AI Strategy Consulting: What the Deliverables Should Actually Contain

Most owners of Indian small and mid-sized businesses have reached the same conclusion: "we need to do something with AI". The next step is where the money gets wasted. AI strategy consulting is sold under many names, and a lot of what is delivered is a slide deck describing possibilities rather than a plan that tells you what to build first and why.

This post describes what a real AI strategy contains, the deliverables you should insist on, and how long the work should take for a business in Mumbai, Pune, Bengaluru or Delhi NCR with a few dozen to a few hundred staff. It does not cover how to choose between firms (that is a separate question) or how to calculate return on a given pilot, which our guide on measuring automation ROI handles.

What AI strategy consulting is for, and what it is not

A strategy engagement has one job: to turn "we need AI" into "we will run this pilot, on this process, with this measure, by this date". Everything in the deliverables should serve that.

It is not a technology survey. You do not need a comparison of model providers before you know which process you are fixing. It is not a vision statement. And it is not a procurement exercise; vendor selection comes after the pilot is specified, when you know what you are buying.

A useful test: at the end of the engagement, could a competent operations manager who was not in the room pick up the documents and start the pilot? If not, the strategy is incomplete.

Consider a Pune components manufacturer with sixty staff. The owner knows that quotes take too long, that purchase bills are entered twice, and that dispatch confirmations go out late. Good AI strategy consulting does not tell him that AI can help with all three. It tells him that purchase bill entry is the first pilot because the bills already arrive as PDFs in one inbox, the output can be checked against Tally in seconds, and a mistake is caught before payment. It tells him quoting is second because the price list needs consolidating first. That is the difference between a strategy and a brochure.

The five parts of an AI strategy

Whatever a firm calls them, these five parts should be present. Ask to see each one.

1. Process inventory

A list of repetitive, rule-bound work across the business, department by department. For each process: what triggers it, who does it, how often, how long it takes, which systems and documents it touches, and what goes wrong. In an Indian SMB this usually surfaces fifteen to forty candidates, from enquiry handling and quote preparation to purchase bill entry, GST reconciliation, dispatch confirmations and collections follow-up.

The inventory should be built from interviews and from watching the work, not from a questionnaire. Volumes should be real counts from your systems where possible, not estimates.

2. Prioritisation

The inventory is then ranked. A sound method scores each process on frequency, clarity of rules, whether inputs already sit in one place, how easy the result is to check, and the cost of getting it wrong. The output is a shortlist of five to ten processes with the scoring visible, so you can argue with it. Our post on ranking automation workflow candidates describes one such method in detail.

A good consultant will also mark which shortlisted processes need AI at all. Many need only workflow rules. AI belongs where the input is free text or judgement: reading an enquiry, extracting fields from a supplier invoice, drafting a reply for someone to approve.

3. Data readiness

For each shortlisted process, where does the data live, how clean is it, and who can grant access? This is where many plans quietly die. A quoting pilot is attractive until you discover the price list exists in three versions across two laptops. The data readiness check should produce a short table: process, data needed, where it is, condition, and what must be fixed before a pilot can start.

This is also the point to flag personal data. India's Digital Personal Data Protection Act, 2023 requires consent that is free, specific and informed, a notice explaining what is collected and why, and breach reporting to the Data Protection Board and affected individuals; the DPDP Rules notified in November 2025 set an 18-month phased timeline for compliance. If a candidate pilot messages customers or processes their records, the strategy should say what consent you hold and what you need.

4. Pilot plan

One process, specified. Inputs and outputs. The systems it reads from and writes to. The exception rule for when the system is unsure. The named person who reviews output and at what point. The baseline measure taken before the pilot starts, and the same measure taken after a fixed period. A stop rule: what result means you switch it off.

The pilot plan should fit on one or two pages. If it needs twenty, the scope is too large.

5. Governance

Not a committee. A short set of operating rules: who owns the pilot, who can change it, what is logged, how errors are reported, and what the system is never allowed to do without a person (commit a price, make a promise to a customer, move money). The NIST AI Risk Management Framework, a voluntary framework built around four functions (govern, map, measure, manage), is a sensible reference for the questions, and its generative AI profile lists risks specific to language models. A consultant should translate that into a one-page rule sheet, not hand you the framework.

Deliverables to demand in writing

Put this table in the proposal before you agree to it.

Notice what is missing: a technology comparison, a vendor shortlist and a multi-year roadmap. Those belong later, if at all.

How long AI strategy consulting should take

For an SMB with one or two sites and a handful of core systems, a realistic timeline is:

• Week 1 to 2: interviews across departments, process mapping, pulling volumes from systems.
• Week 3: scoring and shortlisting, data readiness checks on the top candidates.
• Week 4: pilot plan and governance rules written, walkthrough held.

Larger businesses with more sites or more systems take longer, but the shape is the same. If a proposal runs to a quarter or more before a pilot is specified, ask what you will be able to measure at the end of month two. The honest answer is usually "nothing yet", and that is the problem.

AI consulting for small business: how to decide if you need it

Not every business needs a strategy phase. Use this checklist.

You probably need AI strategy consulting if:

• You have more than five candidate problems and no agreed way to rank them.
• The work crosses departments or systems, so nobody owns the whole picture.
• Previous automation attempts stalled and you do not know why.
• Customer data is involved and you are unsure what consent you hold.

You can probably skip straight to a scoped build if:

• You already know the one workflow to fix and can explain its rules in ten minutes.
• The inputs sit in one place and the output is easy to check.
• The cost of a wrong first pick is low because you can redo it in weeks.

Either way, the first thing to run should be small, measurable and reviewed by a person.

What to bring to the first workshop

The engagement goes faster, and costs less, if you arrive prepared. Before the first session, gather:

• A list of the systems you run: accounting (Tally, Zoho Books or similar), CRM, inventory, HR, and the shared drives and WhatsApp groups where work actually happens.
• Rough monthly volumes: enquiries received, quotes sent, orders processed, invoices raised, purchase bills entered.
• The three complaints you hear most often from staff and customers about slow or repeated work.
• The names of the people who own each major process, so interviews can be scheduled in week one.
• Any previous automation attempt, what it was meant to do, and why it stopped.

None of this needs to be polished. A page of notes is enough. What matters is that the consultant spends the first week confirming facts rather than discovering them.

Common ways the work goes wrong

• The inventory is built from a survey. People under-report what they do by hand. Walk the floor.
• The ranking is hidden. If you cannot see the scores, you cannot challenge them.
• Data readiness is assumed. Check access and condition before choosing the pilot, not after.
• The pilot has no baseline. Without a before measure there is no after.
• Governance is a committee. One owner and one page beat a steering group.

Next step

Before engaging anyone, spend an hour listing your own repetitive processes and the systems each one touches. That list will make any strategy engagement shorter and cheaper, and it may show you that one process is obvious enough to pilot straight away. If you want help ranking the list or want a second opinion on a proposal, our AI consulting page explains how we run a first engagement, and you can talk to us about your workflow.