AdalFi: Powering credit infrastructure across emerging markets

Hi friends! 👋

In Pakistan, barriers to credit run deep: blank credit reports, broken underwriting, and a legacy banking system that simply doesn’t scale to the needs of real consumers or SMEs.

That’s the gap Salman Akhtar set out to close – not as a lender, but as the builder of a new kind of credit infrastructure.

Drawing on decades of experience implementing core banking systems at Techlogix, Salman knew the problem wasn’t capital – it was the inability to assess and execute lending decisions at scale. So he launched AdalFi, a digital credit platform that transforms how banks score, underwrite, and disburse loans – using only the data they already have.

No paperwork. No CAC. Just real-time lending to verified customers, scored on the floor of their financial capacity – not the ceiling.

Today, AdalFi is powering over 400,000 loans across 14 banks with a blended NPL rate of just 0.2% – compared to a market average of 6%. And as open banking looms and Gulf banks look beyond salaried borrowers, AdalFi is positioning itself as the federated intelligence layer that can scale smarter credit – without ever becoming a lender itself.

In this conversation, Salman and I unpack what it takes to digitise lending inside legacy banks without triggering resistance, how to underwrite customers with no formal credit history, and why AdalFi’s refusal to lend may be its biggest strategic moat.

We also dive into how you design models that get smarter with scale, why seasonality breaks traditional credit products, and what makes the Gulf such fertile ground for infrastructure-led expansion.

Here’s what we covered:

• 💳 Why lending to the floor of someone’s financial capacity is safer than chasing the ceiling

• 🔁 How AdalFi built a zero-CAC credit model by embedding into liability-side bank infra

• 📊 Why most “simple” credit signals (like balance trends) are statistically fragile – and how to fix them

• 🧠 How federated learning across banks creates compounding model defensibility

• 🚫 Why not being a lender is the most scalable lending strategy of all

Let’s get into it 👇

Actionable insights 🧠 🛠️

If you're short on time, here's what the smartest operators, investors, and founders should understand from Salman Akhtar's AdalFi playbook.

• Underwrite to the floor, not the ceiling. AdalFi’s risk model doesn’t try to impute full income – it lends against the verified floor of a borrower’s financial capacity by using just one bank account. That single-account constraint (which many would view as a limitation) is actually a feature: it conservatively anchors credit decisions to what is at least visible, rather than guessing what might exist. If the borrower holds more funds elsewhere, that’s upside — not exposure. In frontier markets with limited data sharing, this is an elegant way to price risk more safely.

• Mathematically rigorous ≠ overly complex — it just means you know where the edge cases live. AdalFi doesn’t use rule-of-thumb shortcuts. They run mathematically robust trend analysis on balance data (e.g. detecting whether account balances are rising or falling over time) – something many banks get wildly wrong because they use simple averages. The practical implication: if you’re scoring based on balance trends without accounting for volatility sensitivity and boundary skew, your model isn’t just noisy – it’s actively misleading.

• Risk-aligned monetisation > flat SaaS fees. The unlock with banks wasn’t software – it was trust, proven by a fully risk-aligned rev share. No license fees. No integration fees. No annual charges. They only get paid when the loan performs. That flips the procurement cycle on its head – sales teams don't need to fight for budget sign-off, and risk teams see genuine skin in the game. Founders building fintech infra: consider how aligning incentives can collapse sales cycles, especially in legacy institutions.

• Distributed learning architecture is a stealth moat. AdalFi’s on-prem model means banks don’t share raw data – but anonymised outcomes and scoring variables are fed back centrally. The result is a federated learning effect: each deployment becomes a sensor node for system-wide improvements. Over time, this makes the entire network smarter – while still preserving bank-level control and privacy. It’s a defensibility layer disguised as compliance accommodation.

• Seasonality is an underwriting blind spot – and a product design failure
  AdalFi’s next evolution isn’t scoring better – it’s building more flexible credit products. Traditional EMI schedules break for seasonal SMEs (e.g. ice cream shops doing 90% of sales in 4 months). By shifting repayment burdens back onto product design (rather than the borrower’s discipline), they’re targeting an entirely excluded segment. If you’re building SME finance infra, ignore this at your peril – or you’ll miss half the TAM.

• Open banking isn’t plug-and-play. You need a model that knows what to do with the data. Open banking doesn’t inherently enable credit access – it just improves data liquidity. Without a risk model capable of parsing transaction-level behaviour, the data is inert. AdalFi has a pre-baked scoring system ready to ingest OB data – meaning that the moment it’s live, banks can lend to non-customers with confidence. In regions racing toward open banking regulation, whoever has the ready-to-consume model will win.

• Embedded credit infra is a wedge – not just for users, but for regulators. AdalFi embeds itself inside the bank’s infra – not as an overlay or app, but by integrating directly with core systems. This creates both switching friction and institutional comfort: it’s not an add-on, it’s part of the general ledger. It also gives them transparency into the full lending decisioning stack – letting them track decisions and outcomes end-to-end, which builds a tighter feedback loop for model evolution. For regulators, that means auditability. For banks, it means trust.

• Don’t be the lender – be the infra. Especially in capital-scarce, CAC-heavy markets. AdalFi could lend. But they’ve actively chosen not to. Why? Because CAC kills fintech lenders. By staying behind the scenes, they avoid acquisition costs entirely – instead, they extract value by activating already-onboarded customers inside the bank. The result is a zero-CAC, no-balance-sheet model with infinite TAM and none of the margin compression. In markets like Pakistan or even Saudi, this is how you play the long game.

Okay, let’s dig into it 👇

This article is part of FWDstart’s partner programme series. We follow stringent editorial standards, and while we collaborate closely with our partners, final editorial control rests solely with FWDstart to ensure the authenticity and creativity of our content. All partnerships are disclosed clearly, and we only collaborate with organisations we genuinely admire – AdalFi is one of them.

You’ve said, “Lending in Pakistan is fundamentally broken.” What was the aha moment when you realised the issue wasn’t demand – but banks’ credit models?

So, some background first – I’m also the CEO of a services company called Techlogix, which I started around 30 years ago. One of the things we do at Techlogix is implement Oracle’s banking solutions – we’re an Oracle partner. That means we work on everything from core banking to compliance, digital banking, and risk management. So we’re very familiar with how banks operate and how their systems are structured – we’ve been doing this for over 20 years.

One of the things we observed early on was the stark contrast between what was happening on the asset side versus the liability side. You might have five million customers and a $20 billion deposit base, backed by a $15 billion asset book. But if you look closely at the asset composition, it’s almost entirely made up of large corporates, a few medium-sized ones, and government treasuries. The consumer lending space was – and still is – tiny.

Just to give you some context: in Pakistan, the consumer lending-to-GDP ratio is about 4%. In Egypt, it’s around 15%. The U.S. is obviously a much more advanced market. So that gives you a sense of how underdeveloped consumer lending is here.

Now, in Pakistan, consumer lending offers spreads of around 1,000 basis points. That’s a huge margin – there’s real money to be made. And while NPLs (non-performing loans) on consumer portfolios are in the range of 5-6%, that’s pretty standard for this region. About a decade ago, there was a spike in consumer defaults, and many bankers still point to that as a cautionary tale – but overall, it’s manageable.

So the natural question becomes: what’s stopping banks from lending more to consumers?

The short answer is that they don’t have effective credit scoring models. Pakistan does have credit reporting—there’s a central bank-managed system, and banks are actually required to check it to see a person’s outstanding debt and repayment history. But scoring? That’s missing.

And because so little consumer lending has been done historically, most of those credit reports are essentially blank. So banks resort to manual verification—sending someone to check where you live, verifying employment history, that sort of thing.

But that comes with two major problems. First, it’s expensive. Second—and this was backed up by surveys from the Gates Foundation and the World Bank—the biggest reason people don’t access formal credit is the time it takes to get a decision. Banks take two weeks, sometimes longer. But when people need a loan, they need it tomorrow or the day after—not two weeks from now. And often, they don’t even get an answer—they’re just left waiting until they’re eventually denied.

And then there’s all the friction: paper forms, physical signatures, branch visits. Every step is inefficient and broken. What we saw was that for things like auto loans—where the ticket size is big and the purchase is planned—people were willing to tolerate the bank’s processes. They didn’t like it, but they dealt with it.

But for personal loans, where the volumes could be huge, none of this scaled.

What were the first lending use cases you tackled – and how did they inform your current approach to both consumer and SME lending?

Let me take a step back. From the beginning, we saw that the core problem had two dimensions. First, the bank’s ability to take a credit decision. Second, the ability to execute on that decision – meaning, to actually disburse the loan in a highly efficient, digitised way. And both of those were broken.

So we started working on both fronts at the same time. On one side, we focused on building digital journeys that allowed banks to disburse loans in real time. Of course, there were regulatory and technical challenges, but we were uniquely positioned to tackle those because of our experience as service providers to these banks. We knew their systems intimately and had the ability to automate those processes.

The good news was that many of these banks were already running relatively modern core banking systems – Oracle FLEXCUBE, Temenos, and so on. The infrastructure was there – it just hadn’t been put to use in this way. So we built on that.

The second part was figuring out how to score customers who had never been scored before. That was the insight: if someone has never taken a loan, you can’t assess their propensity to repay – there’s no data to build that model.

So instead, we shifted the focus: let’s measure their ability to repay. Do they have the actual financial capacity to pay back the loan? Because that comes before propensity.

And the best way to assess that was through the information the bank already had – what’s in the customer’s bank account.

So we took two years’ worth of transaction history, looked for patterns and trends, and used that to build a scoring model for these customers who couldn’t be assessed using traditional methods.

Then we combined both elements:

• A scoring model built on internal bank data

• Digital journeys that enabled a real-time, straight-through lending process

The beauty of this approach is that it allowed us to start by lending to the bank’s existing liability-side customers – people who already had accounts and were KYC’d.

The result? You could disburse a loan in real time. No paperwork, no manual checks – the money just went straight into their account. And it worked beautifully.

Banks are notoriously slow to adopt new tech. You’ve onboarded 14 banks, with 7 actively lending. What was the unlock?

Honestly, if we’d just been a brand-new startup, we wouldn’t have gotten past the front door. Banks have their reasons – and they’re valid. But our legacy as long-term partners made a big difference.

The second unlock was a very deliberate business decision we made early on: we said we’d do a risk share. And not only that – we’d do it with a pure revenue share model.

That meant:

• No license fee

• No annual maintenance charge

• No implementation fee

• Zero upfront cost

We only make money when the bank makes money. If you disburse a loan and the money comes back – that’s when we get paid. Our interests are perfectly aligned.

That made it much easier for the business teams at the banks to say yes. They didn’t need budget approvals because it was essentially free. No capital expenditure – just upside.

And part of the story we shared with them was: “You already have these customers. You don’t need to spend big on marketing. You know who they are. They’re already onboarded. Just message them directly.” That really resonated – it was smart, targeted growth without big marketing costs.

Now, the one team that required a bit more work was the risk team. That was the toughest part.

What I had to do – which wasn’t easy – was convince the first three, four, five Chief Risk Officers that their bank would be authorising loans without a vetting signature. No physical signature. No manual review. It would be a fully automated, programmatic process.

Some risk teams were open to that. They’d say, “We’ll take your model as-is.” And we were very open – we’d say, “If you have views on the model’s weightings or logic, let’s talk. We’re taking risk with you – we’re aligned.”

Others took the opposite approach. They’d say, “It’s your model. You own it. If it blows up, that’s on you. We won’t make any changes.”

So we got both ends of the spectrum. And we’re fine with that – we love all of them. We work with them every day. That’s just the game. You learn to play it.

Your SME model only uses digitally verifiable data. How do you define data integrity – and how do you ensure it’s robust across very different partners, from banks to POS providers?

In my view, there are three tests that any data must pass before it can be used in a scoring model.

First, the data has to be third-party collected – it can’t be self-reported. There are a lot of reasons for that, but the most important is that once people know you’re using self-reported data to make lending decisions, it starts to corrupt the data. So, third-party is essential.

Second, the data must be digitally collected – that’s fairly obvious, but worth stating.

And third, it has to be directly relevant to the lending decision – financially meaningful.

Let me give you a few examples to illustrate.

Take bank transactional data. That clearly passes all three tests. It’s third-party collected – by the bank. It’s digitally collected – it’s sitting in a core banking system. And it’s directly relevant – it reflects actual inflows and outflows, so you’re seeing how much money a person has and how they manage it.

Now let’s look at an example where the tests begin to fail – telco data.

Yes, it’s third-party – the telco collects it. Yes, it’s digitally collected – through their billing system. But when it comes to financial relevance, it starts to break down. In most countries – including Pakistan – there’s a very narrow dynamic range in telco billing data. What I mean by that is: I can’t really spend more than about $10 a month on my phone. No matter how much data I use, the ceiling is quite low. At the same time, I probably won’t spend less than $1. So, the range is about $1 to $10.

Compare that to income levels in Pakistan – where the range is 5x or more. Telco data just doesn’t carry enough financial resolution to help you underwrite medium or large loans. It’s useful for nano lending – very small-ticket loans – where that narrow billing data matches the income bracket. But you can’t use it to give out a car loan. It doesn’t make sense.

Let me give you a third example – one that actually works: weekly retail sales data from corner stores. For instance, we have data from a partner like Pepsi, showing weekly sales from mom-and-pop retailers going back five years.

That data checks all three boxes. It’s third-party – Pepsi collects it. It’s digital. And it’s financially relevant – it directly reflects the revenue those stores are generating, which is a strong signal of their ability to repay.

So that’s the lens we use. The data must pass those three tests. Some datasets will work for certain financial products – like telco data for nano loans – but won’t work for others.

Your blended NPL rate is 0.2%, compared to a market average of 6%. That’s a staggering delta. What specifically enables that? Is it your model design, the data inputs, or how you update it in real time?

I think there are three elements going on here. And I should add – the lending vintage we have now is coming up on four years. Our SME portfolio vintage is about two and a half years. That’s plenty of time to start seeing defaults in the space. Obviously, when you start lending, your NPL ratio looks great because nobody defaults in the first month – so it’s zero. But over time, you start to see real patterns.

So, the first aspect is the model-driven approach, which is really about trying to understand what’s happening in your core financial strength as reflected in your bank account.

Now, one thing to note – because we’re only looking at your account with the bank we’re working with, we’re not seeing your entire financial life. You might have other bank accounts we can’t access. But that actually creates a benefit: we’re lending against the floor of your financial capacity – the minimum. You at least have this much money. If you have more money elsewhere, that’s great, but it doesn’t negatively affect our view.

At the same time, we do have full visibility on your liabilities across the system, because we can see all your loans and indebtedness through the central credit bureau. So we’re working with a base of asset data and a comprehensive view of liabilities – which, on balance, is a good setup.

It’s important that we’re not lending to some hypothetical or imputed ceiling. We’re lending to a verified floor. That’s a more conservative and safer approach.

The second part is the rigour of the model itself – how we interpret what’s going on in the account. I’ll give you an example. Suppose you want to use bank data to answer a simple question: is this person’s balance going up, staying the same, or going down over time?

Intuitively, you’d think: going up is good, going down is bad. Seems obvious, right?

But it turns out to be an incredibly hard question to answer. When you look at end-of-day balances, the data is all over the place – boom, up and down. You can’t just draw a neat trendline through it.

And here’s the kicker: trendlines are very sensitive to the starting and ending points. Shift those slightly, and suddenly the trend says the balance is increasing, or decreasing – depending on where you begin and end. So the conclusions become fragile.

That’s why we use sophisticated mathematical techniques just to answer that one seemingly straightforward question. Because while the data looks simple, to use it in a statistically valid, robust way – that’s not easy.

I know a lot of banks use very simple techniques. Some just use rules of thumb – like taking the ten biggest credits, dropping the largest and smallest, and averaging the rest to get some indicative figure.

What we do is far more robust – because the signals are there, but they’re not surface-level. You have to excavate them mathematically. That’s the real challenge.

And, the third point I want to make is about the totality of what’s going on in the system. By that, I mean two specific things.

First – because we start with the bank’s existing customers, liability-side customers – we’re able to do something important: we design the loan to auto-liquidate. So, when the repayment is due, we deduct it directly from the customer’s account.

That gives you automatic repayment capability. If the money isn’t there at the time, we place a track on the account, so the next time a credit comes in, it repays the loan before the customer can use those funds for anything else.

And the reason we can do that confidently is that we know this is a frequently used account. If it weren’t, there wouldn’t be enough transaction history for us to score the customer in the first place.

In fact, roughly half the accounts banks send us – we reject. We tell them: there’s not enough data here. The account might be technically active, but it’s not meaningful. It’s not that customer’s main account. We just throw it away.

And most people have one main account they use – their salary comes in, they pay bills from it, etc. That’s the account we’re scoring. So not only are we getting an accurate view of their financial behavior – we also know this is the account most likely to have funds in it.

Second – because we operate on a risk share model, we’re not just giving a score and walking away. All of the underwriting rules run through our system. So we have full transparency.

Compare that to a traditional model – say with FICO. FICO gives you a score, but they have no idea whether the bank actually disbursed a loan, how much was lent, or what other variables went into that decision. They’re blind to all of that.

In our case, all that logic – the scoring, the underwriting rules, the decisioning – runs through our system. So if something goes wrong, we have a full view of what happened. That gives us a feedback loop that lets us improve the model continuously.

And because we have skin in the game, we’re highly incentivized to evolve the models over time. Banks are very comfortable with that because we’re aligned – we don’t make money unless the loans are performing.

For example, both on the consumer and SME sides, we’re now running third-generation models. And we’ll keep evolving them. That’s in everyone’s interest – we just want to make sure the loans go to the right people, and nobody loses money. Same as the bank.

And just to touch on data network effects – to what extent do those tie in here? Does every new partner make the model smarter? Or do you have to be cautious about overfitting?

That’s a great question. In all of our deployments, the models run on-premise at the customer’s site. That was important from day one, because banks had understandable concerns about data privacy. So we said, “No problem – everything runs on your servers. You’re not sharing any customer data with us.”

The only thing we ask is this: once you disburse a loan, share the repayment history and the scoring parameters with us – on an anonymized basis. That’s the data we use to learn and improve.

Now, all of our partner banks have agreed to this setup. So yes – there’s absolutely a federation effect happening. We’re learning from every single bank, not just one. And when we evolve the models based on those learnings, we update them for everyone.

It’s a win–win. Everyone benefits from a better model.

Of course, you could say the smaller banks get a bit of a smoother ride – they’re benefiting from improvements driven by the larger banks’ data.

But honestly, that’s just like an ATM network – smaller banks get a free ride on the bigger network. And hey, the big boys can afford it, so it’s okay.

You’ve disbursed over 400,000 loans – with half of those just in 2024 – and enabled $44M in lending this year alone. That’s immense acceleration. How are you maintaining precision as you scale? Have there been stress tests or near-misses that changed how your systems evolve?

So as loan volumes grow over time – and more loans go out – the model is actually becoming more accurate. Our NPL rate remains at 0.2%, but now that 0.2% is being held across 400,000 loans, not just 20,000. That’s a much stronger signal – we’re working with richer data and more patterns to learn from.

The second thing is that every bank focuses on slightly different segments – different customer profiles, different geographies. So each new bank we partner with effectively becomes another sensor for the system. We get visibility into new behaviour clusters and can learn from each one.

Here’s an example. In one province in Pakistan, we noticed a localised hotspot of willful defaults. People had figured out they could get away with it – and that behaviour started to spread. It became clear there was a kind of social contagion happening.

We learned about that through one banking partner – but once we saw the pattern, we fed that insight back into all of our models across all of our partners. Because we’re in a risk-share model, we’re highly motivated to catch these things and communicate them early.

What’s really interesting is how the conversations with risk teams and business teams evolve. They’ll ask us, “What are you seeing at other banks?” – because they know we have that visibility. And that’s a conversation they can’t easily have with each other. A CRO can’t just pick up the phone and ask another bank, “Where are your problem areas?” No one’s going to answer that honestly – everyone says, “Everything’s fine.”

So we’ve become this kind of neutral partner – the common thread that everyone can speak to. We share what we’re seeing (in aggregate and anonymised, of course), and that helps every bank we work with.

So as we scale, we’re not just increasing volume – we’re deepening and broadening our capability set in a meaningful way.

You’ve said the model is GCC-ready. What structural differences between Pakistan and the Gulf (e.g. digital banking maturity, regulatory comfort, data availability) affect how you approach GTM in the region?

First, it's worth saying that the GCC isn’t monolithic – there are real differences across markets. But if we talk in broad strokes…

GCC banks have done a great job lending to salaried consumers. The UAE, in particular, has done this phenomenally well – they’ve really shaken that tree. Saudi has also done a solid job. So that segment is well addressed.

That said, lending remains a challenge everywhere – even in the U.S. So, in a sense, you could land blindfolded in almost any country, walk into a bank, and say, “I want to help you with lending,” and they’d probably say, “Yes, we could use some help.” There’s always room to improve credit delivery.

But then there’s this other, netherworld – individuals who aren’t salaried, and who don’t quite fall into the SME bucket either. Say I run a hair salon. I bank in my personal name – so I’m not categorised as an SME – but my income is tied to that business.

The bank knows I’m not salaried. In places like Saudi or the UAE, they can verify that through national payroll databases or other checks – and once they know I don’t earn a salary, they often don’t want to lend to me. Even though I have consistent income, I’m outside the box.

That’s a huge opportunity. One of the things we do really well is look into those transaction flows and find patterns – uncovering viable borrowers in spaces that traditional scoring misses.

So in the GCC, we see two high-potential areas:

1. The SME market – which is still significantly underserved.
   

2. The adjacent “individual operator” segment – people who aren’t salaried and aren’t labeled SMEs, but whose bank accounts clearly reflect real business activity.

Let’s talk about the buzzword that’s got everyone excited right now – especially here in the region – and rightly so: open banking. You’re speaking at the Dubai Fintech Summit next week on a panel about open banking as a catalyst for inclusion. I’d love to hear your thesis – how do you see open banking reshaping lending in the region?

I think open banking will be a phenomenal driver of SME lending – if you’re set up to use it properly.

But here’s the key – you need to have the kind of models that can actually consume that data and make lending decisions from it. Just turning on open banking doesn’t mean much on its own.

Sure, now I can pull your data – but that doesn’t mean I know how to score it. If I haven’t been scoring data that way already, it’s meaningless.

To give you an example: if open banking went live in Pakistan tomorrow, our partner banks could instantly start lending to every banked person in the country, as if they were their own customers. That’s huge.

Right now, there’s a lot of friction – if you’re not a customer, we need your bank statement, maybe a PDF, maybe a file upload... it’s slow, messy, full of drop-off. Open banking eliminates all of that – it’s real-time, rock solid, seamless.

But again, the benefit only comes if you’re ready to use that data. If you’re still relying on a basic credit score from the central bank, and you don’t have the infrastructure to read and interpret transaction-level data, then open banking adds nothing. It’s just noise.

You’ve made a conscious decision so far not to be a lender yourselves. Is that a permanent stance? Are there edge cases where you’d consider going direct? 

We’ve had this discussion extensively at the board level. And I’m extremely clear in my mind: we should never be a lender. Not now. Not ever.

Here’s why.

First – if I wanted to be a lender, how big a balance sheet could I realistically build? Let’s say $500 million. That would be massive for a fintech. But compare that to any bank – even a small one – and it’s tiny. I mean, $100 million is nothing. Even in Pakistan. In the GCC? It’s a rounding error.

And because my balance sheet is small, I’d be structurally forced to go short and small – short tenors, small ticket sizes – just so I could churn and manage risk. But the economics of lending work the other way: you make money by going big and long. That’s where the yield is.

You can run the math – it’s not a marginal difference. It’s orders of magnitude.

So what ends up happening is that you see these fintechs boasting: “We’ve lent $500 million!” But in reality, they’ve lent $10 million for one week at a time, over and over again. Meanwhile, a bank lends $20 million to one guy for a year – and makes twice the money the fintech did. So what’s all the hoopla about?

That’s reason one.

The second reason is this: the biggest hurdle for most fintechs is customer acquisition cost. CAC is the rock on which so many fintechs flounder – and literally die.

In our model – where we partner with banks – our CAC is literally zero. The customer is already there. We don’t need to acquire anyone. We just need to uncover the right customers from within the bank’s existing base.

And our algorithms do that – at scale. Doesn’t matter if it’s 100,000 or 10 million customers. The algos sort it out. We find the right customer. The bank lends. CAC: zero.

If you become a lender, that entire dynamic shifts. Now you’re in an extraordinarily expensive business.

I’d love to touch on product expansion and building for specific market needs. For instance, in Pakistan, you’ve got salary advances. But as you expand into the GCC, do you lean more toward building general-purpose infrastructure that scales, or tailor products more locally? 

You can look at this in a few different ways. From a credit product perspective, things are actually quite standardised. Across most markets, commercial banks tend to offer the same basic set of products: personal loans, credit cards, auto loans, sometimes salary advances. On the SME side, you’ve got working capital loans – sometimes structured as revolving lines of credit, running finance, or just simple term loans on EMIs. There are only five or six broad categories.

From a system standpoint – we already support all of those. The infrastructure is in place.

Technically, what matters next is integrating with the bank’s core system – because that’s where the loan actually sits, where the general ledger lives, and where repayments and accounting are managed. So on that front, we’re covered. We can originate, structure, and disburse directly into the bank’s core system.

Now, what I’m really excited about is how we’re thinking about product evolution. We recently brought on board a Chief Product Officer who’s based in Jordan. Until recently, he was the CEO of an SME-focused lending platform that operated across Jordan, Iraq, Lebanon, Palestine – and even Romania – with a $300 million balance sheet. A solid, real-world lender.

What excited me about bringing him on was the opportunity to go beyond standard products. Given that we already do some interesting things on the scoring side, I wanted us to explore how we could build more innovative credit products – especially ones that align better with the business cycles of SMEs.

Because look – consumers are fairly predictable. Their financial needs are stable, often spread over 12–24 months, and you can model their income and repayment capacity in a fairly consistent way.

But SMEs are a different story. You’ve got businesses that operate entirely seasonally – they might do 90% of their business in four months of the year.

That kind of seasonality is nearly impossible to accommodate with a traditional EMI model. Right now, most banks just hand the burden over to the borrower. They say: “Here’s your EMI. You figure it out.” That means the borrower has to manage cash flow, store earnings from high season, and make fixed payments throughout the year. It creates both a cognitive and financial burden – just because the product is inflexible.

Let’s take a concrete example: say you run an ice cream shop, and your entire business is concentrated in the summer. You want to buy a new freezer – some capital equipment – but there’s no way to repay that in four months. You need a two- or three-year loan, but the standard EMI model doesn’t match your income profile.

Right now, you’re forced to carry that mismatch. But what if we could design a more flexible credit product – something that actually scales with how your business works? Maybe it adjusts payments based on revenue cycles or offers repayment holidays in off-peak months.

If you can do that – if you can build credit products that fit the customer instead of making the customer fit the product – you dramatically expand who can borrow. You reduce default risk. And you unlock a much larger segment of the SME market that is currently excluded simply because the product design isn’t aligned with their reality.

It would seem remiss to end a conversation like this without asking: in five years, AdalFi – where are we? What does success look like?

So we are confident that what we’ve built is something that has broad applicability. It’s not just a Pakistan thing – everything we’ve done can be used more widely.

Hopefully, in the next 30 days, we’ll announce our first customer outside of Pakistan. We’re in final-stage discussions with a bank outside the country. So we know we’ve got something that travels – and we’re ready.

I see us really being at the heart of the lending infrastructure across emerging markets, right? Because the problems are quite similar – whether it’s countries like Pakistan or Bangladesh, or countries like Saudi and Malaysia. They’re a little further along in the journey, okay – but that just means there are things you can put right now, so you don’t end up in situations where even a country like the U.S., with its huge and complex financial infrastructure, still has SMEs struggling to access credit.

There’s been a lot of focus in the last 15 years on financial inclusion – rightly so. But what we see now in a lot of countries is a three-tier structure:

1. At the centre, you’ve got medium-sized businesses. They’re banked, they have access to credit – that part of the market works.

2. Then on the outer edge, you’ve got micro-businesses. They’re often unbanked, but they’re actually served reasonably well by microfinance institutions, which were built to solve that problem.

3. But the people in the middle – the small businesses, the “S” of SME – they’re banked, but they don’t have access to credit. And that’s the real problem.

Microfinance institutions don’t serve them – they’re not unbanked enough. And commercial banks don’t lend to them – they’re not formal or visible enough. So they just fall through the cracks.

And it’s really my wish and dream that we can fix that. That we can bring credit access to that missing middle.