The RBI's Fair Practices Code for collections is not aspirational guidance — it is a binding conduct framework with specific prohibitions on language, contact timing, third-party disclosure, and intimidation tactics. Every agent call made to every borrower is either compliant or it is a regulatory and reputational liability in progress. The Collections Head AI listens to every call, scores it in real time, and escalates the ones that matter — before a complaint does.
What the RBI Fair Practices Code Actually Requires
The RBI's Master Circular on Fair Practices Code for NBFCs, supplemented by the Regulatory Framework for Microfinance and the guidelines on digital lending, creates a specific and enforceable set of conduct requirements for collection activity. These are not principles — they are rules, and their violation is a supervisory finding waiting to happen.
The core prohibitions are well-known but inconsistently enforced: agents must not contact borrowers before 7 AM or after 7 PM, must not use abusive or threatening language, must not contact third parties (employers, family members, neighbours) about a borrower's debt without explicit consent, must identify themselves and the institution at the start of every call, must not misrepresent the legal consequences of non-payment, and must not make false threats about legal proceedings that have not actually been initiated. Less well-known but equally binding: agents must not call a borrower on a number not registered in the borrower's loan account record, and must not make repeated calls within a short window in a manner that constitutes harassment.
The institution is liable for every breach of these requirements by every agent of every agency it has contracted — regardless of whether the breach was the direct result of an instruction given by the institution. Principal liability for agency conduct is established in the regulatory framework. This means monitoring is not optional. It is the only way to manage a liability the institution cannot disclaim.
The 8 FPC Rules the AI Monitors on Every Call
Calls only between 7 AM and 7 PM
AI checks call timestamp against permitted window for the borrower's registered state time zone. Calls outside the permitted window are flagged immediately regardless of outcome.
RBI FPC 5(b) · NBFC Master DirectionFull identification at call start
AI transcribes the opening 30 seconds and verifies that the agent stated their name, the institution's name, and the purpose of the call within the first 60 seconds. Failure to identify is a hard FPC breach.
RBI FPC 5(a) · NBFC Master DirectionNo abusive, threatening or intimidating language
NLP model trained on 50,000+ annotated collection calls detects prohibited language patterns — threats, profanity, derogatory terms, and implied physical intimidation — in real time during the call. Alert fires within 30 seconds of detection.
RBI FPC 5(c) · Zero tolerance thresholdNo disclosure of borrower's debt to third parties
AI detects when an agent calls a number not in the borrower's loan record, or when call content suggests a third party (employer, family member) is being engaged about the debt. Immediate escalation to Collections Head.
RBI FPC 5(d) · DPDP Act overlayNo false threats of legal proceedings
AI identifies language patterns that assert legal action, property seizure, or criminal proceedings that are not consistent with the account's actual legal status. False legal threats constitute misrepresentation under the FPC and consumer protection law.
RBI FPC 5(e) · Consumer Protection Act overlayNo excessive repeated contact within short windows
AI monitors call frequency per borrower per agent per day. Three or more calls within a 4-hour window constitutes a harassment pattern under the FPC. AI flags the pattern and suppresses further calls to that borrower for 24 hours from the flagging agent.
RBI FPC 5(b) — frequency clauseContact only via registered numbers
Calls must be made to and from numbers registered in the loan account. AI cross-references every call's from/to number against the borrower's registered contact record. Unregistered number contact is a data protection and FPC breach.
RBI Digital Lending Guidelines 7 + DPDP ActSettlement and waiver offers must match authorised terms
AI compares settlement and waiver terms verbally offered by agents against the authorised resolution matrix for that borrower's account. Agents quoting terms they are not authorised to offer create liability — both legal and financial.
Internal resolution policy · Customer communication standardsThe Live Call Monitoring Feed
The Collections Head AI does not analyse calls only after they are complete. For agencies with integrated telephony, it monitors calls in near-real-time — generating alerts within 30 to 90 seconds of a prohibited language event, giving the supervisor the ability to intervene in an ongoing call before the conduct escalates further. For post-call analysis, every call is scored and the findings are available within 15 minutes of call completion.
The Escalation Architecture: From Detection to Resolution
Real-Time Call Alert to Agency Supervisor
For in-progress calls: alert fires to agency supervisor immediately with the specific conduct issue and the audio timestamp. Supervisor can intervene in the ongoing call. For post-call: scoring complete within 15 minutes.
Collections Head Notified with Evidence Package
For Rule 3, 4, and 5 breaches (language, third-party disclosure, false threats): Collections Head receives an evidence package — call recording, transcript excerpt, agent ID, borrower account, specific FPC rule breached, and a draft response action for approval.
Agency Notified — Agent Suspension Triggered
Repeat or critical breach: Collections Head AI initiates agent suspension from the institution's calling list pending agency investigation. Agency is required to provide a root cause response within 48 hours. All ongoing calls by the suspended agent are paused immediately.
Systemic Agency Conduct Pattern → Governance Escalation
If an agency accumulates 3+ critical FPC breaches in a 30-day window, the Collections Head AI escalates to the Board Risk Committee with a pattern analysis brief. This is not an operational escalation — it is a governance one. The board needs to know when a contracted agency is creating systemic conduct risk.
The Complaint You Never Received Is the One the AI Caught First
Every FPC breach that escalates to a formal borrower complaint costs the institution: complaint management time, potential RBI observation, reputational exposure, and the relationship with a borrower who may have recovered and re-borrowed. The Collections Head AI catches the breach before the complaint is filed. At 2,847 calls a day across 50 agencies, no human team catches them all. The AI catches every single one. That is not a marginal improvement — it is a structural elimination of conduct risk.