“Which use cases will actually pay back?”
Strategy and roadmap.
You leave in two to four weeks with a ranked roadmap of three to eight use cases, a one-page business case for each, and the list we scored and rejected, with reasons.
Consulting
Analysis got cheap.
Judgment did not.
For fifty years, consulting sold analysis at a markup. Rooms of junior people gathering data, reading the market, building the deck. AI does most of that now, in hours, for a fraction of the cost. So the question is no longer who can analyze more. Everyone can. It is who reads the evidence, makes the call, and puts a name on it. That part does not scale, and it is the part we sell.
- Baseline
- dated
- Kill criterion
- written first
- Decision
- signed
“Now that AI can do the analysis, what are we actually paying a consultant for?”
The pyramid
The old model is quietly broken.
The pyramid billed you for the base. A wide layer of analysts, doing work that took human time, because human time was the scarce input. It is not scarce anymore. One specialist with AI now covers what used to take a team, so paying pyramid prices for analysis buys you a constraint that no longer exists.
So what you buy changes
You still need a consultant. What you need one for has changed. When the analysis is cheap, the value moves to the parts AI cannot do: deciding what the numbers actually mean, owning the recommendation, and calling a stop when the evidence says so.
// who does the analysis
The analysis
hours, at a fraction of the cost
Judgment and accountability
unchanged
The split
What we let AI do. What we will not.
Everything on the left got faster and cheaper in two years. Everything on the right is why you hire a person.
AI does the analysis
- Reads your market, competitors, and documents in hours.
- Scores use cases on feasibility, data readiness, and impact.
- Drafts the roadmap, the business cases, the diagnostic.
- Watches a live pilot, metric and cost, in real time.
A person makes the decision
- Sets the baseline. A number counts only if someone dated it.
- Writes the kill criterion. Defining failure is a judgment call.
- Signs the recommendation. When it is wrong, a model cannot answer for it.
- Reads the room. The politics and the hard conversations are not in the training data.
AI removes the limit on how much thinking gets done. It does not remove the need to decide what matters. Intelligence is scalable. Accountability is not.
The shape
So the work looks different.
We did not start out working this way. One engagement taught us, and it is worth telling straight. A food and drink company came to us wanting AI across the whole marketing surface at once. Five jobs in one pilot, and running all five together was the first mistake. Every pilot now runs the same shape, and this is that engagement, walked through it.
Five jobs, one pilot
- Content
- Media
- Brand visibility
- Market monitoring
- Creative assets
// only one survived the narrowing, and that is why it worked
The lesson that runs the practice
Context is everything, and context comes from data. You cannot prompt your way past thin inputs. Fixing them is a data job, and it comes before the AI one.
A person owns this stageAI runs this stage
Weeks 1–2a person owns it Scope and context
a person sets the baseline and writes the kill criterion
Narrow to one metric and check the material behind it. This is the stage the engagement skipped: five jobs went in at once, and the first outputs came back off. Copy that missed the brand voice, and assets generic enough to belong to any competitor on the shelf.
Weeks 3–4AI runs it Foundations, if needed
fix the inputs before the AI touches them
The reflex was to blame the model, but the AI was working in a vacuum: brand guidelines scattered across old decks, product data that did not agree with itself, tone references pulling in three directions. So we stopped and fixed the source material. One agreed set of brand guidelines instead of five, product facts that finally matched across documents, and a single place the voice was defined. This is where a pilot is really won or lost.
Weeks 5–6AI runs it Production on real data
cost and quality visible, judged against the standard
We re-ran production on the repaired inputs, and the quality came up. AI runs the work at volume with cost and quality visible, reviewed against the brand standard, not against a demo.
Weeks 7–8a person owns it Measure and decide
a person signs the call: scale, adjust, or stop
Same measurement, taken again. The content use case scaled. The other four were parked, in writing, until their foundations were ready. A person signed that call.
Getting started
Four ways in.
That is the shape. Where an engagement starts depends on where you are.
“Are we even ready?”
Readiness assessment.
A gap list priced in effort, every gap tied to the use case it blocks, in two to three weeks. Where the honest finding is foundations first, that is what the document says.
“How do we prove it works before we scale it?”
The pilot.
It runs on AZIMUTH, our five-phase method, in the four-to-eight-week shape above, from dated baseline to signed decision.
“It works, but nobody uses it.”
Adoption and operating rhythm.
The last one to two weeks build the operating rhythm around it: spend caps, clear ownership, and training on real accounts instead of a sandbox.
// end to end, when an engagement starts from nothing
Readinesstwo to three weeks
Roadmaptwo to four weeks
Pilotfour to eight weeks
Adoptionone to two weeks
Most engagements start in the middle of this line rather than at its head, and every stage ends with something you can act on without booking the next one.
The lines
What we will not do.
Saying what we do is easy. What we refuse to do is the part that tells you how we work.
Recommend a build when a rule, a spreadsheet, or a hire is cheaper. In writing.
Run a pilot without a written kill criterion. No definition of failure, not ready.
Let a model make a call no person will sign.
Create dependency. Every deliverable is built to run without us.
FAQ
Questions we get.
If AI does the analysis, why are you not cheaper than everyone else?
The AI work does lower our cost, and that shows up in the price. But the invoice was never for the analysis. You are paying for the baseline, the decision, and the accountability behind it. The firms worth watching are the ones charging old pyramid rates while quietly pocketing the AI savings.
Do you work with our existing agency or IT partner?
Yes, and it is common. We define the metrics and run the evidence; your partners keep executing.
What size of company do you work with?
Both ends. Small and mid-sized companies, and large corporates. Size matters less than people expect, and it does not cut the way you would assume: smaller companies often move faster, because there is no legacy data estate to untangle and no committee sitting between the decision and the work. Large organizations bring scale and deeper data, and pay for it in coordination. What we need on either side is a team that owns a real number. The budget to pilot properly usually starts in the low five figures.
Can we start with something small?
Yes. The readiness assessment exists so you can judge how we work first.
Who actually does the work?
The people you meet in the first conversation. There is no pyramid underneath us, because the layer a pyramid was built to sell is the layer that stopped being scarce. What that changes for you is simple: the person who sets the baseline is the person who signs the recommendation, and they are in the room when you argue with it.
How much of our time does this take?
Less than most engagements, and it is concentrated rather than spread. Expect a working session a week with the team that owns the metric, plus access to the systems and documents behind it. The one commitment we do not flex on is the person who owns the number: if they cannot make the sessions, the baseline is guesswork and the pilot is worth postponing.
What if the honest answer is that we should not build anything?
Then that is the deliverable, in writing, with the reasoning and the cheaper alternative priced beside it. A rule, a spreadsheet, or a hire beats a model more often than the market admits. The roadmap always ships with the list we scored and rejected, so the “no” is documented as carefully as the “yes”.
What happens if the pilot fails?
It gets stopped, on a criterion written before it started. That is the point of defining failure up front: nobody has to argue their way out of a project that is not working, and nobody quietly rebrands a miss as a learning. You keep the baseline, the measurements, the cost per run, and a written record of what did not clear the bar, which is what makes the next attempt cheaper.
What happens to our data, and do our documents feed anyone’s model?
No. We work under your confidentiality terms, on your infrastructure or on accounts configured with training switched off, and vendor terms that claim rights on your inputs get flagged before anything is uploaded. Nothing from your engagement is reused as an example elsewhere without your written agreement.
Do you know our industry?
We ask about it rather than claim it. The parts that transfer are the mechanics: the baseline, the cost per run, the kill criterion, the adoption rhythm. The parts that do not transfer are yours, and the first two weeks exist to load them properly. Where a sector rule genuinely decides the design, regulated data, residency, product claims, we say so early and bring your people or your counsel into the design instead of guessing.
Related work
What this looks like in practice
Three engagements where the consulting work was the deliverable: definitions argued out before extraction, pricing logic written down for the first time, constraints agreed before anyone opened an editor.
Executive reporting
An executive dashboard the CEO can question
Data sat in five systems and reached the CEO through five people. Definitions took three sessions. One metric was published as a known gap instead.
One metric published as a gapRead the case→
Sales and pricing
Quotations assembled from client context
Pricing logic lived in three people’s heads. Writing it down was most of the project. Sales now checks and sends rather than assembling from scratch.
The pricing logic, written downRead the case→
Group finance
A shorter close, with the ledger untouched
Coding, matching, and commentary drafting are pattern work. The ledger stayed off limits by design. Constraints were written before the build, and auditors were briefed early.
Nothing touches the ledgerRead the case→
July 202613 min read
AI orchestration: the architecture after one model
One model rarely ships. Orchestration coordinates specialized models and tools, then adds decision points you should govern before you build.
Read the piece→
July 202613 min read
Code or platform? The question after the POC
The demo worked. Now: build it in code, or run it on a platform? Four questions, ordered so the first hard answer ends the discussion.
Read the piece→
June 20269 min read
12 NotebookLM prompts that do real work
Most people give NotebookLM lazy prompts and get book reports back. Twelve our team uses to pull real work out of it instead. Each prompt included in full and formatted to copy.
Read the piece→Talk to us
Talk to us about AI.
A conversation with the senior team about your markets, your data, and the one or two places AI is worth trying first. No slides, no obligation, and if the honest answer is that AI is not your next move, you will hear that too.