The Indian pharmaceutical industry employs, by reasonable estimates, somewhere between four and five hundred thousand Medical Representatives across operating companies. The MR force is the single largest line in any pharma company's commercial budget. It is also, by a wide margin, the most pitched-to function inside Indian pharma when an AI vendor walks into a Brand Manager review meeting. The pitch usually runs along these lines: our system helps your MRs see more doctors per day, file expense reports faster, target the right prescribers, and clean up your CRM. We have seen the deck. You have seen the deck.
This essay is about the gap between that pitch and what actually moves MR-force productivity inside an operating Indian pharma company. The gap is wider than vendors admit and narrower than the more cynical Brand Managers assume. The version of MR-force AI that ships into production is concrete, unsexy, and reasonably well-understood by the teams who have done it. The version that does not ship is the one that confused activity with productivity and called it a day.
What MR-force productivity usually means in the pitch
Read enough vendor decks pitching the MR-productivity workload and a pattern emerges. The proposed measurements are activity-level: doctor calls per MR per working day, percentage of target doctors covered per cycle, average call duration, time spent on administrative versus client-facing work, expense-report turnaround. The proposed mechanism is data-aggregation: pull the MR's call log, pull her territory's prescriber master, pull the call plan from the CRM, run a model. The proposed deliverable is a dashboard for the Territory Manager and a tablet experience for the MR herself.
There is nothing wrong with any of this. The objection is not that it is harmful. The objection is that, by itself, none of this changes the unit Indian pharma cares about, which is moved prescriptions, attributed to specific brands, in specific territories, against the IQVIA market share number that the Brand Manager will be measured against next quarter.
The standard pitch ends one layer above the layer that matters.
Why the standard pitch underperforms
Three reasons. They compound; the result is that the most-pitched MR-productivity workload is also one of the most-pilot-and-stalled workloads inside Indian pharma.
The MR is not the layer where the call plan decisions get made. What the MR does on a given day is, in most operating pharma companies, decided two or three layers above her. The Territory Manager builds the call plan from the prescriber master, the brand priorities, and last cycle's coverage. The Area Sales Manager reviews. The Brand Manager pushes priorities top-down from HQ. By the time the MR opens her tablet, the room she has to make decisions in is small: which order to make the calls in, how to handle a doctor who is not in, when to slip in an off-plan call she thinks is a good idea. An AI tool aimed at the MR herself, without an upstream tool that improves the call plan she has been given, produces a more efficient version of the same problem.
An MR with a better tablet running a worse call plan is a more efficient version of the same problem.
— The shape of the most-pitched MR-productivity failure.
The metrics that prove activity are the wrong unit. Calls per day, coverage percentage, average call duration, expense-report turnaround: every one of these is an input metric. Prescription movement at the territory level, lagged through SMSRC Rx data and IQVIA secondary, is the output. The two are correlated but not interchangeable. An MR can hit her call quotas and her coverage targets and have her brand's IQVIA market share decline in her territory, because the doctors who switched were not the doctors she was prioritising, or because the prescriptions she generated were captured by a competitor's better-stocked chemist, or because the chemist network in her HQ is structurally hostile to her stockist. The standard pitch will measure her as productive. The Brand Manager will measure her as not.
The MR alone cannot tell you which doctors are switchable. She has qualitative signal, accumulated by visiting these prescribers every cycle for years. Her TM has slightly more abstracted qualitative signal, derived from her conversations with the MR and her own field visits. None of this is in any database. Every commercial AI workload that pretends the MR's call plan can be optimised from data alone is missing the most important input. Every workload that pretends the MR's qualitative signal can be ignored is missing the second most important input. The version that works treats the MR's tacit knowledge as a model input, not as noise to be smoothed away.
What the version that actually ships looks like
It is recognisable. It does these things:
It operates one layer up from the MR. The unit of intervention is the Territory Manager's weekly call plan, not the MR's daily tablet flow. The model re-ranks the doctor list each cycle, surfaces confidence scores per doctor, and explains its reasoning in terms a TM already uses: market share movement, recent Rx trend from SMSRC, secondary movement in this chemist cluster, competitor stock-out signal, the doctor's switching history relative to similar prescribers in nearby territories. The MR's tablet still works; the calls still happen; the difference is upstream of the MR.
It reconciles all four data layers against a consistent territory definition. This sounds trivial; it is not. Most Indian pharma companies have at least three different definitions of a territory in active use across primary sales, secondary sales, IQVIA's geography mapping, and SMSRC's panel mapping. A workload that quietly normalises these to a single internal territory master is doing 60% of the engineering work the production deployment will need; a workload that does not, will return numbers nobody trusts within six weeks of going live.
It serves the Brand Manager and the Territory Manager off the same source of truth. The Brand Manager wants to know which of her brands' priority targets moved, against the IQVIA layer, in each region. The TM wants to know which doctors to re-prioritise in her MRs' call plans next cycle. These are the same model used twice with different aggregations. A workload that produces a Brand Manager dashboard the TM cannot drill into, or a TM dashboard the Brand Manager cannot roll up, is going to live in exactly one of those two reviews and die a slow death in the other.
It measures itself in moved prescriptions, not moved tablet sessions. Output metrics are SMSRC Rx and IQVIA secondary, attributed to the brands the system was meant to help, in the territories the system was meant to help, with a 60 to 90-day lag accepted in the measurement plan. Input metrics (calls per day, coverage, etc.) are tracked because they have to be, not because they are the proof of value. The lag is part of the deal; teams that try to shorten it by measuring closer-in metrics quietly substitute activity for productivity again, and the workload's credibility decays.
The unsexy part, which is 80% of the project
If you have built a production-quality MR-force AI workload, the most valuable thing you learned is that the model is the easy part. The data engineering is the project. It looks like this.
Normalising territory boundaries across primary, secondary, IQVIA, and SMSRC. Indian pharma companies acquire and reorganise their field structures often enough that there is rarely a single canonical territory master in clean form. Building one is months of work and a permanent operational commitment to keep it clean as the org changes.
Cleaning stockist secondary feeds. Stockists report secondary sales to the manufacturer with varying degrees of completeness, varying SKU coding conventions, varying delays, and varying degrees of incentive to misreport (positively or negatively, depending on the scheme). The engineering pipeline that turns raw stockist files into a reliable secondary-sales view is opinionated software with operational guardrails.
Mapping IQVIA SKU codes to the internal SKU master. IQVIA does not see SKU IDs the way the manufacturer does. Reconciling them, including the long tail of strength-and-pack variants, is the work everyone underestimates. The reconciliation is also unstable: every time the manufacturer launches an SKU or IQVIA reclassifies a therapeutic area, the mapping has to be updated.
Setting up SMSRC delivery pipelines and tying the panel doctors to the manufacturer's own prescriber master. The doctors SMSRC audits are mostly a subset of the doctors the field force calls on. The intersection is what the workload can actually reason about; the non-intersection has to be acknowledged as an unmodelled tail.
Linking all of the above to a consistent MR-to-TM-to-territory identifier through the field-force pyramid (NSM, RSM, ASM, TM, MR) including org-change history. The day the org reshuffles, the model's attribution math breaks unless this layer is maintained.
None of this is glamorous. All of it is necessary. The vendor that has done it, even once, is a different conversation from the vendor that has not. The Brand Manager who has been through a deployment like this knows the difference, and asks about the territory master in the first thirty minutes of the meeting.
What this means for the meeting where the workload gets bought
If you are the AI vendor walking into the Brand Manager's office, the meeting you need to win is not the demo. The meeting you need to win is the next one, the one in the conference room near the data-engineering team. They will ask you about your territory-reconciliation logic, your treatment of incomplete stockist secondary feeds, your IQVIA SKU mapping, and your SMSRC-to-prescriber linkage. If the answers are concrete, the deployment can probably ship in six months. If the answers are vague, the demo was the highest point of the project.
If you are the Brand Manager being pitched, the question that filters most useful from less useful is exactly the same: ask about the territory master, ask about secondary reconciliation, ask about the SKU mapping, ask about the SMSRC panel coverage in your top territories. The vendor who answers in working detail is the vendor who has shipped this somewhere. The vendor who pivots back to the dashboard demo is not.
If you are running the MR organisation, the question is again the same but from the inside: which of your TMs will own the new call-plan tooling, what reporting line will surface the attribution numbers to the relevant Brand Manager, and what is the change-management plan for the day the territory master gets re-baselined. The answers to these questions decide whether the workload makes it from launch to the cycle after launch.
Why this conversation belongs in the consortium
The MR-force productivity question is the conversation Indian pharma has been having, in some form, for the last twenty years. Generative AI is the third or fourth wave of technology that has been pitched as the answer; each wave has produced some real value and a great deal of frustration, and the boundary between value and frustration has moved each time. The version of the answer that 2026 will produce is a version that takes the field-force pyramid, the four data layers, the Brand Manager review cycle, and the reconciliation work seriously, all at once, by people who already know the vocabulary.
That conversation does not happen well in a vendor pitch. It does not happen well on a webinar. It happens in a room small enough to hold a disagreement, between peers, on a schedule recurrent enough that the disagreement gets resolved by the next edition. That is what AIforPharma is being convened to host.
Applications are open.