Business Opportunities

What every executive should know

1. 1. Test every idea with the same five questions

   Before approving any AI use case, run it through these five questions. They work the same way for a one-person pilot and a company-wide rollout:

   • Does this solve a real business problem, or is it AI looking for a use?
   • Does it save meaningful time or improve quality, enough to justify setting it up and maintaining it?
   • What information would the AI need access to, and does that match the data-handling rules from Briefing 3 and the policy from Briefing 4?
   • What happens if the answer is wrong? A rough first draft of a blog post and a wrong number on a client invoice carry very different weight.
   • Who reviews or approves the result before it goes anywhere that matters?
   • A use case that can't answer all five yet isn't a bad idea. It just needs more groundwork before it's ready to approve.

2. 2. The best first projects share three traits

   Look back at the expensive AI mistakes in Briefing 3, and most of them are missing one of these three. Picking projects with all three traits is most of what separates a useful first pilot from a cautionary story.

   • Frequent: something your team does often enough that the time saved actually adds up.
   • Easy to review: someone who already knows the subject can check the output quickly, without redoing the work.
   • Low consequence if the first draft is wrong: a rough outline is a minor miss; a client-facing number is not.

3. 3. Where this creates value, function by function

   The specific tasks look different by department, but the underlying move is the same everywhere: AI produces a first draft or organizes something large, and someone who already understands the work finishes it.

   Leadership: a board deck built from three departments' updates used to take an afternoon of copying and formatting. AI can assemble the first draft from source material in minutes, leaving leadership time to focus on what the numbers actually mean rather than the layout.

   Sales: a rep who used to spend twenty minutes after each call writing up notes and a follow-up email can now get a solid first draft of both in two, and still personalize and send every message themselves.

   Marketing: turning one piece of long-form content, a webinar, a whitepaper, into five smaller pieces for social and email used to be a half-day task. AI handles the repurposing pass; a person still decides what's worth publishing.

   Operations: procedures that used to live only in a supervisor's head, or scattered across old meeting notes, can be turned into a documented, shareable SOP in an afternoon instead of never getting written down at all.

   HR: a first draft of a job description or a set of interview questions, built from a rough description of the role, gives a hiring manager something to edit instead of something to write from a blank page.

   Finance: a monthly variance report that used to take an analyst half a day to write up in plain language for non-finance leaders can get a solid first draft in under an hour, still checked line by line before it goes out.

   Customer service: a rep facing a complicated ticket can get a summary of the customer's history and a few response options pulled together in seconds, instead of digging through five past tickets to piece it together.

   None of these remove the person from the work. They remove the blank page, which is usually the slowest part.

4. 4. Where a good opportunity turns into Briefing 3's risk

   Three of the items above deserve a second look, because they sit right on the line.

   HR: drafting a job description is a good use of AI. Making the actual hiring, discipline, compensation, or termination decision is not, and Briefing 4 already placed that decision off-limits.

   Finance: drafting commentary and organizing numbers is a good use of AI. Treating the calculations as correct without checking them is the accuracy risk from Briefing 3, just wearing a finance department's clothes.

   Any client-facing deliverable: if AI did meaningful work on something a client will see, Briefing 4's disclosure requirement applies whether the project succeeded or not.

   The review step that makes each of these safe is usually the same step that makes the output better.

5. 5. How you'll know a pilot is actually working

   Before launching anything, decide what "working" looks like, in a form you could show someone else.

   • Time: track how long the task took before, and how long it takes now, for at least a handful of repetitions, not just the first one.
   • Quality: ask the reviewer whether the output needed heavy editing, light editing, or almost none, and whether that's changing over time.
   • Adoption: check whether the people the pilot was built for are actually using it, or quietly going back to the old way.
   • Cost of mistakes: if something did go wrong, was it caught by the review step before it mattered, or after.
   • A pilot with a real answer to all four is worth expanding. A pilot with no answer to any of them isn't necessarily failing, but nobody can tell yet, which is its own kind of problem.

Action plan

This is the fifth in an eight-part series designed to give your leadership team a shared, working understanding of AI, without turning your business into a training exercise.