Ami choli kotha bolar age, ek cup cha dei. Bosen. Thirty years I have been on this corner of Motijheel, behind the Janata Bank gate where the rickshaws park when it rains, and every variety of bhai walks past this stall. The bank manager in the morning, the broker after lunch, lately the SaaS bhai with the laptop bag and the slightly nervous smile. Tara shobai vaben ora ekta naya jinish niye eshechen, mostly things which my brown khata has already had since 1996. Saquif bhai sat on this bench for two months in 2020 asking me about that khata, and his company kept running long after he flew off. He says I should put my version down somewhere, so this is it, mostly his words in my voice.
There is a mudi dokan around the corner from me on Mohammadpur Town Hall road. Mostofa bhai runs it, six feet by nine, painted a green that hasn't been actually green since 2007, with a tin awning that doubles as a percussion instrument when it rains. He sells Pran biscuits, Square detergent, Igloo ice cream when his freezer is working, and seventeen brands of cigarette in fractional sticks. Forty thousand taka a day moves through that shop, and none of it is recorded anywhere a bank could see.
What is recorded is a brown notebook on a hook by the calendar. In that notebook, in pencil, Karim bhai (his wholesaler, comes from Karwan Bazar in a CNG twice a week) keeps track of who got how much haldi powder on credit, when, and against what kind of promise for the following week. Mostofa bhai owes Karim bhai, the neighbourhood owes Mostofa bhai, Karim bhai owes the distributor in his turn, and four hops further up the chain there is a brand nobody in this lane has ever met. There is a bank in the picture too, but the bank sits outside the chain even though every bank manager I've served chai to thinks his bank is the centre of it. The actual centre of gravity is that pencil notebook on the hook by the calendar.
This is what Saquif bhai's company SupplyLine bet on. He cofounded it in March 2020 and stepped back in March 2022, and the company still operates today under his cofounders, who are by all accounts much better at running a lending book than he ever was.
The classic move that every FinTech bhai flying in from Dubai or Singapore opens his pitch with is a story about formalisation. Bring the unbanked into the system, onboard the shopkeeper, digitise the ledger, KYC everyone. The pitch always says the bank is the source of truth, and the startup is the kind extender of its reach. This is wrong in a way that took Saquif bhai four months and a great many declined loans to fully appreciate, and which the rest of the formalisers on this street have not appreciated at all.
The bank is not the source of truth in Bangladesh. On a good week the bank lands somewhere ninth or tenth in trust on a shopkeeper's phone, below the wholesaler, the wife, the nephew working in Doha, the bKash agent, the local Pran salesman, two landlords, and the imam at the mosque. Credit doesn't live there. It lives in the brown khata, between two people who have known each other since Karim bhai's father was selling rice to Mostofa bhai's father.
What SupplyLine did instead was assume the khata between wholesaler and retailer was the source of truth, and that the company's job was to compute on top of it. A stack of alternative signals fed the credit rating model, and none of them were salary slips.
The rest of the stack ran through supplier diversification, repayment latency on existing trade credit, MFS exposure through family members, the ratio of stick cigarettes to packed cigarettes ordered, and one signal Saquif bhai still won't describe to me because, he says, it is the only thing that ever felt proprietary. The point of the whole stack is that a basket of weak signals, properly weighted, beats one strong signal you do not actually have. There is a Peruvian economist Saquif bhai keeps quoting at me, Hernando de Soto, who once wrote in The Mystery of Capital that the poor are not poor in assets, they are poor in the legibility of those assets. SupplyLine wasn't creating new information so much as making legible what Karim bhai already knew with his pencil.
There was a retailer in Mirpur 11, paan and cigarettes, six square metres of shop, who applied for forty thousand taka of trade credit through a wholesaler SupplyLine had partnered with. By the bank's standards he was unbankable, with no collateral, no formal accounts, just an NID and a trade licence. By the SupplyLine model he was a 7 out of 10, decent, not stellar. The model wanted to give him twenty five thousand, while one cofounder pushed for the full forty and Saquif bhai wanted to give him fifteen and see what happened. They went with forty in the end, mostly because the cofounder shouted slightly louder that afternoon.
He paid back in six days. The trade credit was a seven day, zero percent invoice discounting product, and he paid on day six, in full, through bKash, and immediately took another forty. The last time I asked, he had not missed a payment in three years.
The reason the bank would have declined that retailer was not in SupplyLine's data either. It was in the head of a wholesaler like Karim bhai, where eleven years of buying from the same supplier (and his father from that supplier's father) sat as a soft pencil scratch in the khata that SupplyLine eventually learned to read.
The opposite case is more painful and I will keep it short out of respect. The team approved a pharmacy in Uttara that scored a clean 8, with good signals across the board, and it defaulted within 45 days. Turned out the owner was running two NIDs, two trade licences, and a quiet gambling habit on a third phone the model had not picked up. They added device graph signals after that. There is a book by Sendhil Mullainathan and Eldar Shafir called Scarcity with a lovely chapter on how the cognitive load of running on the edge produces exactly this kind of failure. SupplyLine had treated the shopkeeper as a rational agent, but he was a tired one, and most of us on this street are.
The product principle Saquif bhai pulled out of all this, and now applies to every product he touches in a market like this one, comes down to one plain idea. The trust network already running between two people for thirteen years is a protocol the FinTech is supposed to compose with, not a competitor to be replaced.
The product trick, the one that took the team a year to find and ten minutes to explain, was that the retailer never directly took a loan from SupplyLine; the wholesaler did, on his receivables. SupplyLine sat on the trade credit relationship that already existed and made it bigger, faster, and priced by a model instead of a human. The retailer experienced it as Karim bhai being a little more generous this month, and Karim bhai experienced it as SupplyLine just buying his receivables and absorbing the default risk. Neither version was a lie, and neither party had any reason to dig past the version they were given.
The cleanest before and after story SupplyLine had was approval rate. They ran a comparison cohort over four weeks with one wholesaler in Old Dhaka in early 2021, against the indicative numbers a partner NBFI would have approved on the same applicants under traditional underwriting. It was not a controlled study; it was a slide in a fundraising deck, with that caveat noisily up front. What actually mattered wasn't the gap in approval rate so much as the slope behind it. The traditional underwriter was learning slowly, because every new application was a new file, while SupplyLine was learning fast, because every transaction in the wholesaler's notebook was a new data point on a population they already knew. That compounding is where the moat lives, and it was never really in the model architecture at all, but in the plain fact that SupplyLine was the only company in the country reading the brown khata.
I will keep this gentle, because Saquif bhai knows most of the people who run the companies he is about to gesture at, and a few of them also drink my cha. The pattern that keeps repeating, with small variations, goes something like this. A team raises a seed in Singapore or Dubai on a deck full of TAM slides about the unbanked SME segment, and then they build a beautiful app with a dark mode that expects the shopkeeper to upload his trade licence as a PDF. The team flies in for a launch event in a hotel ballroom in Gulshan. The first cohort of users is recruited through a friendly chamber of commerce, and the numbers look good for two quarters because the friendly chamber's members are already the most formalised SMEs in the country. Then the team tries to scale beyond the chamber and hits the actual median retailer, the one with no working email address. So they go home and write a thoughtful blog post about how the regulatory environment is challenging.
The Bangladesh microfinance heritage, Grameen and BRAC and the rest, is sometimes invoked as a kind of moral cover by the FinTech bhais who arrive in Gulshan, as if having Yunus on the national wikipedia entry means your B2B FinTech has cultural permission to skip the time on the bench. That permission isn't there to be borrowed. Microfinance worked because it took the trust network of the existing solidarity groups seriously, and most modern B2B FinTech in this country has the iconography of that heritage and none of the architecture. There is a BIS working paper from 2022 on underwriting with alternative data in emerging markets that makes roughly the same point in much more diplomatic English, worth reading if you are about to raise on a TAM slide.
Saquif bhai left active operations at SupplyLine in March 2022 to take a senior product role at Optimizely. The honest reason, he tells me, is that he is better at building the kind of product that defines a category than running a lending book, and SupplyLine had reached the point where running the book well mattered more than imagining the next feature. His cofounders are better at that than he is, they still run the company, and from where I sit it has weathered two currency crises, a regulatory squeeze on consumer lending, and the gradual entry of the bigger MFS players into adjacent products. As far as I can tell from this corner, the only reason the company has survived is that the core architecture has not been tampered with since 2021. Karim bhai still writes the order in the notebook the way he always has, and the SupplyLine system still reads what he writes there instead of asking him to relearn his job inside an app. The retailer pays back through bKash on day six the way he was already going to, and the trust score updates upstream into the wholesaler's ceiling without anyone in the chain having to perform a new ritual to make it happen.
What I keep handing to every bhai with a laptop bag who has come through here since 2020 is one plain idea, that you cannot replace a protocol already running quietly between two people who have known each other for thirteen years. The most you can do, if your team is good and the wind is at your back, is wrap a thin layer of computation around the protocol so that its existing intelligence becomes legible to the kinds of capital that previously could not read it. Whatever the term legacy system usually implies about something creaking and barely alive, it doesn't apply to the brown notebook on the hook in Mostofa bhai's shop, which the neighbourhood has been auditing for fifteen years just by using it. The bank is just one of several apps trying to interoperate with it from the outside. If anyone ever manages to beat what happens in that notebook, I expect I will hear about it from the Pran salesman who comes by every Thursday, well before any banker on the planet does.