Imani Brooks manages a seven-person marketing operations team at a mid-size events company. For months she felt the same low-grade pressure every manager feels right now: leadership kept asking what her team was doing with AI, and she kept giving vague answers because she did not actually know where it would help. Her first instinct was to grab the flashiest tool she had seen demoed and roll it out to everyone. She caught herself just in time. Instead of starting with a tool, she started with her team's work, listing what they actually did all week, and within an afternoon she had a ranked shortlist of opportunities and a clear top candidate to pilot. This lesson walks through exactly what she did, because spotting the right opportunity is a skill, not a lucky guess.

What This Lesson Covers

The goal here is not to adopt AI everywhere. It is to find the specific spots where AI creates real value for your team, and to ignore the spots where it does not. Most managers make one of two opposite mistakes: "AI is everywhere, let's use it for everything," or "AI is hype, we don't need it." Both are wrong. The truth sits in the middle. AI helps a lot in some contexts and not at all in others, and your job is to tell them apart deliberately rather than by mood or marketing. This lesson gives you a simple way to inventory your team's work, score each task, separate good candidates from bad ones, and pick the one or two worth piloting first.

What A Good Candidate Looks Like

Before scoring anything, it helps to know the shape of a good opportunity. AI earns its keep on work that is repetitive, high-volume, and language-heavy, where the output is reviewable before it does any harm. The strongest candidates cluster in a few categories. Drafting: emails, proposals, announcements, first-pass copy. Summarizing: meeting notes, long documents, email threads. Research and extraction: pulling key facts out of reports, categorizing feedback, organizing messy information. These are tasks where a fast, imperfect first draft that a human then reviews is genuinely useful, and where the volume is high enough that small time savings add up.

The common thread is the time sink. If your team does something many times a week, dislikes it, and the task is mostly about producing or digesting text, you are probably looking at a real opportunity. Imani noticed her team groaned every time they had to write the weekly event recap or sort through post-event survey responses, and those groans turned out to be a reliable signal.

What A Bad Candidate Looks Like

Just as important is knowing where not to point AI, because a confident wrong answer in the wrong place is expensive. Avoid AI for high-stakes, irreversible decisions and for anything touching regulated or sensitive data. Specifically, keep it away from performance reviews and hiring or firing, which need human judgment, empathy, and carry legal risk. Keep it away from compensation and financial decisions, where the legal implications are serious and the context is subtle. Keep it away from high-stakes predictions like deciding which employee is "high risk" or assessing someone's credit, which carry real ethical and legal complications. And be cautious with anything involving regulated data, such as customer health records, where compliance rules govern what you can even feed a tool.

A simple test: if a wrong output would be hard to undo, would harm a specific person, or would put you on the wrong side of a regulation, it is a bad candidate no matter how much time it would save. Time savings never outrank those risks.

A Simple Triage: Frequency, Time, and Feasibility

You do not need a fancy model to rank opportunities. Imani used three numbers per task that anyone can estimate in a few minutes.

• Frequency: how many times per week the task happens.
• Time per instance: how long one instance takes, in minutes or hours.
• Feasibility: how good a fit the task is for AI today, on a 1 to 5 scale, factoring in AI fit, data readiness, and how risky or sensitive the work is. A task that is a natural text-drafting fit with no sensitive data scores a 5; anything touching regulated data or needing deep human judgment scores low.

The math is deliberately simple. First compute the time the task consumes each week as frequency times time per instance. That tells you how big the prize is. Then weight it by feasibility, because a huge time sink that AI cannot safely touch is not actually an opportunity. A clean way to rank is: weekly time cost in hours, multiplied by feasibility. The tasks with the highest product are your primary targets. This keeps you honest, a painful task that AI simply cannot do well falls down the list where it belongs, even if everyone hates it.

Worked Example: Imani Scores Her Team's Tasks

Imani listed six recurring tasks her team complained about or spent real time on. For each she estimated frequency per week, minutes per instance, and feasibility from 1 to 5. Then she calculated weekly hours and multiplied by feasibility to get a triage score.

• Drafting post-event recap emails: 10 per week, 30 minutes each = 5 hours/week. Feasibility 5 (pure drafting, no sensitive data). Score: 5 x 5 = 25.
• Summarizing attendee survey responses: 4 per week, 45 minutes each = 3 hours/week. Feasibility 5 (summarization and categorization, AI's sweet spot). Score: 3 x 5 = 15.
• Updating the weekly status report for leadership: 1 per week, 90 minutes = 1.5 hours/week. Feasibility 3 (some drafting help, but high-stakes wording needs her judgment). Score: 1.5 x 3 = 4.5.
• Manually reconciling vendor invoices: 8 per week, 20 minutes each = 2.67 hours/week. Feasibility 1 (financial data, needs accuracy and accountability AI should not own). Score: 2.67 x 1 = 2.7.
• Researching venues for upcoming events: 3 per week, 40 minutes each = 2 hours/week. Feasibility 4 (research and first-pass shortlisting, human picks). Score: 2 x 4 = 8.
• Writing individual performance feedback notes: 2 per week, 30 minutes each = 1 hour/week. Feasibility 1 (judgment, empathy, legal sensitivity, a clear bad candidate). Score: 1 x 1 = 1.

Ranked by score, the order is clear: recap emails (25), survey summaries (15), venue research (8), status report (4.5), invoice reconciliation (2.7), performance notes (1). The two losers are instructive. Invoice reconciliation eats nearly three hours a week and everyone hates it, but its feasibility of 1 (financial accuracy, accountability) correctly pushes it down. Performance notes feel painful too, but they need exactly the human judgment AI lacks. The triage protected Imani from her own instinct to "save time" on precisely the tasks where time savings would have created risk.

Pick The Top One Or Two, Then Pilot

The biggest mistake at this stage is enthusiasm. Imani's six tasks produced four plausible candidates, and the temptation was to tackle all of them. She did not. Chasing too many at once creates adoption fatigue, the team gets overwhelmed and none of the changes stick. So she picked the top two: recap email drafting and survey summarization. Both were high time cost, high feasibility, and low risk, the ideal profile for a first win.

She ran them as small pilots rather than mandates. For recap emails, one team member used AI to draft and the others kept doing it by hand for two weeks, so Imani could measure the difference. The drafting cut the 30 minutes per email to about 8 minutes of editing, roughly 3.5 hours saved per week for that one person, and the recaps actually read more consistently. She kept her estimate conservative on purpose: she told leadership "this could save a few hours a week," not "this will cut our workload in half," because overpromising sets a pilot up to look like a failure even when it succeeds.

Start Small, Measure, Expand

The pilot is not just a rollout, it is a measurement. Imani tracked three things: time actually saved, output quality (did the AI drafts need heavy rework, or light edits?), and team reaction (did people want to keep using it?). After two weeks the recap pilot was an easy yes, so she expanded it to the whole team and brought survey summarization online as the next pilot. Only once those two were humming did she look at the next tier, venue research. This start-small, measure, expand rhythm is what separates managers who get durable value from AI from those who launch five tools and quietly abandon all of them.

Involve Your Team And Avoid False Opportunities

One more thing Imani did right: she did not build her list alone. She asked the team directly, "what takes the most time, what is most frustrating, what would actually help you?" Their answers reshaped her scoring, the survey summarization pain was bigger than she had assumed, and a task she thought was painful turned out not to bother anyone. Your view and your team's view always differ, and theirs is closer to the work.

Finally, watch for false opportunities, the ones that look attractive but are not. "This tool is so cool, everyone must need it" is a shiny object, not a problem you have. "A competitor uses it" does not mean it fits your context. "The salesperson says it will transform our business" is a sales pitch, not evidence. And "we can't not do AI" is fear, not strategy. Being a little late with a good fit beats being early with a bad one. Start from a real problem on your ranked list, and let that, not the hype, decide where you point AI.

Practice: Build Your Own Shortlist

Do what Imani did, on your own team, this week. Set aside a single afternoon and write down every recurring task your team performs, not the projects, the routine work that happens again and again. Aim for six to ten items rather than an exhaustive list, because the point is to surface the obvious time sinks, not to audit the department. Then put three numbers next to each one: how often it happens in a week, how many minutes a single instance takes, and a feasibility score from one to five that honestly reflects AI fit, data readiness, and risk. Multiply weekly hours by feasibility and sort the list.

Then check your work against two questions. First, did anything score high that involves a person's livelihood, a regulated data set, or a decision you could not easily reverse? If so, your feasibility number was too generous, revise it down. Second, did anything you personally find painful land near the bottom? Sit with that. The triage is doing its job when it tells you that a task you hate is still yours to do.

Before you commit to a pilot, take the list to your team the way Imani did and ask three plain questions: what takes the most time, what is most frustrating, and what would actually help. Expect the answers to move your rankings. Then reflect on one more thing honestly: was the reason you were looking at AI at all a real problem on this list, or was it leadership pressure, a competitor, or a demo that impressed you? If it was the latter, you have just avoided the most common and most expensive mistake managers make at this stage.

Related Lessons

This lesson opens a sequence of four that build on each other, and while you can jump to whichever is most relevant to what you are facing right now, working through them in order gives you the full arc from inventory to habit.

• Mapping Your Daily Workflow goes deeper on the first step Imani took. Understanding what AI can do is one thing; taking a systematic inventory of your own actual week so you can see where it fits is another, and that inventory is what every later decision rests on.
• High Value AI Use Cases for Managers puts concrete examples behind the categories described here. Once you have a mapped workflow and a ranked shortlist, seeing how other managers actually apply AI to drafting, summarizing, and research helps you judge whether your feasibility scores were realistic.
• Tasks AI Should Not Do expands the guardrails that pushed invoice reconciliation and performance notes to the bottom of Imani's list. It gives you the clear boundaries for work where AI should not be the primary tool or the decision maker, which is what keeps a time-saving instinct from turning into a risk.
• Building an AI Opportunity Mindset closes the loop. Spotting one good opportunity is a project; spotting them consistently is a habit, and this lesson covers how to keep noticing the right openings long after the first pilot has become routine.

Key Takeaways

• Start from your team's work, not from a tool. Inventory what your team actually does each week, then look for AI fit. Grabbing the flashiest tool first is how managers waste effort on problems they do not have.
• Good candidates are repetitive, high-volume, and language-heavy. Drafting, summarizing, research, and extraction are where AI earns its keep, because a fast first draft that a human reviews is genuinely useful and the volume makes small savings add up.
• Bad candidates are high-stakes, irreversible, or regulated. Keep AI away from hiring, firing, performance reviews, compensation, high-stakes predictions, and sensitive or regulated data. No amount of time saved outranks that risk.
• Triage with frequency, time, and feasibility. Weekly hours (frequency times time per instance) multiplied by a feasibility score gives a simple, honest ranking that correctly demotes painful tasks AI cannot safely do.
• Verify the scoring catches the traps. In the worked example, recap emails topped the list at 25, while a three-hour invoice task scored just 2.7 because its feasibility was 1, exactly the result you want.
• Pick one or two, then pilot and measure. Too many changes at once causes adoption fatigue. Run a small pilot, measure time saved, output quality, and team reaction, then expand only what works.
• Be conservative and involve your team. Estimate savings modestly so a real win is not seen as a miss, ask your team where the pain actually is, and let a real problem, not hype or competitors, decide where you start.