Iro AIWhat you'll be able to do
- Scope a market or industry in an hour with a source-flagged first brief
- Structure a messy problem with the right framework (MECE, pyramid, 2x2)
- Turn analysis into a slide narrative a partner would present
- Synthesize a stack of interview notes into themes and evidence
- Catch a confident, wrong figure before it reaches the client
Inside the path
A focused set of five-minute lessons. Each one ends with a hands-on exercise, not a quiz you can guess.
Scope a market fast 6 min
Prompt patterns for sizing a market, mapping players, and getting a source-flagged first brief.
Structure with a framework 5 min
Get AI to organize a messy problem MECE, or into a 2x2 or issue tree you can defend.
Build the slide narrative 6 min
Turn analysis into a pyramid-principle story: one recommendation, the arguments, the evidence.
Synthesize interviews 5 min
Cluster interview notes into themes with quotes and a flag on where the signal is thin.
Verify before it ships 5 min
The checks that catch an invented number or source before it lands in a client deck.
Try a sample exercise
This is the kind of card you'd practice inside Iro: you do the thinking, then get feedback.
◆ Sample exercise · Prompt practice
You're building the executive-summary slide for a client's cost-reduction project. You already have the analysis; you need AI to structure the story for a CFO. "Make me a slide" gives you a generic bullet dump that buries the recommendation.
Your task: Pick the prompt that produces a slide narrative a partner would actually present.
- "Make a slide summarizing our cost-reduction findings."
- "Act as a management consultant preparing an executive-summary slide for a CFO. Using the pyramid principle, structure the findings below into one governing recommendation, three supporting arguments, and the key evidence under each. Keep it to what a CFO cares about — impact, cost, and risk — and flag any claim that isn't backed by the data I've given you. Findings: [paste]."
- "Write a 10-slide deck about cutting costs for this client."
- "What cost savings should I recommend to the client?"
See why the second prompt wins
The winning prompt sets a role (management consultant), names a framework (pyramid principle: one governing recommendation, supporting arguments, evidence), targets the audience (a CFO who weighs impact, cost, and risk), and asks the model to flag any claim not backed by your data. It structures analysis you already did rather than inventing it. The losing options either ask the model to make the recommendation ("what should I recommend?") or produce unstructured volume ("a 10-slide deck") with no story and nothing verified. In Iro you'd write your own and get feedback on role, framework, audience, and where you left the model room to guess.
Why AI fits consulting — and its one big risk
Consulting is research, structure, and communication under a deadline, and AI is fast at all three. It can pull together a first market brief, organize a tangle of findings into a framework, and shape a slide narrative in minutes instead of an afternoon. For a job that runs on billable time, that speed is a real advantage.
The risk is that a deliverable carries your firm's name. AI will state a market size, a growth rate, or a competitor fact with total confidence and no source, and if that slides into a deck, it's your credibility, not the model's. So the rule is simple: AI drafts and structures; you verify every number and claim before the client sees it.
Where AI earns its keep on an engagement
- Research: first-pass market and industry briefs, company backgrounds, sharper questions for expert calls.
- Structure: MECE issue trees, 2x2s, hypothesis lists, framework-driven analysis.
- Communication: slide narratives, executive summaries, first-draft client emails and memos.
- Synthesis: clustering interview notes and survey data into themes with evidence.
What stays with you: the recommendation, the judgment behind it, and responsibility for every figure on the page. AI accelerates the work; it doesn't sign off on it.
How do you turn 20 interview notes into a synthesis?
The mistake is asking for "the key themes" in one shot. That makes the model compress twice at once, and you get five headings bland enough to describe any industry. Split it into three jobs instead: pull the claims out, group them, then try to break the groups.
Extract, one interview at a time. Run each set of notes separately: "From the notes below, pull every distinct claim as one sentence. Tag each with the speaker's role and whether it's a fact, an estimate, or an opinion." Say explicitly not to summarize or merge yet. You end up with a claim list you can sort, count, and trace back to a named person.
Group the claims, not the notes. Feed it the claim list and constrain the grouping: "Every theme needs at least three claims drawn from two different interviews. List any claim that doesn't fit a theme separately instead of forcing it in."
That orphan list is worth reading twice. The finding nobody expected usually sits there, one person saying something odd that turns out to be right.
Then try to break it. Ask which themes rest on a single source, where interviewees contradict each other, and what a skeptic would say is missing. Those are the first questions in a partner review, so you'd rather hear them from a model on Tuesday.
The so-what is still yours to write. Iro's Prompt Lab, part of Pro, is where you practice multi-pass prompting like this and get feedback at each step instead of discovering the gaps on a live engagement.
Where consultants hand AI too much of the thinking
Three habits show up on almost every engagement, and they share one root: the model got handed a decision that was yours to make. The cost never lands in the moment. It lands two days later, in a partner review.
Asking for the answer instead of the issue tree
"What should we recommend here?" returns confident, generic strategy: consolidate vendors, renegotiate contracts, pilot then scale. It reads fine and falls apart in the first partner review, because you can't say why those options and not four others. Ask for the issue tree, the hypotheses worth testing, and the data each one needs. Then go test them.
Letting it size the market
A market size is the easiest thing to ask for and the hardest to defend in a room. Ask for the method instead: which segments to count, which multiplier, which public sources carry the inputs. You pull the real numbers yourself. Any figure that reaches a slide should trace back to a named source with a year on it.
Generating the whole deck before the story exists
Ten slides of plausible filler take longer to fix than five real slides take to build. Get the narrative right first (one recommendation, the arguments under it, the evidence under those), then build pages against that spine. If you can't say the storyline out loud in thirty seconds, formatting won't save the deck.
The pattern underneath all three is timing: where in the work your judgment enters. Early, and AI genuinely saves you hours. Late, and you're auditing a stranger's reasoning against a deadline.