything in practice.
Step 3: Pick the top two or three and ask the right question (20 minutes)
For your highest-scoring tasks, the question to answer is: what would a useful AI output actually look like for this task?
This is where most AI assessments go wrong. People ask “can AI do this?” (usually yes, in some form) instead of “would the output be useful enough to save us meaningful time?” (often, it depends).
Be specific. For example:
- “We spend two hours a week writing client proposals. If AI produced a first draft that needed 20 minutes of editing rather than two hours of writing from scratch, that’s a real saving.”
- “We get 50 customer enquiries a week. If AI could correctly categorise 80% of them and draft a response for 60% of those, we save roughly four hours a week.”
- “We produce a weekly operations report. If AI could pull the data and produce a draft that needed 15 minutes of review rather than 90 minutes of manual compilation, that’s worth pursuing.”
This framing also reveals where AI won’t help. If the task requires relationship knowledge, professional judgement, or information that isn’t written down anywhere, AI is unlikely to produce output good enough to act on. Knowing that before you start saves you the cost of finding out the hard way.
Step 4: Run a free experiment before you spend anything (variable)
Most AI tools worth considering have free tiers or trials. Before committing to any paid tool or implementation project, test the actual task with the actual tool on real examples.
Take three to five real examples — three actual proposals you’ve written, five real customer enquiries, a real week’s worth of data for your operations report — and run them through the AI manually.
What you’re assessing:
- Is the output good enough to be a useful starting point?
- How much editing does it need?
- Does the time saving hold up in practice, or is the review and editing eating up the gain?
This is the fastest way to cut through the hype. The answer is sometimes yes, this works well. Sometimes the output quality isn’t there. Sometimes it works for some inputs and not others. You won’t know until you test with your actual content, not a demo.
Step 5: Calculate the real return before you commit (15 minutes)
If the experiment is promising, do the arithmetic before spending money.
Example: your team spends three hours a week writing standard client communications. An AI tool reduces that to 45 minutes of editing. You save 2.25 hours per week. At a fully-loaded cost of $60/hour for that person’s time, that’s $135/week, or roughly $7,000/year in recaptured capacity. A $35/month AI tool pays for itself in less than two days of savings.
Do that calculation explicitly — both to justify the spend and to set a realistic expectation with your team about what success looks like.
If the numbers don’t work, the use case isn’t the right starting point. Go back to your scored list and try the next candidate.
What this process is actually doing
It’s forcing specificity in a space where vagueness is the norm. “We should be using AI” is not a strategy. “We’re going to use AI to reduce first-draft time on client proposals by 70%, starting with a two-week trial of Claude’s paid tier, targeting a saving of four hours per week across the team” is a strategy.
The businesses getting real value from AI right now aren’t the ones that ran a big AI initiative. They’re the ones that identified three specific, repetitive tasks, tested carefully, implemented where it worked, and ignored everything else.
What to do if the assessment says “not yet”
Sometimes the honest conclusion from this process is that there isn’t a clear, high-value AI use case in your business right now. That’s a legitimate outcome — and more valuable than spending money on tools that don’t deliver.
“Not yet” doesn’t mean never. It means the current processes aren’t the right candidates, or the tools aren’t mature enough for your specific use case, or the volume doesn’t justify the investment right now. Review it in six months. The tools are improving quickly.
If you’d like help running this assessment for your business — or if you’ve done it and want an independent view on the results — get in touch. No agenda, no preferred vendor. Just a straight read on where the real opportunities are.