AI can now produce respectable first drafts of research, proposals, process maps and technical options. Charging a premium for producing those artefacts alone will become harder.
That does not make consultants obsolete. It moves value towards work that a general model cannot own: discovering the real constraint, gaining access to evidence, making trade-offs, implementing change and accepting responsibility for what happens next.
This is a strategy, not a revenue promise. The old headline's claim that six frameworks could produce a particular income had no supporting evidence and has been removed.
Stop selling the document
A report is useful when it changes a decision. A workshop is useful when it creates an agreed next move. An automation is useful when somebody owns its exceptions.
Package the decision and the operational change, not the number of slides.
For example, replace AI readiness assessment with a specific outcome:
- identify one costly workflow
- map the data, permissions and failure risks
- prototype the smallest useful change
- test it with real cases
- produce an owner, runbook and evidence record
The client can see what exists at the end. You can also say clearly what has not yet been proved.
Six practices that remain valuable
1. Diagnose before recommending a tool
Interview the people doing the work and inspect the artefacts they use. Separate the visible complaint from the mechanism causing it.
AI can help cluster notes, but it cannot verify an organisation's unwritten rule unless you find the people and evidence behind it.
2. Use AI backstage
Use models to challenge a hypothesis, draft alternatives, generate test cases and expose missing questions. Keep confidential material inside tools and contracts approved for that data.
Do not hand an unreviewed model output to a client as expertise.
3. Make evidence visible
For each important claim, record its source, date and confidence. Label estimates and forecasts. When a vendor feature or licence can change, link to the current primary documentation.
This is slower than pretending certainty. It is also much easier to defend.
4. Price the bounded outcome
Define what the client receives, what decisions are included and what remains outside scope. Avoid pricing language that assumes a saving or uplift before you have a baseline.
If the engagement includes a pilot, agree the success and stop conditions before building it.
5. Design for adoption and failure
A good prototype can still fail because the owner leaves, permissions drift, staff do not trust it or exceptions overwhelm the process.
Include handover, monitoring, rollback and an escalation route. That operational design is often more valuable than the first automation.
6. Build reusable methods, not cloned answers
Turn repeated work into checklists, interview guides, test packs and decision records. Reuse the method while keeping each client's evidence and recommendations separate.
A weekly test for your offer
Ask four uncomfortable questions:
- Could a prospect get most of this from a generic AI prompt?
- What private evidence or workplace access makes the work specific?
- Who is accountable for the recommendation and implementation?
- What observable state will be different when the work is complete?
If the answer to the first is yes and the other three are vague, the offer needs work.
What nobody can honestly promise
No consultant can guarantee that a practice is AI-proof. Model capability, buyer behaviour, regulation and competitors will keep changing. Predictions about specific jobs or fee levels are forecasts, not settled facts.
The defensible aim is simpler: get better at selecting real problems, using AI carefully and proving useful outcomes.
For practical ways to turn AI ideas into work that can be checked and handed over, join the Microsoft Copilot & AI Mastery Space.