The first thing we do when we audit a business for AI opportunity isn't to talk about AI.
We ask questions. Specific ones, in a specific order, designed to get past the self-diagnosis problem, the gap between what a business thinks its biggest AI opportunity is and where the real one actually lives.
We've written before about why that gap exists. The short version: when you're inside a business, you see the output of problems, not their source. The most painful tasks aren't always the biggest time drains. The work that consumes the most hours is often the work people have normalised.
Good questions break through that. Here are the five we use, and why each one is asked the way it is.
The five questions
Question 1
"If I looked at your team's week, what are the 3 things that consume the most hours, not the most important things, the most hours?"
Why this phrasing matters:
The distinction between "most important" and "most hours" is deliberate. Ask for the most important tasks, and you get strategy, the things the business values most. Ask for the most hours, and you get operations, where time is actually going.
These are rarely the same list.
The phrasing also forces people to think rather than reach for the pre-packaged answer. In almost every session, there's a pause here. People know immediately what the most important things are. They often don't know, off the top of their head, what takes the most time. That pause is informative.
What it typically surfaces:
Volume work, the things that consume time predictably and repeatedly. Usually communication, reporting, or data handling. The work that's always there, every week, every month, in roughly the same shape.
Question 2
"What task, when it arrives, makes your team visibly sigh? The one that has to get done but everyone knows it's painful."
Why this phrasing matters:
Emotional signal is the best proxy for friction. The thing that causes a visible reaction when it lands in the inbox is the thing nobody has thought to question. It's always been that way, because the pain has been accepted. After all, it's easier to absorb it than to redesign it.
"Visibly sigh" is specific language. We're not asking what's frustrating in a vague sense. We're asking for the thing with a name. The task with a face.
What it typically surfaces:
High-friction, repetitive work that feels manual in a way that irritates people. This is almost always the best first automation target, not just because it's time-consuming, but because the team will actually use the solution. They're already motivated. You're removing something they've wanted removed for a long time.
People answer this one immediately. They know exactly what it is. Sometimes they laugh. The answer comes without thinking.
Question 3
"What does a new hire spend most of their first few months doing, the work you give them because it doesn't need senior judgment?"
Why this phrasing matters:
Work you give to someone with no experience is, by definition, work that doesn't require expertise. Which means it's work that doesn't require judgment. Which means it's the category of work AI handles best.
The question is a way of asking "what's mechanical in your business?" without using language that requires people to think in terms of AI capability. They don't need to know what AI can do. They just need to tell you what a first-week employee does all day.
What it typically surfaces:
The templated, formulaic layer of the business, the data entry, the standard communications, the recurring reports, the first drafts that always follow the same structure. The work that experienced people also do, just faster and with more confidence in the output.
It also reveals something important: in most businesses, senior people are doing this work too. They just don't think of it as junior work because they're doing it themselves.
Question 4
"Where does work slow down or pile up, what's the step where things sit waiting?"
Why this phrasing matters:
Bottlenecks aren't always obvious from the outside, and they're often not where people expect them to be. The question is designed to find the step that requires manual consolidation, human judgement, or simply someone's attention before the next thing can happen.
Automating a bottleneck gives the most visible relief because the entire downstream process moves faster, not just the step itself. The people waiting for that step notice immediately.
What it typically surfaces:
Handoff problems. Steps that require pulling data together from multiple places before anything can progress. Communication steps that are waiting for one person to draft something. Approval chains that stall because the right information isn't in the right format for the decision-maker.
The answer often points to a workflow problem rather than just a task problem, which is useful because it means the automation opportunity is larger than it initially appears.
Question 5
"If I gave you a smart resource, available 24/7, handles any volume, never gets tired, what's the first thing you'd give them?"
Why this phrasing matters:
This question is a deliberate reframe. People have accumulated assumptions about what AI can and can't do, and those assumptions often get in the way of honest answers. "Could AI do this?" triggers a filter. "What would you delegate to an infinitely capable resource?" doesn't.
The question asks them to think in terms of delegation rather than technology. It surfaces the honest answer: what they would actually offload if they could, without the mental edit that happens when AI is explicitly mentioned.
What it typically surfaces:
Specific, named tasks that people have clearly thought about before. Often something that comes up immediately, because it's been sitting on the list of "things someone else should be doing" for a long time.
This answer is almost always directly actionable. Whatever they say is the first thing to look at for a viable automation.
What we have at the end of 60 minutes
By the time we've worked through these five questions, the picture is usually clear.
We have a rough map of where time is going. We know what's painful, what's high-volume, what's mechanical, and where the system is backing up. We know what the person running the business most wants off their plate.
What we don't have yet is an assessment of which of those things AI can actually address, in what order, with what effort. That comes from the audit itself, the shadow sessions, the process mapping, the analysis that turns the picture into a prioritised list.
But the 60-minute session tells us whether there's a real opportunity, and usually where the biggest one is. In almost every case, the answer surprises at least one person in the room.
Try them yourself
These questions work outside of a formal audit too. Run them on your own team, or on yourself, and pay attention to where the answers don't match what you expected.
If the answers are obvious and unsurprising, the self-diagnosis is probably accurate, and the opportunity is where you think it is. If there are pauses, contradictions, or answers that don't match the official version of how things work, that gap is worth investigating.
If you want to run through these questions with someone who's done it across a range of businesses and knows what the answers typically mean, that's what the free Grapine Discovery Session is for. 60 minutes. No pitch. You leave with a clearer picture of where your biggest AI opportunities actually are.