AI strategy consulting

AI strategy consultant: what a roadmap should contain and what it should cost

Updated 27 August 2026 9 min read

Short answer

An AI strategy consultant decides where artificial intelligence belongs in a company and in what order — which processes to change, what each will cost, what has to be true for it to work, and what to deliberately skip. A useful roadmap has four things in it that most do not: a measured cost baseline per process, a build estimate per item, a named owner per item, and an explicit not-now list with the reasoning attached. Engagements run four to six weeks. Below roughly thirty people you do not need one yet — you need one working automation and the discipline to measure it.

Every roadmap item is tied to a named process with a measured cost today
Sequenced by payback and dependency, so item one funds item two
Includes what not to do, and why — usually the most valuable page
The engagement ends with a shipped workflow, not a handover meeting

Discuss a roadmap See how the audit works

For companies of roughly 30–300 people where several departments are experimenting separately and nothing is consolidating.

What makes a roadmap defensible

Roadmaps get abandoned when the reasoning behind the sequence is missing, so the next reorganisation reshuffles it arbitrarily. These are the five properties we hold every item to.

01 It names a process, not a capability. "Reduce invoice handling time" is an item; "leverage AI in finance" is not.
02 It has a measured baseline. Hours, headcount, cycle time, or error rate as they stand today, gathered rather than estimated.
03 It has a build estimate produced by people who will do the building, not by a research team handing over to a delivery team later.
04 It names an internal owner who has agreed to it. Items nobody will own get moved to the not-now list rather than quietly carried forward.
05 It states its dependency. If item four requires clean data that item two produces, the sequence is a constraint, not a preference.

The strategy phase is where most AI budgets quietly disappear

Not spent badly — spent on deciding. Companies typically arrive here after twelve months of scattered pilots with nothing consolidated.

Every department bought their own tool

Marketing has one subscription, support has another, finance is running something through a spreadsheet. Nobody can state the total spend, and none of it connects to anything else.

The pilots never ended

Three proofs of concept, all technically successful, none in production. Usually because nobody defined what result would justify a rollout before the pilot started, so there is no threshold to have crossed.

The roadmap has no numbers in it

A list of ambitions ranked by enthusiasm rather than payback. Without a cost baseline per process there is no way to defend the sequence when priorities shift — and they always shift.

How the roadmap gets built

Four to six weeks, structured so you can act on the output before the engagement formally ends.

01

Map the current estate

What is already running, who bought it, what it costs, and what it actually does. Most companies find spend they had forgotten and two teams solving the same problem separately.

You get An inventory of current AI and automation spend, tools, and owners.
02

Cost the processes, not the ambitions

We take the candidate areas and measure what the work costs today: hours, headcount, error rates, cycle time. Without this baseline nothing later can be prioritised honestly.

You get A cost baseline per process, comparable across departments.
03

Assess what has to be true

Data availability, integration surface, approval requirements, regulatory constraints, and whether anyone internally can own the result. This is where most roadmap items get deferred or killed, and that is the point.

You get A readiness assessment with blockers named and owners assigned.
04

Sequence and cost the roadmap

Items ordered by payback and dependency, each with a build estimate and a success measure, plus an explicit not-now list with reasoning attached so the decision survives the next reorganisation.

You get A sequenced roadmap with costs, owners, and success measures.
05

Build item one

The first roadmap item gets implemented by the people who wrote the roadmap. This is where estimates meet reality, and it is the only real test of whether the strategy was grounded.

You get A production workflow and a revised roadmap informed by what building it taught us.

What consolidation typically uncovers

An illustrative estate model for a 120-person company — the shape we see most often when nobody has been tracking departmental AI spend centrally. Run it against your own subscription list.

Four departments, each having independently bought AI tooling over eighteen months, with no central inventory and no shared measurement.

Departments with their own AI subscriptions: 4
Separate tools in use: 9
Tools doing substantially the same job: 3
Combined monthly subscription spend: $2,400
Seats paid for but unused after 90 days: ~35%
Recoverable monthly spend from consolidation alone: ~$800
Pilots technically finished but not in production: 3
Internal hours already invested in those pilots: ~240 h

The consolidation saving is real but small — it typically pays for the assessment, not for the roadmap. The expensive line is the last one: hours already spent on pilots that never shipped because nobody had defined the threshold for rolling one out. That is the loss a roadmap prevents from repeating, and it is the number worth putting in front of a finance director.

What the engagement covers

A strategy engagement is easy to inflate. This is the boundary we hold.

In scope

Inventory of current AI and automation spend across departments
Cost baseline per candidate process, comparable company-wide
Readiness assessment: data, integrations, approvals, ownership
Sequenced roadmap with build estimates and success measures
An explicit not-now list with reasoning
Implementation of the first roadmap item
A governance note: who approves what, and where humans stay in the loop

Not in scope

Organisational design and headcount planning
Vendor selection scored for procurement rather than for fit
Change management programmes
Roadmap items nobody internally is willing to own
Board-level narrative decks — we will support yours, not write it

Engagement shape

Priced as a project with a defined end date, not as an open retainer.

Readiness assessment

$990

Typical market rate: $10,000–25,000

You want to know what is already running and whether the company is ready to scale any of it.

Current-estate inventory
Cost baseline for up to five processes
Readiness and blocker assessment
A go or wait recommendation

Roadmap

$2,900

Typical market rate: $25,000–75,000

You need a defensible, costed sequence you can take to a budget conversation.

Everything in the assessment
Sequenced roadmap with build estimates
Success measures per item
Explicit not-now list

Roadmap plus first build

$4,900

Typical market rate: $40,000 and up

You want the strategy tested by shipping the first item rather than by discussion.

Everything in the roadmap
Implementation of item one
Measured before-and-after
Roadmap revised against what the build taught us

These are the prices for up to four departments and five baselined processes, which covers most companies in the 30–300 range. Beyond that the scope genuinely grows and so does the fee — quoted in writing after a scoping call, always with a fixed end date rather than a rolling engagement.

Why a roadmap costs $2,900 here and $40,000 elsewhere

The gap is real and it is mostly structural. Two of these reasons are advantages; the third is a genuine limitation you should weigh.

No research team handing off to a delivery team

The people gathering the baselines are the people who would build the workflows. A large consultancy has to staff both functions and bill for the handover between them, which is where a lot of the fee goes.

Four to six weeks, not four to six months

The fee is mostly time, and we bound the scope hard: a fixed number of departments and processes. Most of what a longer engagement adds is consensus-building, which your team can do better and cheaper than we can.

We reuse the assessment framework

The inventory template, the cost model, and the readiness checklist are the same every time. You pay for them to be applied to your company, not designed for it.

The honest limitation

We are a small team without the political weight of a recognised brand. If the roadmap has to survive a hostile board or settle a conflict between department heads, that brand is worth paying for and we are not a substitute for it.

How this differs from a management consultancy

Large consultancies are genuinely better at some of this. Being specific about which parts is more useful than pretending otherwise.

Criterion	This engagement	Large management consultancy
Deliverable	Costed roadmap plus a working first build	Strategy document and implementation proposal
Who does the analysis	The people who will build it	A research team, then a separate delivery team
Typical duration	4–6 weeks	3–6 months
Better for	Operational processes with measurable cost	Board-level narrative, market positioning, M&A
Weakness	Not the right partner for organisational restructuring	Recommendations often outlive the team that could execute them

Whether you need a roadmap at all

The most common honest answer at this stage is "not yet".

Several departments are buying AI tooling independently

Start with the readiness assessment. Consolidation usually pays for it, and you will know within three weeks whether a full roadmap is justified.
You have pilots that finished but never shipped

Roadmap engagement. The problem is not capability, it is that nobody defined the threshold for production — that is exactly what the sequencing work fixes.
You know the processes and just want them automated

Skip strategy entirely. Go to the automation consultant page — a roadmap would be overhead between you and a working workflow.
You are under thirty people

Too early. One measured automation teaches you more than any roadmap will, and it costs a fraction of the fee.
A board or investor has asked for an AI strategy

Be careful what gets bought here. A document that satisfies a board and a plan that changes operations are different artefacts. We build the second and will help you present it, but we do not write the first on its own.

A two-week readiness check you can run internally

This is the assessment phase in compressed form. If it produces a clear answer, you may not need the engagement at all.

Days 1–2 Pull every AI and automation line item from the last twelve months of card statements and invoices. Include the ones bought on personal cards and expensed.
Days 3–4 Ask each department head what they are using and what it replaced. Compare against the list — the gap between the two is your real finding.
Days 5–6 For every pilot that ran, write down what result would have justified rolling it out. If nobody can answer, that is why it did not ship.
Days 7–8 Pick the three processes most often named as painful and measure them properly: volume, hours, error rate, cycle time.
Days 9–10 For each, check data availability and integration surface. Note anything that would be blocked by a closed system or an approval requirement.
Days 11–12 Ask who would own each one after launch. Cross off anything with no willing owner — this is usually the shortest list.
Days 13–14 Rank what survives by cost and dependency. That ranking is a first-draft roadmap.

If your team produces that ranking and agrees on it, you may not need a strategy engagement — you need to start building item one. Bring us in when the internal version stalls on disagreement about sequencing, or when nobody has time to gather the baselines, which is the more common outcome.

Depending on where you actually are

Frequently asked questions

What does an AI strategy consultant do?

Decides where AI belongs in a company and in what order: which processes to change, what each will cost, what conditions must hold for it to work, and what to skip. The output should be a sequence you can fund, not a list of possibilities.

How is this different from an AI readiness assessment?

A readiness assessment answers whether you can. A roadmap answers what to do, in what order, and at what cost. The assessment is the first phase of the roadmap engagement and can be bought on its own for $990 if that is the only open question; the full roadmap is $2,900, or $4,900 including implementation of the first item.

Do we need an AI strategy at all?

Under about thirty people, usually not — you need one working automation and the discipline to measure it. Strategy becomes worth paying for when several departments are spending separately and the duplication is starting to cost real money, or when pilots keep finishing without shipping.

How long does a roadmap engagement take?

Four to six weeks for the roadmap itself. Building the first item typically adds another two to four weeks. Longer engagements usually mean the scope was not bounded properly at the start, and the extra time rarely improves the recommendations.

Will you recommend tools we already pay for?

Frequently, yes. A large part of the assessment is finding capability you are already licensing and not using. We take no vendor commissions, so there is nothing to gain from recommending a new purchase over an existing one.

What happens to the roadmap after you leave?

It belongs to you, with the cost baselines and assumptions written down so your team can update it. Roadmaps go stale when the reasoning is missing — the numbers behind each decision matter more than the sequence itself, because the sequence will change and the numbers are what let you re-derive it.

Can you implement the whole roadmap?

We build the first item as part of the engagement and can take on later items, but that is a separate decision made after you have seen how the first one went. Committing to a whole roadmap before item one ships is exactly the pattern that produces abandoned programmes.

How do you handle governance and human oversight?

Every roadmap item specifies where a human approves, what gets logged, and what happens on failure. For regulated processes we design the approval gate first and the automation around it, because retrofitting oversight after the workflow exists usually means rebuilding it.

What if the roadmap concludes we should do less than we planned?

That is a common and legitimate outcome. Two or three well-chosen items beat a twelve-item programme nobody has capacity to execute, and the not-now list exists precisely so those decisions are recorded with their reasoning rather than relitigated every quarter.

Do you work alongside an existing internal AI lead?

Often, and it is the easier version of the engagement. An internal lead usually knows which processes matter but lacks time to gather baselines across departments, or needs an outside view to settle a sequencing disagreement. We do the legwork and hand the roadmap to them to own.

Next step

Scope a roadmap engagement

Tell us how many departments are involved and what is already running. We will come back with a scope, a fee, and an honest view on whether you need a roadmap yet.