Now the fundamentals are in place, this is your edge. "AI consulting" is the hottest and most hype-soaked corner of advice work right now. To be credible — and to build a product that isn't snake oil — you need to know exactly where AI creates real value, where it doesn't, and what can go wrong. Your job is to be the calm, honest voice in a noisy room.
The state of play, honestly
The data is sobering and useful. McKinsey's late-2025 survey found that although AI use is now nearly universal, most organisations are still stuck between piloting and scaling, and only a minority report meaningful enterprise-level profit impact yet — the gap is rarely the technology, and almost always the workflows, data, and operating model around it. The lesson: value isn't in "adding AI." It's in redesigning a workflow so AI removes a real bottleneck. Bolt a chatbot onto a broken process and you get a faster broken process.
Where AI genuinely creates business value
- Automating repetitive, language-heavy work — drafting, summarising, classifying, answering routine questions. Clear, provable time savings.
- Augmenting experts — a faster first draft or wider search, so skilled people spend time on judgment, not grunt work.
- Making sense of unstructured data — finding patterns in piles of text, tickets, reviews, or documents no human has time to read.
- Personalisation and triage at scale — routing, recommending, and tailoring across volumes a human team couldn't touch.
The pattern: AI shines where work is high-volume, language-based, and tolerant of a human check. The framing that wins: "where is your team spending hours on work a machine could draft, so they can do the work only they can do?"
Prioritising use cases — impact vs feasibility
Clients will have ten AI ideas; your value is sequencing them. Plot each on two axes — business impact and feasibility (data readiness, technical difficulty, risk). Start in the high-impact, high-feasibility corner to bank an early win that builds belief and budget; park the high-impact, low-feasibility ideas for later (they need groundwork); and quietly drop the low-impact ones however shiny they look. This is the AI version of the impact-effort grid from Module 8, and it's how you turn AI enthusiasm into a credible roadmap rather than a scatter of pilots.
Data readiness — the unglamorous gate
Most AI ambitions hit the same wall: the data is messy, siloed, incomplete, or not allowed to be used the intended way. AI on bad data produces confident bad answers faster. Before recommending an AI build, ask the boring questions — what data exists, is it clean and accessible, who owns it, and are we permitted to use it? Often the honest first recommendation is "fix the data foundation," which is less exciting and more valuable than another pilot.
Where AI is oversold (and you should say so)
- Anywhere accuracy is non-negotiable and unverifiable — models produce confident, fluent nonsense ("hallucinations"). High-stakes outputs need a human in the loop, always.
- When the underlying data or process is a mess — fix the foundation first.
- As a replacement for strategy or judgment — AI is a tool inside a decision, not the decision-maker.
- When a spreadsheet or fixed rule would solve it more cheaply and reliably — recommending the boring solution when it's right is a mark of integrity.
The risks you must be able to discuss
Clients will (and should) ask about these — fluency here is a major credibility signal: accuracy & hallucination (what's the cost of a wrong answer, and what catches it?), data privacy & security (where does the data go, and is that allowed?), bias & fairness (could it disadvantage a group, and how would you know?), regulation & liability (rules are tightening; who's accountable?), and over-reliance & skill erosion (does leaning on AI hollow out the team's capability over time?).
Measuring ROI honestly
The decisive question is "did it pay off?", and it's where most AI projects stay vague. Tie any AI initiative to a measurable before-and-after on a metric that matters — hours saved, cost reduced, conversion lifted, errors avoided — and net out the real costs (tooling, integration, oversight, change management). A pilot that "feels impressive" but can't show a number is a pilot at risk of cancellation, and being the person who insists on the honest measurement protects both the client and your own credibility.
Good AI consulting isn't evangelism. It's matching a specific, high-volume, language-heavy bottleneck to an AI capability, inside a redesigned workflow, gated by data readiness, sequenced by impact and feasibility, with honest guardrails and a measurable ROI. The value is in the workflow change, not the technology. Being the person who says "AI won't help here, and here's what will" is what makes the rest of your advice trustworthy.
This is your subject-matter spine, so it must be the most rigorous part of your product. The app should diagnose whether AI is even the right answer before recommending how, prioritise use cases by impact and feasibility, check data readiness, and model the honesty above — including telling users when AI isn't the fix. An AI consulting tool that recommends AI for everything is the exact credibility trap this module warns against; one that sometimes says "don't use AI for this" would stand out sharply.