As India reopens the economics of UPI, the harder problem may not be deciding whether merchants should pay MDR. It may be determining which merchants fall into which category, and applying the rules consistently.
India’s UPI ecosystem has reached a scale where even a narrowly targeted policy change can create a significant technology challenge.
The current debate around the Merchant Discount Rate (MDR) is a case in point.
The Government has clarified that consumers will not face UPI transaction charges. Person-to-person payments will remain free. It has also indicated that any future MDR could apply only to a limited set of merchant transactions above a specified threshold, at a nominal rate. The vast majority of merchant transactions are expected to remain free.
That clarification changes the nature of the technology question.
The issue is no longer simply whether UPI should have MDR.
It is whether banks, acquirers and payment institutions have the merchant intelligence infrastructure needed to apply a selective framework accurately.
UPI MDR Is Becoming a Classification Problem
A universal charge is relatively straightforward to administer.
A selective charge is not.
The moment a policy distinguishes between merchants, transaction values, business categories or other qualifying parameters, the payment ecosystem needs reliable answers to several questions.
Who is the merchant?
What does the merchant actually do?
Which category does the business belong to?
Is its classification still accurate?
Has its business model changed?
Does the transaction qualify under the applicable policy?
And, perhaps most importantly, what happens when the system gets the answer wrong?
These are fundamentally data, classification and decisioning problems.
That makes merchant intelligence an important part of the next phase of UPI infrastructure.
Why Merchant Identity Matters
Merchant onboarding traditionally focuses on establishing whether a business is legitimate.
That remains essential.
But a selective MDR framework would potentially require institutions to know much more than whether a merchant exists.
They may need an accurate and continuously updated picture of the business.
A merchant’s legal identity, tax information, declared activity, operating model and transaction behaviour can all contribute to that picture.
The challenge is that merchant information is not necessarily static.
Businesses expand into new activities. Ownership can change. A retailer can add online sales. A restaurant can introduce delivery. A professional services company can add new revenue streams.
A classification that was accurate at onboarding may therefore become inaccurate later.
That creates a strong case for moving from one-time merchant verification to continuous merchant intelligence.
MCC Accuracy Could Become More Important
Merchant Category Codes, or MCCs, already play an important role in payment processing.
They help identify the nature of a merchant’s business and can influence payment rules, pricing, reporting and risk controls.
If future UPI policies introduce differentiated treatment for particular merchant categories, MCC accuracy could become even more consequential.
An incorrect classification could produce several outcomes.
A merchant could be charged when it should not be.
Another merchant could escape a charge that should apply.
A legitimate business could face unnecessary compliance reviews.
A financial institution could also have to deal with disputes and manual remediation.
The problem is therefore larger than an incorrect database field.
Classification becomes part of policy execution.
AI Can Help — But It Should Not Become the Policymaker
Artificial intelligence could play a meaningful role in this environment.
Modern merchant-verification systems can combine information from multiple sources, identify inconsistencies, assess business characteristics and support MCC classification.
For example, AI-based merchant intelligence can help institutions examine business credentials, digital presence, declared activities and other signals before producing a classification or risk assessment.
CARD91 is one example of a payments technology provider developing this capability. Its merchant-verification offering combines business verification, AI/ML risk assessment and MCC mapping for banks and acquiring institutions.
But there is an important distinction.
AI can support classification. It should not determine public policy.
Whether a merchant is subject to MDR must ultimately depend on the rules established by the relevant authorities and payment ecosystem.
An AI model should help an institution understand the merchant and apply those rules consistently.
It should not independently invent the rules.
The Bigger Shift: From Merchant Onboarding to Merchant Intelligence
This is where the MDR debate exposes a broader transformation taking place in payment technology.
Merchant onboarding used to be largely a front-door activity.
A business submitted documents.
The institution verified them.
The merchant was approved.
The relationship then moved into transaction processing.
That model becomes less effective as digital payments become more sophisticated.
Modern acquiring increasingly requires institutions to understand merchants throughout their lifecycle.
That means:
- verification during onboarding
- classification before activation
- risk assessment
- transaction monitoring
- behavioural analysis
- periodic re-verification
- reclassification when circumstances change
- transparent audit trails
This represents a shift from merchant onboarding to merchant decisioning.
The distinction matters.
A merchant can be successfully onboarded and still have an inaccurate classification.
Likewise, a merchant that was correctly classified at onboarding may require a different assessment months later.
The Risk of Treating Merchant Data as Permanent
One of the weaknesses of static merchant databases is that they assume yesterday’s information remains valid tomorrow.
That assumption is increasingly difficult to sustain.
Consider a hypothetical merchant that starts as a small physical retailer.
Its business later expands into online commerce.
It begins selling a new category of products.
Its transaction volume increases dramatically.
Its ownership changes.
Or its transaction pattern suddenly differs from its historical behaviour.
None of these events necessarily makes the business illegitimate.
But they could change how the institution should understand the merchant.
A modern merchant intelligence layer therefore needs to detect business-profile drift.
The objective should not be perpetual surveillance of legitimate businesses.
It should be proportionate intelligence that identifies material changes while minimising unnecessary friction.
The Fairness Problem Is Just as Important as the Risk Problem
Financial institutions naturally focus on fraud, compliance and revenue leakage.
Merchants have a different concern.
They want predictable treatment.
If two similar businesses are classified differently, one could potentially face costs or controls that the other does not.
That can create disputes.
A selective MDR environment therefore requires more than sophisticated algorithms.
It requires explainability.
A merchant should be able to understand why it has been classified in a particular way and what information influenced that classification.
There should also be a process for correcting inaccurate information.
This is especially important for smaller businesses that may not have dedicated compliance teams.
Bad Data Could Create a New Operational Burden
The scale of UPI makes this particularly significant.
At very high transaction volumes, even a small classification error rate can translate into a large number of exceptions.
That can create a chain reaction.
Incorrect merchant data → incorrect classification → incorrect policy application → merchant dispute → manual review → operational cost.
Technology should therefore not merely automate the existing process.
It should reduce the number of errors entering the process in the first place.
That requires better data quality, stronger verification and continuous monitoring.
The Architecture Behind Selective MDR
If a selective MDR framework eventually emerges, financial institutions will need technology that can connect several layers.
1. Merchant identity
The system needs reliable information about the legal entity and the business operating the payment acceptance point.
2. Business verification
Merchant information should be validated against appropriate authoritative and contextual sources.
3. Merchant classification
The institution needs an accurate understanding of the merchant’s actual business activity.
4. MCC management
Classification needs to translate into appropriate payment-category codes and remain reviewable when circumstances change.
5. Transaction intelligence
The merchant profile should be considered alongside transaction patterns and applicable policy parameters.
6. Rules and policy engines
The system should apply the actual MDR framework and related rules established by the authorities and payment ecosystem.
7. Human oversight
Exceptions, ambiguous cases and disputed classifications should reach appropriately trained human reviewers.
8. Auditability
Every material decision should be traceable.
This last layer may prove particularly important.
When a merchant disputes a classification, an institution should be able to explain not only the outcome but also the information and rules that produced it.
Automation Should Reduce Friction, Not Create It
There is a temptation to interpret greater intelligence as greater control.
That would be a mistake.
The objective of merchant intelligence should be to make legitimate businesses easier to manage while directing deeper scrutiny towards genuinely ambiguous or higher-risk cases.
That means technology should support risk-based friction.
A straightforward merchant with consistent information should not repeatedly face the same verification burden.
A merchant displaying material inconsistencies may require additional checks.
This approach is more scalable than treating every merchant identically.
It also aligns better with the fundamental promise of digital payments: speed and accessibility.
The Human Oversight Question
There is another issue that deserves attention as AI becomes more deeply embedded in payment infrastructure.
Who is accountable when an automated classification is wrong?
An AI model can identify patterns.
It can compare business descriptions.
It can recommend an MCC.
And, it can flag anomalies.
But accountability remains an institutional responsibility.
Banks, acquirers and other regulated entities cannot simply outsource policy interpretation to an algorithm.
A robust architecture should therefore maintain a distinction between:
data → intelligence → recommendation → institutional decision.
That separation becomes particularly important when financial treatment differs between merchant categories.
What Banks and Acquirers Should Prepare For
Even before a detailed MDR framework is established, payment institutions can strengthen the underlying infrastructure.
The priorities are relatively clear.
Clean the merchant database.
Legacy merchant records may contain outdated or incomplete information.
Review classification quality.
Incorrect or overly broad merchant categorisation can create problems well beyond MDR.
Build re-verification capabilities.
Merchant intelligence should not stop when onboarding ends.
Connect merchant data with transaction intelligence.
A merchant profile becomes more useful when institutions can identify meaningful behavioural changes.
Create explainable decision workflows.
Merchants and internal teams need to understand why a classification or action occurred.
Maintain human escalation.
Automated systems should identify and prioritise exceptions rather than remove accountability.
The MDR Debate Is Really About Digital Infrastructure
The debate over UPI MDR is often framed as an economics question.
Who should pay?
How much should they pay?
Who should remain exempt?
Those questions matter.
But once a policy becomes selective, implementation becomes equally important.
A rule is only as effective as the infrastructure used to apply it.
If the underlying merchant information is incomplete, outdated or incorrectly classified, even a carefully designed policy can produce inconsistent outcomes.
That is why UPI MDR merchant classification deserves attention as a technology issue.
The future of India’s payment ecosystem will not depend only on transaction volumes or pricing models.
It will also depend on the quality of the intelligence sitting behind those transactions.
What Comes Next
India does not need to choose between frictionless UPI and stronger merchant controls.
The more realistic objective is to build systems capable of distinguishing between legitimate complexity and genuine risk.
That requires better merchant data.
It requires dynamic classification.
It requires AI that assists rather than dictates.
And it requires transparent institutional decision-making.
The UPI ecosystem was built around simplicity at the point of payment.
If selective economics are introduced, maintaining that simplicity behind the scenes will require considerably more sophisticated infrastructure.
The next UPI challenge may therefore not be collecting MDR. It may be knowing precisely when, from whom and under which rules it should be collected.