Affordable housing is getting a lot more scientific. Underwriting is shifting from one-size-fits-all to profile-specific playbooks, and behavior is now a key signal in secured lending. To unpack what’s changing and what still needs fixing, we spoke with Ram Kishan Kolli, Executive Vice President at Credit Saison, who’s led mortgage portfolios across prime and affordable segments. 

Ram’s expertise and experience in this space is invaluable. It spans decades of being at the frontlines of credit, risk, product and business scale. The past, present and future of housing finance market is revealed through these conversations.  

Read on for the full interview.  

Professional journey 

Q: Hi Ram, thanks for joining us. To start, could you walk us through your experience in home loans and secured lending and the big shifts you see shaping the next five years? 

RAM KISHAN KOLLI : 

I’ve spent about 20 years in lending, including roughly a decade in mortgages. My career has spanned wholesale to retail, and within retail, from prime into affordable. Over this period, affordable housing has become far more scientific. Lenders increasingly underwrite by profile: kirana owners, carpenters, cash-salaried workers, rather than applying a single broad template. Large players have matured, core credit-underwriting challenges are better addressed, and borrower segments once overlooked are now clearly defined. With portfolio experience, lenders have leaned into more resilient profiles and de-emphasized those with pronounced seasonality. 

Financial inclusion has reshaped the landscape. With broader unsecured adoption, the “new-to-credit” pool is smaller, and repayment-behavior signals now matter even when collateral exists. Underwriting has evolved—policy design, collateral lenses, and customer segmentation are more disciplined. The next leg is about stronger models and cleaner data pipelines to cut TAT without compromising asset quality—and that’s where most teams are focused now. 

Underwriting 

Q: When you look at underwriting decisioning, several challenges stand out: policy complexity across borrower groups, data quality (especially in affordable), and regulatory constraints. For example, the push for explainable models rather than black-box AI. What big limitations still persist, and what innovations are helping address them? 

 RAM KISHAN KOLLI : Some constraints are structural, and they’ll stay. Policies differ by institution—risk appetite, how household income is derived, acceptable LTV, and typical ticket sizes. In prime you often see common multipliers, whereas affordable varies by lender and by profile.  

An affordable housing borrower isn’t a single archetype. A kirana owner, a dairy supplier, a carpenter, and a cash-salaried employee, each of these requires a different approach. Most lenders have now built profile-specific playbooks and personal discussion (PD) questionnaires, but true uniformity isn’t realistic.  

Data quality is the other big one. Semi-documented income needs triangulation—bank statements, UPI trails, supplier cycles, and local comparators (what a similar shop in that area usually does). PD still needs trained probing. The aim isn’t full automation, it’s to make PD consistent and auditable.  

On regulation, constraints show up through collateral and end-use rules: LTV caps, how value is recognized (agreement value vs market), and plan-approval requirements shape what qualifies as a home loan. In smaller towns, self-declared plans are common; HFCs typically classify those as LAP rather than home loans. On modeling, decisions must be explainable—opaque, black-box AI isn’t acceptable. 

What’s helping today is a practical stack: codifying profile playbooks into policy-as-code, tightening PD training and rubrics, cleaning up data pipelines, and moving vendor flows (PD/legal/valuation) to API-based uploads. That combination applies policies more consistently while preserving room for on-ground judgment. 

Q: So as a lender, you essentially distinguish between already-constructed homes and those under construction. How do you handle end-use monitoring? 

RAM KISHAN KOLLI :We look at two things: the stage of the property and the end-use of the money. If construction is still going on, we release the loan stage by stage (foundation, slab, finishing) and we pay directly to the builder. That’s a home loan (HL).  

If the customer has already spent on construction and now wants a reimbursement, that portion isn’t HL anymore, it’s LAP under current norms. Earlier, smaller tickets sometimes slipped in as HL, but the expectation now is clear: reimbursements = LAP.  

Any remaining construction (often the last 10–15%) can be financed as HL, again tied to verified milestones. Practically, there are two lanes: past spend is LAP; future spend is HL. We monitor end-use by linking each release to a verified stage and keeping HL disbursals builder-direct rather than crediting the borrower’s account. 

Q: In affordable housing, say with carpenters or kirana owners, you often hear about questionnaire-based score carding. Can this ever be fully digital, or will a credit underwriter still need to go out and do a PD? 

RAM KISHAN KOLLI : Scorecards help, but they don’t replace the PD. Accroding to me, two things matter. First, standardize the questionnaire, by profile. Unlike a prime salaried case where you can just read a payslip/CTC, or a clean P&L for a self-employed borrower, these customers need deeper probing—gross vs. net sales, missed expenses, seasonality, supplier cycles, cash vs. UPI mix. A tighter, profile-wise questionnaire is the starting point. 

Second, who asks, and how they ask matters. The PD officer has to drive the conversation. This bit can’t really be automated unless a lot of data already exists somewhere. So yes, it still needs a physical discussion in many cases. The challenge is the subjectivity; different underwriters probe differently. Standard questions, digital capture of answers, and validating against nearby comparable profiles (same locality/pin code) helps shrink that gap. 

Can it be fully digital? Not right now. Where data exists (bank statements, UPI trails, GST), you can pre-fill, record the PD, run calculators, and give nudges/prompts to the officer. But a human conversation and probing will stay important because the profiles are so varied. So, the practical model is hybrid. Standardized digital questionnaires along with human PD, with explainable scoring layered on top. 

Digital maturity of the housing finance ecosystem  

Q: In housing finance, a lot has digitized quickly, but the product still feels less mature than unsecured retail. What are the real sticking points today? 

RAM KISHAN KOLLI : 
The first mile is fairly strong, bureaus are well integrated, and most lenders pull both CIBIL and CRIF because coverage differs. But not everything sits neatly in one place. 

The big gap is income assessment for semi-documented or undocumented borrowers. That still hinges on personal discussion, where a trained officer runs a structured conversation and uses local context to judge cash flows. Much of the capture remains manual and will likely stay hybrid. We do use surrogates where available. Say, a shopkeeper with ~50% UPI receipts, to triangulate, but it isn’t universal, so you can’t standardize end-to-end. You can record PD inputs and run calculators, yes, but full automation beyond that is hard. 

Once PD is in, the digital flow picks up again. Profile-specific journeys and rules in the LOS help, because a carpenter doesn’t look like a kirana owner in a high-footfall junction with ₹1 lakh inventory, which again doesn’t look like a similar shop on an inner road. Pre-visit comparators (past cases in that pin code, typical daily sales/expenses) make the underwriter’s assessment more consistent and reduce overestimation or fraud risk. After you’ve got a clean view of household expenses vs. business income, it’s straightforward to underwrite, just align to policy. 

On collateral and legal, prop-tech tools and some state digitization help with valuation benchmarks and document pulls, and vendors increasingly integrate via API. But online data isn’t perfect. You still need sub-registrar searches, local office checks, and physical visits, especially in Tier-3/4. The good news is that this pain should ease as more institutions participate and more on-ground data flows into systems. 

Q: Within the value chain, some parts need manual probing, and others rely on third-party vendors for valuation and legal. Is that data reliable enough yet to tell fraud control units, “we’ve got this via API, no visit needed”? Are we there yet? 

RAM KISHAN KOLLI : 
No not really, not end-to-end, yet. We’re getting closer, but we’re not at “pull this through API and it’ll be done.” 

On PDs, a few vendors are scaling nationally, but most are still regional, so we usually empanel multiple vendors. The upside is better uniformity (common tags, formats), less branch manpower, and a basic QC layer before files hit our desk. But delivery today is mostly portal uploads, rather than clean, universal APIs. 

On technical/legal, it’s further along. The same vendor often serves 15–20 lenders in a district, so they bring cross-case intelligence (“this property was rejected last month for title issues”). They still do site visits and sub-registrar/local searches, then upload standardized reports. Again, typically uploads, not one API everyone plugs into. 

Integration-wise, unless you work with an aggregator that normalizes vendor formats and pushes reports straight into your LOS, you won’t get deep API integration across the board. Most of us still juggle multi-vendor panels and asynchronous reports. Having said that, aggregators are emerging now, and you can plug in APIs into them, but as a market we’re not API-only yet. There’s still a place for FCUs/field checks in the loop. 

Q: Now, at a housing discussion the other day, I remember someone asking, “what would it take for a ₹5 crore home loan to be fully digital?” For a prime customer, you’d expect valuation, builder integration, and employment/financials to be clean. Is that realistic, or still far off? 

RAM KISHAN KOLLI : 
It’s realistic for clean salaried profiles. We already do ₹1 crore salaried cases smoothly; scaling that isn’t hard. It’s also doable for well-documented self-employed borrowers where GST/ITR trails are complete. 

Where it gets tricky is multi-income files (rental income, trading income, spouse income etc.) spread across different bank accounts and formats. You need triangulation and some surrogate methods, and the data rarely arrives in one neat package. The issue isn’t the calculation, it’s aggregation. Customers have to pull statements, GST returns, rent agreements, etc., from multiple sources, and that impacts the TAT. 

Q: But from what I understand, some parts of the process still feel manual. Think occupancy/completion certificates, and the whole “papers-in-mortgage” step. Are those solved digitally yet, or do they still need boots on the ground?  

RAM KISHAN KOLLI : It’s a bit of both. For instance, e-signing is now sorted for the loan docs. In some states, parts of the mortgage process, like Notice of Intimation (NoI) in Maharashtra, can be done online. For builder/RERA-registered projects, most title documents are available at a click, and you can even run basic analytics on litigation or status. 

But then, a registered mortgage still needs a physical visit to the registrar in most states. For resale properties, you still do sub-registrar and local office searches, and there’s variation by state in how closely online data matches ground reality. Then there are NOCs (society/authority), share-certificate updates, and a few other local nuances. These are typically offline. You also want a field check to verify that the documents you received actually came from the issuing authority; there isn’t always an online registry to confirm authenticity. 

  A big chunk of signing and intimation is digital, but the perfection steps such as registrar touchpoints, NOCs, and authenticity checks, remain physical, especially in resale and Tier-3/Tier-4 contexts. 

Fraud in housing finance  

Q: You mentioned fraud risk earlier; can we double-click on that? What kinds of fraud do you actually see in home loans, especially in affordable housing? 

A: In prime, fraud is relatively limited and the data trails usually catch it, unless you run into a large, coordinated forgery (like syndicates faking salary slips). Those are outliers. In affordable, a few patterns show up more often. One is the classic “setup”: the applicant claims to run the shop, but it’s actually the brother or someone else. With common surnames and familiar local networks, it’s easy to miss unless you ask the right reference questions and do on-ground cross-checks. 

Another is buyer–seller collusion with a small builder or an individual seller to raise short-term cash. A sale deed gets executed and loan money moves, but there’s no real intent to transfer possession; the family may continue living there. If you probe intent to move, current housing situation, and timing, the story often doesn’t add up. 

On income papers, outright forgery is less common because PDs are conversation heavy. The bigger risk is on the property side in Tier-3 and Tier-4 markets: laminated or high-quality colour copies presented as originals, or cases where the true originals are already deposited with private lenders. If you skip last-mile verification, you can end up holding a convincing copy while the real title is pledged elsewhere. That’s why registrar checks, society/authority verification, and field confirmation still matter—even as more of the journey goes digital. 

Q: So it sounds like multiple funding is a real risk, right? 

RAM KISHAN KOLLI : It can be. One common edge case is the timing gap between a sale/registration and when that change actually shows up in sub-registrar searches. Say a flat is sold this morning, but another lender, working off a file logged a month ago, disburses tomorrow. If the lender skips a last-day search, they might miss that the property was already sold before their mortgage was perfected. 

Another variant is two institutions underwriting the same borrower/property in parallel; one disburses first, the other follows without catching it. This is rarer now because recent CIBIL enquiries are visible and teams do sanity checks, but purely local transactions that don’t surface immediately can still slip through and only show up later—sometimes at default.   

Q : And on CERSAI, how much does it help, and where does it fall short? 

 RAM KISHAN KOLLI : CERSAI is the central registry where lenders file mortgage details. You upload the charge with customer identifiers (name, PAN) and property details. When we run a CERSAI check, it’s pretty effective at answering, “Does this customer already have a registered mortgage?” The weaker spot is property-level matching. Indian addresses aren’t standardized—slashes vs hyphens, missing characters, local formatting—so the same address can be written three different ways. That means a straight match can miss, not because the data isn’t there, but because the string doesn’t line up. There’s also room for human error at entry. 

There are two ways to work around this. First, triangulate, don’t rely on any single source. Combine CERSAI, sub-registrar searches, society/authority checks, and recent bureau enquiries. Second, there’s real upside in address-normalization and fuzzy-match models on top of CERSAI: learn local quirks (where slashes become hyphens, which elements get dropped), score probable matches, and flag them for a credit manager to review. You won’t make it 100% automated, but you can cut the miss rate and push more of the edge cases into a review queue before money goes out. 

 Industry outlook for home loans/LAP 

Q: Got it. Another thing I’ve been hearing is that a lot of housing finance companies are leaning more into LAP with direct sourcing. Is that because of macro tailwinds in India, or mainly because the RBI is cooling sentiment on unsecured loans?   

RAM KISHAN KOLLI : I think it’s macro, not just regulation. India’s mortgage penetration is still around 12%, far below mature markets. Post-COVID real-estate growth started in Tier-1/2 and is now spilling into Tier-3 and Tier-4. Incomes are rising through 2030, nuclear families are growing, and people upgrade homes roughly every 7 years (closer to 10 years in smaller towns, but that’s shortening). Affordable is growing faster than prime, and there’s still a lot of white space: many lenders remain regional; even national players active in 13–14 states often have only 50–60% depth. More states are now following the early movers, so the runway is big. 

So yes, LAP along with home loans via direct sourcing can create real value for both the institution and the customer. It’s hard work, every branch is a new learning curve, but that’s exactly why the opportunity has been left on the table. This isn’t about RBI said slow unsecured. Secured lending itself is a large, underserved journey we’re choosing to build. 

Q: Pulling this back to the customer experience, everyone wants growth (lenders, government, borrowers), but it ultimately lives or dies on the journey. Across the sector, where do you see the biggest drop-offs, and how do you solve for them? 

 RAM KISHAN KOLLI : The first drop-off in affordable housing is really about clarity. Customers often don’t know if their collateral fits a lender’s criteria; dimensions don’t match the papers, approvals are missing, or the title chain isn’t what a lender needs. Given the subjectivity and triangulation in this segment, they can be left hanging for weeks. A clean yes/no early, something like “fix legal/technical/credit and come back” is actually good service and saves everyone time. 

Next, there are bureau issues. We check both CIBIL and CRIF. A fresh delinquency or a lingering remark can stall the file, and many customers can’t immediately produce a No Dues Certificate (especially for small mobile/fintech loans). If we can’t verify closure, the case drops, not because we don’t want it, but because we can’t evidence it. 

Then there’s valuation. In Tier-3 and Tier-4 markets, perceived value is often set by one high offline deal that becomes the ‘village benchmark.’ Our vendors may come in lower, which cuts eligibility and triggers a drop. 

Legal is another big bucket. Ancestral properties without proper mutation, revenue records that don’t match deeds, notarial transfers without stamp duty, and so on. Some lenders will take a collateral call or a profile call, but taking both risks on the same file is rare, so appetite drives outcomes and drop-offs vary by institution. 

 A lot also comes down to how you filter leads. If the front-end gating is weak, junk flows in and then dies later, first at in-principle credit (eligibility), then at technical (valuation), then at legal. And after all that, you still see customer back-outs: they want a higher ticket than eligibility allows, they’ve lost interest, or they’ve got a better offer from another lender. 

Put together, you see why login-to-disbursal in affordable tends to land around 35–40% in well-run, mature branches. New branches swing more until teams, vendors, and local processes align. The fixes are mostly about sequencing, do the right probing on day one, set expectations, clear bureau mismatches quickly, and don’t let valuation/legal checks start late. That’s how you protect the customer’s time and your TAT. 

Q: Speaking of lead quality and borrower pools, there’s a lot of buzz around partnership lending and embedding into places where people are already searching for homes. Do you see that working for housing finance? 

RAM KISHAN KOLLI : It helps at the top of the funnel, but conversions are modest. Property portals and aggregators do a good job of capturing intent and passing home-loan leads, but once you start nurturing, conversion rates are merely 5–10%. 

There are two reasons behind this. First, prime and semi-prime borrowers already have a bank preference and are very rate sensitive. Second, builders run their own channels, CP (channel partner/broker) and DSA, so a lot of serious buyers are pre-tagged to lender panels before they ever hit an online portal. Data flows both ways between CP and DSA networks, so by the time a lead reaches you digitally, it’s often been worked. 

Even when a lead is ‘open,’ you hit practical hurdles. The property may not be finalized yet, or the location might sit outside your branch coverage. Those drop-offs show up early. And once you do move ahead, the physical journey (PD, legal, valuation) introduces the usual friction points, so you see another wave of attrition there. In short: great for discovery, but you still need strong qualification, coverage, and on-ground execution to make partnership leads pay. 

 Q: So, is the digital market for housing loans more likely to build around digital balance transfers (BT)? A lot of borrowers rate-shop, so I think that feels like a natural wedge. 

RAM KISHAN KOLLI : Yes, especially in the prime segment. Some partners already sit on rich signals, full bank/account data via Account Aggregators (AA), card spend patterns, even basic vehicle ownership data, plus the customer’s existing home-loan details and current rate. If you combine clean repayment behavior with what you know about inflows and obligations, you can automate BT offers well - pre-approved, limited-ticket, straight-through flows. That’s where a digital play can really work. 

But the core still has to be direct sourcing. There’s a lot of unmet demand, and in affordable you want to own the customer relationship end-to-end. Digital BT and aggregator-led plays are good adjuncts. Lenders use them for inorganic growth or selective buys but they won’t replace a strong direct engine. Over time, data-driven nurturing will take a bigger share, but today, if you want a resilient mortgage/LAP business, you lead with direct, and layer partnership and BT motions on top.