TL;DR
Portfolio health metrics like PAR, NPA, and collection efficiency are standard, but most lenders overlook the call-level data that feeds these numbers. Reporting metrics to monitor portfolio health from calls include right-party contact rate, promise-to-pay fulfillment, call dispositions, borrower sentiment scores, and compliance rates. Together, these call-derived metrics act as leading indicators that predict where your portfolio is headed weeks before the financial numbers confirm it.
Why Calls Are a Portfolio Health Data Source
Most portfolio health metrics come from loan management systems and accounting ledgers. They tell you what already happened. PAR went up. NPAs climbed. Collection efficiency dropped. By the time these numbers hit your dashboard, the damage is done.
But for collection-intensive lenders (NBFCs, MFIs, small finance banks), calls are where borrower behavior is actually observed in real time. Every collections call produces data: Did the borrower pick up? Did they commit to pay? Did they sound distressed? Did the agent follow compliance protocols? These signals, when captured and structured, become leading indicators of portfolio health rather than lagging ones.
India’s microfinance sector makes this argument urgent. The Sa-Dhan Report 2025 showed PAR greater than 30 days jumping from 2.1% to 6.2% year-on-year. Gross NPAs in the microfinance sector roughly doubled to 16% by the end of FY25, per Brickwork Ratings. Existing monitoring frameworks clearly aren’t catching problems early enough. The missing layer is structured intelligence from calls.
This guide covers two categories of reporting metrics to monitor portfolio health from calls: the portfolio-level “destination metrics” that boards and regulators care about, and the call-level “source metrics” that feed them. More importantly, it explains how one converts to the other.
If you’re evaluating how voice AI fits into this picture, book a demo with Awaaz AI to see how call data becomes structured portfolio intelligence.
Portfolio-Level Health Metrics: The Destination Numbers
These are the metrics your board, regulators, and investors look at. Each one can be informed, predicted, or moved by what happens on collections calls.
Portfolio at Risk (PAR 30/60/90)
PAR measures the outstanding balance of all loans with at least one installment overdue by a specified number of days, divided by the total gross loan portfolio. It is balance-weighted, meaning a single large delinquent loan affects PAR more than dozens of small current ones.
Formula: PAR>30 = (Outstanding balance of loans with payments >30 days late) / (Total gross loan portfolio)
Why it matters: PAR is the most widely accepted measure of loan performance in microfinance and retail lending. Practitioners recommend using PAR>30 for weekly portfolio monitoring and PAR>90 for provisioning decisions. Never report one in place of the other, as they serve different purposes.
The call connection: PTP fulfillment rates by DPD bucket directly predict PAR movement. If your calls are generating promises in the 1-30 day bucket but only 35% of those promises convert to payments, you can forecast next month’s PAR>30 with reasonable accuracy. Call data gives you the leading signal; PAR confirms it after the fact.
Gross NPA and Net NPA
An NPA is a loan where the borrower has not paid interest or principal for more than 90 days. Gross NPA is the total value of these non-performing assets. Net NPA subtracts provisions and write-offs from the gross figure.
India context: Small Finance Banks reported the highest NPA ratios in microfinance, with 22% of their portfolios turning bad, followed by universal banks at 17.5% and NBFC-MFIs at 12.3%.
The call connection: Accounts that exhaust all call-based collections attempts without a cure migrate into NPA territory. Tracking how many accounts in the 60-90 DPD bucket have received calls, how many produced a PTP, and how many of those PTPs were fulfilled tells you exactly how many accounts will cross the 90-day NPA threshold next month.
Collection Efficiency Ratio (CER)
CER is the proportion of debt collected relative to the total amount due in a given period. It directly measures how well your collections operation is performing.
Formula: CER = (Total amount collected during period) / (Total amount due during period) × 100
The call connection: For lenders where calls are the primary collections channel, CER is essentially the summary metric of call-based collections performance. Every other call metric (contact rate, PTP rate, fulfillment rate) feeds into this single number.
Delinquency Rate and Days Past Due (DPD) Buckets
The delinquency rate shows the percentage of loans that are behind on payments. DPD buckets (1-30, 31-60, 61-90, 90+) segment these delinquent accounts by severity.
The call connection: Call dispositions directly categorize accounts across DPD buckets. When an agent or voice AI reaches a borrower, the outcome (payment made, PTP secured, hardship flagged, refusal to pay) determines whether that account stays in its current bucket, moves to a worse one, or cures. Indian retail credit stress is concentrated in the early DPD buckets, where the majority of accounts will self-cure if they get the right reminder in the right language at the right time.
For a deeper look at how automated calls can intervene at the right DPD stage, see this guide on reducing loan delinquency with automated calls.
Cure Rate
The cure rate is the percentage of delinquent loans that return to current status within a specific period. If 80 out of 100 delinquent loans become current within 90 days, the cure rate is 80%.
The call connection: Successful calls are the most common trigger for cures in the early DPD buckets. A call that produces a genuine PTP, followed by payment facilitation (sending a payment link, confirming UPI details), converts a delinquent account back to current. Cure rate is where call-level effort directly translates to portfolio-level improvement.
Write-Off Ratio
The write-off ratio measures the percentage of loans written off as unrecoverable relative to the total portfolio. It represents the endpoint of failed collections.
The call connection: Accounts that reach write-off have typically exhausted all call-based intervention. Tracking the call history of written-off accounts retroactively reveals patterns: Were they ever contacted? Did they give PTPs that were never followed up? Were calls attempted in the wrong language? This analysis identifies systemic failures in your call strategy.
Concentration Risk
Concentration risk arises when too much of your portfolio is tied to a single loan type, geography, industry, or borrower profile. One adverse event in that concentrated segment creates outsized damage.
The call connection: Call pickup rates and engagement rates by geography and language segment reveal where concentration risk lives operationally. If 30% of your portfolio is in a region where pickup rates are 15% (versus a national average of 26%), that concentration risk isn’t just theoretical. It’s a blind spot in your collections coverage.
Call-Level Metrics That Feed Portfolio Health
This is the layer that every existing guide on portfolio health metrics misses. These are the operational metrics extracted from calls that serve as leading indicators for the portfolio-level numbers above.
Right-Party Contact (RPC) Rate
RPC rate measures the percentage of outbound calls that successfully connect with the actual account holder, not a family member, a wrong number, or voicemail.
Benchmark: Industry averages hover around 26%, meaning 74% of call attempts produce zero useful data or outcomes.
Why it matters for portfolio health: RPC is upstream of everything. If you don’t connect with the right borrower, no PTP gets captured, no payment gets made, and no data gets generated. Practitioners at consulting firms focused on collections analytics emphasize that maximizing RPC rate is the single most impactful thing a collections operation can do. A one-percentage-point improvement in RPC can translate to meaningful recovered principal each month for a mid-sized NBFC with 100,000+ active accounts.
What moves it: Contact-time optimization (calling when borrowers are most likely to answer), caller ID reputation management, and language matching all improve RPC. Voice AI systems can execute these optimizations at scale by analyzing pickup patterns across millions of calls. For a broader look at how conversational AI works in contact centers, that guide covers the infrastructure side.
Promise-to-Pay (PTP) Rate and PTP Fulfillment Rate
PTP rate is the percentage of connected calls where the borrower makes a verbal commitment to pay by a specific date. PTP fulfillment rate is the percentage of those promises that actually convert into payments.
Benchmark: Across Indian BFSI, PTP fulfillment rates for human agents remain stubbornly low at 35-45%. Voice AI systems have pushed this to 55-65% in documented deployments by combining structured follow-up sequences with payment facilitation.
Why PTP is the centerpiece metric: PTP sits at the inflection point between operational activity and financial outcome. It is the single most important intermediate metric in the collections lifecycle. Between the moment of delinquency and actual payment recovery, PTP represents the borrower’s commitment to pay. Every improvement in PTP capture quality and every percentage point gained in PTP fulfillment directly translates to portfolio performance improvement.
The nuance practitioners miss is tracking the relationship between how a promise was obtained and whether it converts. Speech analytics can reveal which commitment-securing techniques lead to real payments versus empty promises. A PTP extracted through empathetic engagement converts at a far higher rate than one extracted through pressure.
Call Engagement and Pickup Rate
The percentage of dialed calls that result in an answered, connected conversation. This is broader than RPC because it includes calls answered by anyone (not necessarily the borrower).
What drives it: Language matching is critical. In Indian collections, 70-85% of borrower interactions happen in Hindi or a regional language. Calling a Tamil-speaking borrower with a Hindi-speaking agent tanks pickup rates on subsequent attempts. Time of day matters too, as does caller ID reputation, since numbers flagged as spam see dramatic pickup declines.
Average Handle Time (AHT)
AHT measures the average duration of a call, including after-call work like updating dispositions or scheduling follow-ups.
Benchmark: Industry norms for collection calls range from 4-6 minutes.
The trade-off: Low AHT may signal efficient calls, but it can also mean agents are rushing through conversations without properly understanding the borrower’s situation or securing a quality PTP. High AHT can indicate complex account handling or inefficient call scripts. The goal is not to minimize AHT but to optimize it, getting enough conversation time to produce a reliable outcome without wasting capacity.
First Call Resolution (FCR)
FCR measures the percentage of calls where the borrower’s issue (or the collections objective) is resolved on the first attempt without requiring a callback.
Benchmark: Best-in-class collections centers achieve 70-75% FCR.
Why it matters for portfolio health: Every unresolved first call means a second call attempt, which doubles the cost and delays payment. In early DPD buckets where self-cure rates are highest, a strong FCR means more accounts return to current status faster, directly reducing PAR.
Call Disposition Taxonomy
Every call should end with a structured disposition code. The standard taxonomy for collections calls includes:
- PTP secured (with date and amount)
- Partial payment made
- Hardship flagged (borrower reports inability to pay)
- Dispute raised (borrower contests the amount or obligation)
- Refusal to pay (borrower explicitly declines)
- Wrong number / not reachable
- No answer
Why it matters: Dispositions are how unstructured conversations become structured portfolio data. Each disposition feeds into different portfolio decisions. Hardship flags trigger restructuring workflows. Disputes require investigation. PTPs enter the follow-up queue. Without consistent disposition coding, millions of calls produce no usable intelligence.
For collections teams standardizing their terminology, the BFSI debt collection glossary provides helpful reference.
Call Compliance Rate
Call compliance rate tracks the percentage of calls that adhere to regulatory requirements: proper caller identification, calls within permitted hours, no threats or coercion, grievance redressal path offered.
Why it’s a portfolio metric, not just an operations metric: In FY2024-25, the RBI imposed over ₹48 crore in penalties on NBFCs and banks for violations in their collection practices. Non-compliance is a direct financial risk to the institution, not just a reputational one. Tracking compliance at the call level, ideally through automated speech analytics rather than manual sampling, is a reporting metric that directly protects portfolio value.
For a comprehensive look at how AI-powered calls maintain compliance during debt recovery, see this guide on AI debt collection calls.
Borrower Sentiment Score
Speech analytics can analyze tone, word choice, and conversational patterns to generate a sentiment score for each call. This score classifies borrowers on a spectrum from cooperative and willing to pay, through distressed and requesting help, to hostile and refusing engagement.
Why it matters: The AI identifies signals in the borrower’s speech that predict payment behavior. Phrases indicating willingness to pay versus resistance are classified and scored. Over time, the models learn which combinations of signals predict actual payment versus empty promises. Aggregated across thousands of calls, sentiment trends reveal borrower stress in specific portfolio segments before it shows up in financial metrics.
To understand how domain-specific NLU powers these kinds of financial conversation insights, that technical guide goes deeper.
Cost per Connected Minute
This operational efficiency metric divides total calling costs (agent salaries, telephony, technology) by the number of minutes spent in actual connected conversations with borrowers.
Why it matters for portfolio health reporting: When cost per connected minute is high, lenders tend to under-call lower-balance or early-stage delinquent accounts. These accounts then migrate to worse DPD buckets, eventually becoming NPAs. Reducing cost per connected minute (through voice AI or better contact optimization) allows lenders to work more accounts per cycle, improving coverage across the portfolio.
Agent and AI Utilization Rate
This metric tracks how effectively your collection team uses working hours for productive outbound and inbound activity versus idle time, administrative work, or breaks.
Why it matters: Low utilization means your collections capacity is underperforming relative to cost. In Indian collections operations, tele-calling attrition runs 40%+ annually, which constantly depresses utilization as new agents ramp up. Voice AI addresses this by maintaining consistent utilization rates regardless of staffing fluctuations. One documented case study showed 150,000 additional calls handled without increasing the number of advisors.
How Call Metrics Roll Up to Portfolio Decisions
Understanding each metric individually is necessary but not sufficient. The real value comes from understanding how call-level data flows upward into portfolio-level outcomes.
The PTP-to-Cure-to-PAR Reduction Loop
This is the core conversion chain:
- Call connects with borrower (RPC)
- Borrower commits to pay (PTP captured)
- Payment facilitation sent (link, UPI, NACH reminder)
- Borrower pays (PTP fulfilled)
- Account returns to current (Cure)
- PAR decreases (Portfolio health improves)
Every break in this chain represents a reporting metric worth monitoring. Low RPC means step 1 fails. High PTP with low fulfillment means steps 2-3 happen but step 4 doesn’t. Tracking where the chain breaks, and for which borrower segments, tells you exactly where to focus improvement efforts.
Every percentage point gained in PTP fulfillment and every day saved in broken-PTP follow-up directly translates to portfolio performance improvement and cost reduction.
Early Warning Signals from Call Data
Speech analytics across thousands of calls can reveal consumer behavior patterns that individual collectors never see because they experience only one call at a time. When aggregated, these patterns become early warning signals:
- A sudden increase in “hardship” dispositions in a specific geography signals localized economic stress.
- Rising borrower hostility scores in a particular loan product may indicate origination quality problems.
- Declining pickup rates in a segment could mean those borrowers are avoiding calls, a precursor to default clustering.
These signals emerge from call data days or weeks before they manifest in PAR or NPA numbers.
Resource Allocation Based on Call-Derived Risk Profiles
A McKinsey analysis of a leading North American bank showed how machine-learning models built on customer interaction data improved estimation of customer risk, identifying customers with a high propensity to self-cure as well as those suitable for early settlement offers. These models saved $25 million on a $1 billion portfolio.
The same principle applies to Indian lenders. Call data informs which accounts need intensive follow-up (skilled human agents) versus which accounts just need a timely reminder (automated voice AI). This segmentation, driven by call-level metrics, determines how efficiently your collections resources convert delinquent accounts into performing ones.
For practical guidance on building this kind of segmented approach, the end-of-month collections planning guide for NBFCs walks through the operational cadence.
DPD-Bucket-Specific Call Strategies
Not all calls serve the same portfolio purpose. Reporting metrics to monitor portfolio health from calls should be segmented by DPD bucket:
1-30 DPD (Soft reminder zone): Key metric is pickup rate and first-call resolution. Most accounts here will self-cure with the right reminder. High-volume, low-cost automated calls are ideal. Track cure rate by call timing and language match.
31-60 DPD (Engagement zone): Key metric is PTP rate and PTP quality. Calls need to be more consultative. Track sentiment scores and hardship flags. This is where speech analytics adds the most predictive value.
61-90 DPD (Intensive zone): Key metric is PTP fulfillment rate and escalation rate. Accounts here need skilled agents or well-calibrated AI. Track AHT (expect it to be higher) and whether payment facilitation was offered on the call.
90+ DPD (Recovery zone): Key metric is right-party contact rate (borrowers actively avoid calls at this stage) and any payment secured. Track call compliance meticulously, as aggressive practices increase at this stage.
Turning Unstructured Call Data into Structured Portfolio Intelligence
Here’s the fundamental problem: millions of calls happen every month across Indian lending institutions, but without structured extraction, they remain just audio recordings sitting on a server.
The gap between “we made 500,000 calls last month” and “here’s what those calls tell us about portfolio risk” is a data infrastructure problem. As one lending technology practitioner put it, the limiting factor for most lenders is not knowing what to monitor. It is having the data infrastructure to do it consistently. Loan-level data that is clean, granular, and accessible in real time separates institutions that catch problems early from those that discover them in an audit.
Speech analytics and natural language understanding (NLU) convert conversations into queryable data. Every call can automatically produce a disposition code, a sentiment score, a compliance flag, and a PTP record, all linked to the borrower’s account in the loan management system. This is what transforms calls from an outreach channel into a data source.
For lenders evaluating how to connect voice AI with their existing systems, this guide on integrating voice AI with collection management systems covers the technical integration layer.
A speech analytics study of a leading debt recovery company revealed that agents failed to follow the company’s “ask for payment” process 60% of the time. Without structured data extraction from calls, this failure was invisible. The portfolio impact (lower CER, higher PAR) showed up in the numbers, but the root cause remained hidden until call-level analytics exposed it.
Building Your Reporting Framework
Putting this together, here’s how to structure reporting metrics to monitor portfolio health from calls across different stakeholders and cadences:
Daily (Operations Team)
- RPC rate by campaign and time slot
- PTP count and PTP value captured
- Pickup rate by language and region
- Call compliance exceptions flagged
Weekly (Collections Head)
- PTP fulfillment rate by DPD bucket
- Cure rate attributed to calls
- Agent/AI utilization rate
- Sentiment trend by portfolio segment
- AHT and FCR trends
Monthly (Risk / Portfolio Management)
- PAR 30/60/90 movement with call-attribution analysis
- Collection efficiency ratio
- Cost per connected minute
- Concentration risk by call engagement (geography, product, language)
- Disposition analysis (shift in hardship flags, disputes, refusals)
Quarterly (Board / Regulatory)
- Gross NPA / Net NPA with call-channel performance context
- Write-off ratio with call intervention history
- Compliance rate and penalty exposure
- Speech analytics insights on systemic portfolio risks
Want to see how structured call data feeds into these reporting layers? Explore Awaaz AI’s approach to converting millions of calls into actionable portfolio intelligence.
Frequently Asked Questions
What is the most important call metric for predicting portfolio health?
Promise-to-Pay (PTP) fulfillment rate. It sits at the exact intersection of call operations and portfolio outcomes. A PTP is the borrower’s commitment; fulfillment is the actual payment. Tracking the gap between the two, segmented by DPD bucket, gives you the most actionable view of where your portfolio is headed.
How does right-party contact rate affect PAR?
RPC is the gateway metric. If your RPC rate is 26% (the industry average), 74% of your call attempts generate no borrower interaction at all. Those unreached accounts continue aging through DPD buckets, eventually increasing PAR. Improving RPC through better contact-time optimization and language matching directly increases the number of accounts that can be worked and potentially cured.
Can call data really predict NPA trends before they show up in financial reports?
Yes. Speech analytics aggregated across thousands of calls reveals borrower behavior patterns, such as rising hardship flags, declining sentiment scores, or increasing refusal-to-pay dispositions, weeks before these trends manifest as PAR or NPA increases. This is the core advantage of monitoring reporting metrics from calls rather than relying solely on financial data.
What PTP fulfillment rate should we target?
Human agent benchmarks in Indian BFSI sit at 35-45%. Voice AI deployments have pushed this to 55-65% through structured follow-up sequences and immediate payment facilitation. Any rate below 35% signals a systemic problem in either PTP capture quality or follow-up processes.
How often should we review call-derived portfolio metrics?
Daily for operational metrics (RPC, pickup rate, PTP count), weekly for bridge metrics (PTP fulfillment, cure rate, sentiment trends), monthly for portfolio-level metrics (CER, PAR attribution, concentration risk), and quarterly for board-level reporting (NPA with call-channel context, compliance exposure).
What’s the difference between call engagement rate and RPC rate?
Call engagement rate measures the percentage of dialed calls that result in any answered conversation, including calls picked up by family members or third parties. RPC rate is stricter: it counts only calls where the actual account holder is reached. RPC is the more meaningful metric for collections because only the borrower can make a payment commitment.
How do we track compliance across thousands of calls?
Manual call sampling (listening to a random 2-3% of calls) is the traditional approach, but it misses most violations. Speech analytics can monitor 100% of calls for compliance markers: caller identification, permitted calling hours, absence of threats, and grievance path offered. Given that the RBI imposed over ₹48 crore in penalties on lenders for collection violations in FY24-25, automated compliance tracking is a portfolio protection measure, not optional overhead.
Does language matching really impact portfolio metrics?
Significantly. With 70-85% of borrower interactions in Indian collections happening in Hindi or a regional language, calling a borrower in the wrong language reduces pickup rates, lowers PTP quality, and increases call-back rates. The downstream portfolio effect is real: accounts that aren’t effectively engaged in their preferred language have lower cure rates and higher PAR contribution.