How long does AI strategy consulting take? A realistic timeline
How long does AI strategy consulting take?
AI strategy consulting takes two to four weeks at Eazyware, or ten working days for a narrower discovery sprint. Four weeks covers several candidate use cases and a roadmap. Anything longer usually means the scope grew, not that the work is harder. Access delays are the commonest cause of slippage.
AI strategy consulting takes two to four weeks. Eazyware runs AI Product Strategy and Use-Case Discovery across that window from $4,250 or ₹2,80,000, and a narrower ten-working-day discovery sprint from $3,250 or ₹2,00,000. Engagements that stretch past four weeks have almost always grown in scope rather than in difficulty.
Below is the week-by-week shape of a four-week engagement, how much of your team's time it consumes, the five things that reliably add weeks, what can safely run in parallel, and how soon after it you see something working.
The four-week engagement, week by week
Each week ends in an artefact, not a status call. If a week ends without its artefact, the engagement is behind and both sides should say so in that meeting rather than the next one.
| Week | Focus | Your team's time | Artefact at the end |
|---|---|---|---|
| Week 1 | Framing: watching the work, writing a measurable problem statement | 8 to 12 hours across sponsor, product owner and two operators | Problem statement with baseline, target and a guardrail metric |
| Week 2 | Assessment: data, systems, permissions, regulation | 6 to 10 hours from data owner, platform engineer and security lead | Readiness verdict per candidate, with named remediation |
| Week 3 | Ranking and design: value models, target architecture, evaluation plan | 4 to 6 hours, mostly review | Ranked shortlist, rejected list, architecture, cost forecast |
| Week 4 | Commitment: scope lock, build estimate, rollout design | 3 to 5 hours plus a decision meeting | One funded build with locked scope and a date |
The ten-day version compresses this by assessing one or two candidates rather than five or six. It skips nothing structurally: there is still a readiness verdict, still an architecture, still a build estimate. The AI Discovery Sprint exists because many buyers already know which workflow they mean and need the feasibility answer, not the ranking.
Four weeks is the ceiling rather than the default. Roughly half the engagements we run finish in three, because the framing week eliminates candidates fast: a workflow whose volume turns out to be forty cases a month does not need a value model, it needs a polite paragraph explaining why it was dropped.
What adds weeks, and what does not
Five things extend the calendar. None of them is model selection, which is a day of benchmarking against your own task, and none is writing the document.
- Access delays. The single largest cause. Read access to a database or a sandbox account can take a fortnight to approve in a regulated organisation, and the assessment cannot start without it.
- Too many candidate use cases. Each additional workflow adds roughly two days of interviews, data checks and value modelling. Six candidates is a four-week engagement, not a three-week one.
- Absent decision-makers. If the sponsor is unavailable in week four, the commitment gate slips and everything behind it waits.
- Regulatory review. Work touching personal data under the DPDP Act, or health records under hospital policy, adds a residency and consent workstream of three to five days.
- No ground truth data. If nobody has a set of historical cases with known correct outcomes, assembling two hundred of them is real work and belongs in the plan.
What does not add weeks: the number of business units consulted, as long as one person can decide; the choice of vendor model, which is a benchmark; and the length of the final document, which correlates inversely with its usefulness.
What can run in parallel
Three workstreams compress the calendar without compressing the thinking. Start access requests the day the contract is signed, not the day the assessment begins. Assemble the ground truth set during week one and two, in parallel with framing, because it needs your people rather than ours. And run security review alongside design in week three rather than after commitment, since a security objection raised in week five reopens the architecture.
One thing that must not run in parallel: ranking before assessment. Scoring use cases you have not checked the data for produces a confident list of things that cannot be built, which is the most expensive artefact a strategy engagement can hand over.
Sequencing also protects the fixed price. A vendor working to a fixed date has to refuse late additions, and the honest way to do that is to agree at the start that anything discovered after the ranking gate goes on a second list rather than into the current engagement.
How many people do we need to interview?
Fewer than most buyers expect. Four to six people per candidate workflow, chosen for role rather than seniority, surface almost everything that matters. Nielsen Norman Group's finding that five test users reveal most usability problems holds for process discovery for the same reason: the first few participants expose the common failures, and later ones repeat them.
Watching two full cases end to end beats twenty interviews. What people describe is the documented process; what they do is the actual one, and AI systems have to handle the actual one.
How long until something works?
The short answer: three weeks after strategy for a working proof, six weeks for a production MVP, and eight to sixteen weeks for most scoped builds. Those windows run consecutively after commitment, not inside the strategy engagement.
A three-week ProofRun proves the hardest step on your own data with measured accuracy and cost per task, priced at $6,250 to $10,500 or ₹4,00,000 to ₹6,80,000. A six-week Launch 6 build produces an MVP with evaluations and observability, from $26,500 or ₹17,60,000 through our AI-accelerated MVP programme. What the six weeks actually contains is set out in what a six-week AI MVP actually contains. All prices are published on the pricing page.
Add shadow-mode running time on top before the system acts alone. For an agent that changes things in your systems, expect two to four weeks of proposing actions that people approve, which happens after launch rather than before it.
One scheduling point buyers underestimate: the gap between engagements. If the strategy work ends in late December or in the fortnight before a financial year close, the build will not start for weeks regardless of how fast the strategy ran. Plan the commitment gate to land when the people who fund it are actually available.
How much of your team's time does it take?
Between 30 and 50 person-hours across a four-week engagement, concentrated in the first fortnight. That is the number to put in the plan, because it is the cost most often missed and the constraint that most often slips the date.
The people whose time matters most are the operators who do the work today and the platform engineer who can grant access. A sponsor who attends only the gates is fine. A sponsor who cannot be reached at a gate is not.
Hours are not the whole cost. The operators you take out of the line for a morning of observation are the people currently handling the work, so schedule those sessions away from month end, peak season or exam weeks. A discovery interview during a support surge produces a distorted picture of the process and an annoyed manager.
When a longer engagement is the wrong answer
A six or eight-week strategy engagement is almost always a symptom rather than a scope. It usually means the organisation is using the engagement to build consensus, and consensus-building is not something a vendor can be paid to do on your behalf. Run the two-week version, take the verdict to the people who disagree, and let the evidence do the arguing.
Equally, do not compress below ten days by dropping the data assessment. A one-week engagement that skips looking at real records is a workshop, and workshops produce agreement rather than feasibility. The discipline that actually shortens delivery is scope lock, covered in scope lock: the discipline that makes fast MVPs possible.
What a real timeline looked like
A university running a fifteen-year-old ERP wanted to know whether AI could sit on top of it or whether the system had to be replaced first. Assessment took longer than usual because read access to the student records database required a committee, and that approval was the critical path for nine days. The eventual answer was modernisation without a rewrite, described in the university ERP modernisation case study. The lesson generalises: start access requests before the kick-off call.
Nine days of that engagement were spent waiting rather than working, and the fixed price absorbed it because the delay was foreseeable. The version of the same project where access arrives on day one finishes a full week earlier for exactly the same fee.
How to compress the calendar honestly
- Raise read access and sandbox account requests the day the contract is signed
- Cut the candidate list to three workflows before week one starts
- Name one decision-maker per gate, with a named deputy
- Book all four gate meetings in the calendar before the engagement begins
- Start assembling two hundred historical cases with known outcomes in week one
- Bring the security lead into week two, not week four
- Agree in advance that a negative readiness verdict ends the engagement early
Related reading
The AI discovery sprint explains the ten-day option, from POC to production: the checklist covers the phase after, and our AI Product Strategy and Use-Case Discovery service lists the artefacts each week produces.
Two to four weeks is enough for any strategy question worth asking; if a vendor needs longer, ask which week is doing the work.
Frequently asked questions
How long does an AI readiness assessment take on its own?
▾
Three to five working days once access is granted, covering data availability, retrievability, permissions, integration surface and regulatory constraints per candidate use case. Access approval is usually the longer part of that window in regulated organisations, so raise the requests before the assessment is scheduled rather than during it.
Can AI strategy consulting be done in one week?
▾
Only by dropping the data assessment, which is the part that changes decisions. Eazyware's shortest engagement is a ten-working-day discovery sprint at $3,250 or ₹2,00,000, which still inspects real records and produces a build estimate. A one-week version is a workshop, and workshops produce agreement rather than feasibility.
How soon after strategy can we have something in production?
▾
A three-week ProofRun gives you a measured proof on your data, and a six-week Launch 6 build gives you a production MVP with evaluations and observability. Most scoped builds take eight to sixteen weeks. Agents that act in your systems then run two to four weeks in shadow mode before acting alone.