The AI Consulting Layer

Module 11

The AI Consulting Layer

What AI business consulting really means, where AI delivers value versus hype, prioritising use cases, and the risks you must speak to credibly.

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.

Key takeaway

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.

For Orelis & the app

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.

Test yourself

Q1A retailer says 'everyone's doing AI, build us a chatbot.' How do you respond like a consultant, not a salesperson?
Show a worked answer
Reframe from solution to problem (Module 4): a chatbot is a solution to an unstated problem. Ask what's actually hurting — slow support? lost sales? returns? Then test whether AI fits: is there a high-volume, language-heavy bottleneck, and is the underlying data/process sound (data readiness)? Often the real win is elsewhere, or a narrow use case (triaging support tickets) beats a flashy customer-facing chatbot. Recommend the thing that solves the problem, even if it's less exciting than 'an AI strategy.'
Q2A client has ten AI ideas and wants to do all of them. How do you turn that into a credible plan?
Show a worked answer
Sequence them on impact versus feasibility. Start with the high-impact, high-feasibility idea to bank an early, visible win that builds belief and budget. Hold high-impact but low-feasibility ideas until the groundwork (especially data readiness) is done. Drop the low-impact ideas no matter how shiny. You convert ten simultaneous pilots — a recipe for diffuse effort and no clear ROI — into a phased roadmap with a win up front and the hard, valuable bets properly prepared.
Q3A six-month AI pilot 'feels great' but leadership is asking whether to keep funding it. What was likely missed, and what do you do now?
Show a worked answer
It was likely never tied to a measurable ROI. 'Feels great' isn't a number, so it can't justify continued spend. Now: define the before-and-after on a metric that matters (hours saved, cost cut, conversion lifted, errors avoided), measure it honestly, and net out the true costs including oversight and change management. If it pays off, you have the case to scale; if it doesn't, you've saved the client from pouring money into a pilot that was never going to. Insisting on the honest number protects everyone.