AI strategy consulting produces a scored list of the places AI will pay in your business, the order to build them in, what each will cost, and whether your data can support them. A useful engagement takes two to twelve weeks and ends with decisions and budgets, not a slide about the future.
Most AI programmes need that discipline. BCG’s survey of 1,000 senior executives in 59 countries found that 74% of companies had yet to show tangible value from AI, and only 26% had built the capabilities to move beyond proofs of concept. MIT NANDA’s 2025 study of enterprise generative AI concluded that 95% of organisations are getting zero return on $30–40 billion of investment. And when RAND interviewed 65 data scientists and engineers about why AI projects fail, the first root cause they identified was that people misunderstand, or miscommunicate, what problem needs to be solved.
Most of these are strategy failures that surface later as engineering problems. This guide sets out what a strategy engagement should produce, how use cases should be scored and sequenced, what it costs, and how to choose who does it — including where a firm like ours is the wrong choice.
What you should have at the end
Judge an AI strategy engagement by what you can do the day after it ends. If the answer is “present it to the board”, you bought a presentation. If the answer is “approve the first budget and start”, you bought a strategy. You should be holding four things.
- A scored use-case register. Every candidate use case the engagement considered, including the ones it rejected, with the score on each criterion and the reason. The rejected ones matter: they stop the same ideas coming back every quarter.
- A sequenced roadmap with costs. The order to build in, with build and run costs for each item and the dependencies between them — which use case builds the data pipeline, the integration or the security approval the next one will reuse.
- A data-readiness assessment per use case. Does the data exist, can you get at it, is it labelled, is it clean enough, and what would it cost to fix? Gartner predicts that through 2026 organisations will abandon 60% of AI projects that lack AI-ready data. This is the document that prevents that.
- A build, buy or partner decision per use case. Some use cases are best served by a product you can license next month. Some need a custom build. Some need a partner who will run it. The roadmap should say which, and why.
Each item on the roadmap should also carry a named owner in the business — the person who will be accountable for the result once it is live. An item without an owner is not ready to fund.
How use cases get scored
Scoring turns an argument about opinions into an argument about evidence. The criteria below map directly onto the reasons AI projects fail in RAND’s research — unclear problems, missing data, inadequate infrastructure — plus the two that Indian enterprises in regulated sectors cannot skip: regulatory exposure and ownership.
A starting scoring model — adjust the weights to your business
| Criterion | Suggested weight | How it is measured | Who supplies the input |
|---|---|---|---|
| Value at stake | 30% | Annual rupee impact: hours saved × loaded cost, revenue gained, or losses avoided — with the assumption written down | The function head, checked by finance |
| Data readiness | 20% | Does the data exist, is it accessible, labelled and clean enough, and what would fixing it cost? | The data team |
| Integration effort | 15% | Systems touched, whether they have usable APIs, and batch windows | IT |
| Ownership after launch | 15% | A named owner with a budget for running costs | The sponsor |
| Regulatory exposure | 10% | Personal data under the DPDP Act, RBI or IRDAI touchpoints, and whether the output affects customers directly | Risk and compliance |
| Process stability | 10% | Will this process look the same in twelve months, or is it being redesigned anyway? | The process owner |
Two rules keep the scoring honest. First, the value estimate must name its assumption: “saves 2,000 hours a year at ₹900 an hour” can be challenged; “high impact” cannot. Second, the data-readiness score must come from someone who has looked at the data, not someone who has been told it exists. A roadmap scored by people who have seen neither the numbers nor the tables is a guess with weights attached.
Regulatory exposure deserves its weight in India. A use case that decides credit, pricing or claims for customers carries obligations that a use case summarising internal documents does not — under the DPDP Act, and for NBFCs, digital lenders and insurers under their regulators’ model-governance expectations as well.
Sequencing: why the highest-value use case is rarely the first one you build
The use case at the top of the value column is usually the hardest: it touches the most systems, the most sensitive data and the most people. Build it first and you are solving data access, security approval, integration and change management for the first time, all at once, on the project the board is watching most closely.
The better first build scores well on readiness and ownership, moderately on value, and low on risk. Its job is to prove the path: get the data flowing, get a security review passed, get one integration into production, get one team using it. Everything after it reuses that path and costs less. Then the high-value use cases follow, on foundations that have already been tested.
The evidence supports patience. Gartner found that 45% of organisations with high AI maturity keep their AI initiatives in production for three years or more, against 20% of those with low maturity. McKinsey’s State of AI 2026 found that nearly three-quarters of the highest performers had fundamentally redesigned workflows around AI, against about a quarter of everyone else. The winners sequence and redesign; they do not just deploy.
The operating model: who does what once the roadmap is approved
A roadmap without an operating model is a list. Gartner’s 2026 survey found that only 22% of organisations have scaled AI across multiple business units, and that the high performers reported positive returns on 81% of their AI initiatives. The difference is rarely talent. It is who decides, who pays and who owns.
Four arrangements do most of the work:
- An owner for every use case, in the business. Not in IT and not in a centre of excellence — the person whose numbers improve if it works.
- A small central team for the shared parts. Architecture, security patterns, evaluation standards, vendor management and the integration layer. Its job is to make the second use case cheaper than the first, not to build everything itself.
- Funding released in stages. Proof of concept, then pilot, then production, with a written decision at each gate and a running budget approved alongside the build budget.
- A quarterly review of the register. Every live and planned use case is continued, changed, scaled or stopped against its original business case. Stopping things is a sign the process works.
For regulated entities there is a fifth. The Reserve Bank’s FREE-AI framework recommends that regulated entities adopt a board-approved AI policy and governance across the AI lifecycle. The strategy engagement is the natural place to draft it, because it is the first time anyone has listed every AI use case the business intends to run.
What an engagement costs and how long it takes
Global strategy firms do not publish rates. Indian AI engineering firms do, broadly: Clutch’s AI pricing guide, updated in September 2026, lists India at $25–49 an hour. Clutch’s business consulting pricing guide lists business strategy consultants at $100–149 an hour. The table uses the Indian band and shows the arithmetic, so you can substitute a quote.
Indicative cost by scope, at India rates of $25–49 an hour
| Scope | Duration | What you get | Typical team | Indicative cost |
|---|---|---|---|---|
| Readiness assessment | 2–3 weeks | Data and infrastructure readiness, and 5–10 candidate use cases with first-pass scores | 1–2 people | $2,000–11,800 (about ₹1.8–10.3 lakh) |
| Use-case discovery and scoring across two or three functions | 3–6 weeks | The scored register and business cases for the top three | 2 people | $6,000–23,500 (about ₹5.3–20.7 lakh) |
| Enterprise AI roadmap | 8–12 weeks | The sequenced roadmap with costs, business cases, operating model and governance | 3–4 people | $24,000–94,100 (about ₹21–83 lakh) |
Keep the strategy spend in proportion to the programme it serves. In the EY-CII AIdea of India outlook, more than 95% of Indian organisations said they allocate less than 20% of their IT budget to AI, and 47% already have several generative AI use cases live. For most mid-sized enterprises, a readiness assessment followed by a proof of concept is a better use of the first ₹20 lakh than a twelve-week roadmap.
Who sells AI strategy, and what each kind is good at
Five kinds of firm sell AI strategy in India. None is right for everyone. We have put ourselves in the table because leaving us out would be odd — and the column on where each falls down applies to us too.
The five kinds of AI strategy provider
| Type of firm | Examples | What you get | Where it falls down | Choose it when |
|---|---|---|---|---|
| Global strategy consultancies | McKinsey (QuantumBlack), BCG (BCG X), Deloitte, EY (EY.ai) | Board-level alignment, cross-industry benchmarks, operating-model redesign | Cost, and often a hand-off to a different firm for the build | The programme spans countries and business units, and the board needs the mandate |
| Large IT services firms and systems integrators | Accenture, TCS, Infosys (Topaz), Wipro, HCLTech | Delivery capacity at scale, and existing contracts and relationships | Roadmaps can lean towards large programmes and the firm’s own platforms | You need hundreds of people and already have a master services agreement |
| Analytics and AI specialists | Fractal, Tredence, LatentView | Deep analytics and data science, industry accelerators | Check who runs the system after launch, and how | The use cases are analytics-led: forecasting, pricing, customer analytics |
| Build-and-run AI engineering firms | AIVeda | Strategy from people who build and run production AI; costs in engineering terms; a proof of concept within weeks | A smaller bench; not the choice for an enterprise-wide change programme across many countries | You want the roadmap and the first build from the same team |
| Independent consultants and freelancers | Marketplace listings | Speed and low cost — Upwork lists AI strategy and roadmap projects at $4,000–10,000 | Thin bench and no delivery capacity | Early exploration before any budget is set |
One finding is worth weighing before you decide. In the MIT NANDA study, AI built with external partners reached deployment about 67% of the time, against about 33% for tools built entirely in-house — although the authors caution that the comparison may not account for every confounding factor. The partner matters less than whether the partner stays through deployment.
Five ways a roadmap dies
Roadmaps rarely fail because the use cases were wrong. They fail in one of five predictable ways, and each has a test you can apply before the roadmap is approved.
- No owner. The use case belongs to a steering committee and therefore to nobody. Test: can you name the person who will be accountable for its results in production?
- Assumed data. The data was said to exist and was never examined. RAND lists missing training data as the second root cause of AI project failure. Test: has someone opened the tables?
- A regulatory surprise. The use case touches customer decisions or personal data and meets compliance at the end rather than the beginning. Test: has risk and compliance scored it?
- No budget to run it. The build was funded; the cost of keeping a model accurate, monitored and supported was not. Test: does each item have a running cost for years two and three?
- The sponsor leaves. The one senior person who believed in it changes role, and the roadmap goes with them. Test: is the roadmap tied to business targets that survive a change of sponsor?
RAND’s remaining root causes — chasing the newest technology rather than the user’s problem, inadequate infrastructure, and problems too hard for AI — show up earlier, in scoring. A roadmap that passes all five tests above has usually already dealt with them.
From roadmap to proof: where the 4-week POC fits
The first item on a good roadmap ends in a proof of concept with a written go or no-go decision, not a demonstration. That distinction matters. Gartner found that at least half of generative AI projects were abandoned after proof of concept by the end of 2025, usually for poor data quality, weak risk controls, rising costs or unclear value — the same things the scoring model is meant to catch. A proof of concept designed as a decision tests exactly those things, on your data, against a success measure agreed in week one.
That is how our 4-week AI proof of concept is built: one use case, one agreed metric, your own data, and a written recommendation at the end — build, change or stop. The roadmap tells you which use case to prove first; the proof tells you whether to fund the build. If the build goes ahead, the data and integration work typically follows the patterns in our guides to data engineering consulting and AI integration services.
We offer AI consulting as a company that also builds and runs the systems it recommends, so the people who score your use cases are the people who would have to deliver them. Built fully custom, or accelerated by our own LLM and inference stack where it speeds delivery — the choice is yours. When the roadmap calls for an enterprise-wide platform, our enterprise AI solutions cover the build and the run.
Frequently asked questions
What is AI strategy consulting?
AI strategy consulting identifies where AI will create value in a business, scores and sequences the candidate use cases, assesses whether the data can support them, estimates build and running costs, and recommends whether to build, buy or partner for each. It ends with a roadmap and budgets you can act on.
How much does an AI strategy consultant cost in India?
Indicatively ₹1.8–10.3 lakh for a two-to-three-week readiness assessment, ₹5.3–20.7 lakh for use-case discovery across two or three functions, and ₹21–83 lakh for a full enterprise roadmap, based on India rates of $25–49 an hour. Global strategy firms charge considerably more and rarely publish rates.
How long does it take to develop an AI strategy?
Two to three weeks for a readiness assessment, three to six weeks to discover and score use cases across a few functions, and eight to twelve weeks for an enterprise-wide roadmap with business cases and an operating model.
Do I need an AI consultant?
Not always. If you already know the use case, have the data and have someone to own the result, a proof of concept may be the better first spend. A consultant earns their fee when there are many competing ideas, unclear data, regulatory exposure, or a board that needs a defensible plan.
How do you build an AI roadmap?
List candidate use cases, score each on value, data readiness, integration effort, ownership, regulatory exposure and process stability, then sequence them so the first build proves the data, security and integration path the later ones reuse. Attach build and running costs and a named owner to each item.
How do you prioritise AI use cases?
Use a weighted scoring model with the evidence recorded behind every score. Prioritise high readiness and clear ownership for the first build, even over higher value, and move to the highest-value use cases once the foundations are proven.
How do you choose an AI strategy consulting partner?
Match the firm to the job: global consultancies for multi-country transformation, systems integrators for scale, analytics specialists for analytics-led use cases, and build-and-run engineering firms when you want the roadmap and the first build from one team. Ask who will still be involved when the first system goes live.