AI Consulting Services: Building Systems That Scale Past 100 Crore
TL;DR summary
- Past ₹100 crore, AI projects stall because locations disagree on the number, not because the technology underperforms.
- AI consulting services open with a process and controls audit across locations, never with a tool demo.
- Your statutory auditor already reports on your internal financial controls. Any model touching a financial entry sits inside that scope.
- Fix four things in order. Process maps. Master data across locations. Control points. Deployment.
- Full DPDP obligations apply from 13 May 2027, so anything you deploy in 2026 needs consent and retention design built in now.
Direct answer
AI consulting services establish where artificial intelligence creates measurable value in your business, then sequence the process, data and control changes that must happen first. AI implementation services come next. They cover build, integration with your existing systems, testing and adoption. For a company operating across multiple locations, the consulting layer exists to stop you deploying a model on top of numbers your own units do not agree on.
Your Chennai plant reports one closing stock figure. Your Coimbatore unit reports another. By the time both roll into the consolidated statement, your finance team has spent four days reconciling the gap, and the number your board reviews is already a fortnight old.
Now someone is asking you to approve a forecasting tool that reads from all of them.
This is where AI implementation services quietly fail in mid market Indian businesses. Not on the model. On the fact that the same component carries three codes, two units of measure and a different tax treatment depending on which location entered it.
Why AI projects stall in companies past 100 crore
This is the gap AI consulting services exist to close. At ₹20 crore, the problem is that nothing is documented. Past ₹100 crore, plenty is documented. The problem is that each location documented it differently, and nobody reconciled the definitions when the second and third units came up.
It surfaces in three places.
Every location defines the metric its own way. Chennai books freight into cost of goods. Coimbatore books it below the line. Both are internally consistent. Consolidated, your gross margin by product line means very little, and a model trained on it learns the inconsistency.
The same item carries different identities. One component, three codes, two units of measure, and a different HSN treatment across your GST registrations. Turnover gets counted across all registrations, not per state, so the reconciliation burden compounds as you add locations.
Decision rights stop at the location boundary. Your unit heads control purchase within limits. Nobody holds authority to act on a group level forecast, so the output lands in three inboxes and moves nothing.
The cost of that sits on your balance sheet as excess raw material cover across locations, funded by working capital you borrowed. Add the four days of finance team effort each month that reconciliation consumes, and the annual figure is usually larger than the consulting engagement your promoter is hesitating over.
The threshold your auditor already crossed
Two rules apply to your business at once, and most promoters know only the first.
Under Ministry of MSME notification S.O. 1364(E), effective 1 April 2025, a medium enterprise means investment up to ₹125 crore and turnover up to ₹500 crore. So at ₹150 crore you still register as an MSME and still access those benefits.
The second rule is the one that matters here. Under Section 143(3)(i) of the Companies Act 2013, your statutory auditor must report on the adequacy and operating effectiveness of your internal financial controls. Private companies are exempt only where turnover stayed below ₹50 crore and aggregate borrowings from banks, financial institutions or any body corporate stayed below ₹25 crore during the year. Past ₹100 crore, you left that exemption behind some years ago.
Read the two together. You are an MSME by classification and a fully reporting company by audit obligation. Which means the moment a model influences a purchase approval, a provision or a valuation entry, it sits inside the control environment your auditor already tests. A recommendation with no traceable approval path is a control gap, and it does not stop being one because a model produced it.
What AI consulting services actually deliver
Most boards believe they are approving software. Serious AI consulting services are a systems engagement, and the budget line behaves accordingly.
AI consulting services cover four things. Diagnosis of where the value sits. Control design around the decisions that will change. Implementation planning. And adoption governance after go live.
Here is the split that decides how much you spend.
| Question | AI consulting services | AI implementation services |
|---|---|---|
| What it answers | Where should we apply AI first, and what breaks if we do | How do we build it, connect it and run it |
| Core deliverables | Readiness audit, use case shortlist, process maps, control matrix, business case | Solution build, ERP and accounting integration, testing, training, measurement |
| Who leads it | Process and finance consultants with audit experience | Engineering and deployment teams |
| When you need it | Before the board commits budget | After the sequence gets agreed |
| What happens if you skip it | Pilots that never reach production | A plan nobody executes |
The sequencing question is where the two disciplines meet, and it is why teams that run business process reengineering alongside ERP work read the risk differently. PKC has delivered more than 100 automation projects and worked across 30 plus ERP platforms, and the recurring multi location finding is the same. The integration rarely fails on the technology. It fails because nobody rationalised the item master and the chart of accounts across units before the model went live, so the model learns three different definitions of the same thing.
Five questions to answer before the board commits budget
Run these with your unit heads in the room, not just your finance team. It takes an hour and needs no outside help.
1. Does every location define this metric the same way, in writing?
2. Does the number in this decision come from one agreed source across all units?
3. Does an exception get caught somewhere before it reaches the consolidated books?
4. Who holds authority to act on a group level recommendation, by name and by role?
5. Can you state today’s baseline for the metric you plan to improve, per location?
Now count your yes answers.
Four or five. Deploy one narrow use case now.
Two or three. Spend eight to twelve weeks on master data and controls first.
Zero or one. Any spend today buys a faster version of your current position.
Most businesses score two on the first pass, and question one is usually where it breaks. That is not a failure of management. It is the normal result of adding locations faster than you standardised the definitions between them.
A four stage sequence for AI implementation services
Order matters here far more than speed. Serious AI implementation services follow this shape.
Stage one. Map and rationalise. Document the process as each location actually runs it, not as the group SOP claims. Build the variance log between units. Three to five weeks. You finish when the differences between locations are written down and deliberate rather than accidental.
Stage two. Fix the data spine. Rationalise the item master and the chart of accounts across units. Assign one owner per field at group level. Reconcile the systems that disagree. Six to ten weeks, and this is the stage promoters consistently underestimate. You finish when the same question returns the same number from every location.
Stage three. Build the control layer. Set the approval matrix, including who acts on a group level output. Define exception thresholds. Decide where a human reviews. Log every decision the model touches. Four to six weeks. You finish when the matrix maps to a process flow your statutory auditor accepts under internal financial controls reporting.
Stage four. Deploy and measure. One use case, one location, one baseline metric, one fixed review date. Six to eight weeks. You finish with a before and after number, then roll to the second location.
Reverse that order and stage four produces entries your auditor will question, because a model influenced approval with no traceable path is exactly the kind of gap internal financial controls testing exists to find.
Can your team handle this internally?
More of it than a smaller company could, and it is worth being straight about which parts, because AI consulting services should never get sold for work your own finance team can finish in a month.
What your team can do without help. Documenting one process per location. Recording baselines before anything changes. Running the five questions above. If you have an internal audit function, it can run the variance log between units. You have the capability at this size. What you usually lack is uninterrupted bandwidth during a closing cycle.
Where it usually breaks. Three places. Item master rationalisation across locations, because merging codes touches your GST records and your unit heads each believe their coding is correct. Control matrix design that survives internal financial controls testing, because the matrix has to map to a process flow an auditor accepts, not to a workflow that merely looks tidy. And the authority question, since standardising a definition across units is a governance decision your promoter has to make, and an internal team rarely carries the mandate to force it.
That last one is the honest reason mid market groups bring in outside help. Not capability. Neutrality between locations.
The compliance clock most boards are ignoring
India notified the Digital Personal Data Protection Rules on 13 November 2025. The rollout runs in three phases. The Data Protection Board came into existence on 13 November 2025. Consent manager registration and the penalty machinery start on 13 November 2026. Full obligations covering notice, consent, data principal rights, security safeguards, breach reporting and retention apply from 13 May 2027. The maximum penalty for failing to take reasonable security safeguards runs to ₹250 crore per instance.
Read that as a build deadline, not a legal footnote.
A system you deploy during 2026 without consent capture, retention limits and deletion logic gets rebuilt before May 2027. You pay for the same integration twice. If you process employee records, customer contact data or vendor details through any model, and most demand planning and collections use cases do exactly that, the obligation applies regardless of your MSME classification.
What your scope document should contain
Ask for the scope before you ask for the price. Any proposal for AI consulting services should name four things you can hold a partner to.
- The process owner on your side, by name and role, and the unit heads who must sign off on standardised definitions.
- The baseline metric per location and the date it gets recorded, before any change begins.
- The control deliverables, meaning the process maps and the control matrix, listed as outputs you keep.
- The handover, covering who runs this after the engagement closes.
On pricing, expect three shapes. A fixed diagnostic fee. Phased fees tied to each stage. An outcome linked component on top. Sequencing the diagnostic first protects the board, because you finish four to six weeks later holding process maps and a variance log that belong to you regardless of what you decide next.
Your first 90 days
| Window | What you do | Who owns it |
|---|---|---|
| Days 1 to 30 | Run the five questions with unit heads present. Pick one process running in two locations. Document both versions. Record baselines. | Promoter and finance head |
| Days 31 to 60 | Rationalise the item master for that process across both units. Assign group level field owners. Set exception thresholds and approval limits. | Finance head with unit heads |
| Days 61 to 90 | Deploy one use case in one location. Log every decision the model touches. Review against the day 30 baseline before extending to the second. | Process lead with your consulting partner |
One process proven across two locations beats five pilots started in five.
Common mistakes to avoid
- Approving a group wide AI programme before two locations agree on one definition.
- Measuring tool adoption instead of the business metric the board approved it for.
- Deploying to every location at once, so no single result is attributable.
- Automating approvals before anyone settles who holds authority at group level.
- Treating master data cleanup as an IT task rather than a finance and governance one.
- Starting with a customer facing use case, where errors reach people outside the business.
Frequently asked questions
What are AI consulting services?
AI consulting services assess where artificial intelligence creates measurable value in your business, then plan the process, data and control changes needed before deployment. The output looks like process maps, a control matrix, a use case shortlist and a business case with baseline numbers per location.
How do AI consulting services differ from AI implementation services?
Consulting answers where and why. AI implementation services answer how. Consulting produces the sequence and the controls. Implementation builds the solution, connects it to your ERP or accounting stack, tests it and trains your team. Run them in that order.
Does an AI system affect internal financial controls reporting?
Yes, once it influences a financial entry. Under Section 143(3)(i), your auditor reports on the operating effectiveness of your internal financial controls. A model driven recommendation acted on without a traceable approval path is a control gap, so log the decision, the reviewer and the override.
Can a multi location group run AI implementation services without standardising master data first?
Only where the use case reads from a single location and a single system. Anything drawing on consolidated data needs the item master and chart of accounts rationalised first, or the model learns three definitions of the same metric and produces output your units will each dispute.
How much should a mid market company budget?
Budget by stage rather than by tool. A diagnostic runs four to six weeks. Master data and control work runs another eight to twelve, longer with more locations. Deployment follows. Ask for each stage priced separately so the board keeps the option to stop after the diagnostic.
How long before AI shows measurable returns?
Expect six to nine months for a first defensible number, and only where you recorded baselines before you started. Groups that skip baseline measurement cannot attribute the improvement afterwards, which is how most pilots quietly disappear at review time.
Start with one process, in two locations
Tools will keep changing. Every quarter brings a new model and a louder launch. The requirement underneath stays fixed. Your locations need one definition, your data needs one owner, and your decisions need a control point your auditor can trace.
Fix that once and every tool that arrives later plugs into ground that holds.
If you are not sure your item master reconciles across locations well enough to support a forecast, that is usually the fastest thing to check. Schedule an Appointment and we will work through one process with you.
