Latest / MBA Training Data / Alternative credit scoring: how Tala lends without bureaus
Transcript
- 0:00Host: Welcome to the Leaders Insights podcast.
- 0:07Host: Today's episode, how Tala built a credit engine for the world's most invisible borrowers.
- 0:12Expert: There are roughly 2 billion adults on this planet with no credit history at all.
- 0:17Expert: Tala looked at that and said, we'll lend to them anyway.
- 0:20Host: How is that not just charity that goes bankrupt?
- 0:23Expert: Because charity doesn't return capital and Tala does.
- 0:26Expert: The trick everyone misses is that no credit history doesn't mean no data.
- 0:31Expert: It means no data in the format lenders got lazy about relying on.
- 0:35Expert: The bureau file.
- 0:37Expert: The tidy score.
- 0:39Expert: Tala went looking for signals somewhere else.
- 0:41Expert: The phone.
- 0:42Host: The phone.
- 0:43Host: So they're snooping through people's text messages to decide if they're creditworthy?
- 0:48Expert: With permission, yes.
- 0:49Expert: And less creepily than it sounds.
- 0:51Expert: When you install the Tala app, you consent to share device data.
- 0:55Expert: How many contacts you keep, whether you top up your airtime before it runs out.
- 0:59Expert: How regularly you move money around.
- 1:01Expert: None of that is a moral judgment.
- 1:04Expert: It's a behavioral proxy.
- 1:06Host: Yeti?
- 1:08Expert: A stand-in signal for the stable financial habits a bureau score would normally capture.
- 1:14Host: Come on, my contact list predicts whether I'll repay a loan?
- 1:17Host: That sounds like the kind of thing that gets you sued.
- 1:20Expert: It's not one variable.
- 1:22Expert: It's the pattern across hundreds.
- 1:24Expert: And here's the contrarian bit.
- 1:26Expert: The consensus in fintech is that you need better models.
- 1:29Expert: Wrong.
- 1:30Expert: Tala's early advantage wasn't a fancier algorithm.
- 1:33Expert: It was owning a data source nobody else bothered to collect.
- 1:37Expert: The model is almost boring.
- 1:39Expert: The data pipeline is the moat.
- 1:40Host: That's a convenient thing for a data person to say.
- 1:44Expert: It is, and I'll back it.
- 1:46Expert: Tala has dispersed over $6 billion across markets like Kenya, the Philippines, Mexico,
- 1:52Expert: India, to people who'd be an automatic decline anywhere else.
- 1:57Expert: You don't get there on model cleverness.
- 1:59Expert: You get there because you're the only one holding the raw material.
- 2:03Expert: MIT Sloan Management Review has made this point for years.
- 2:08Expert: In machine learning, proprietary data beats a marginally smarter algorithm almost every
- 2:13Expert: time.
- 2:14Host: Fine, but smartphone behavior changes constantly.
- 2:17Host: People switch phones, apps get updated, habits shift.
- 2:21Host: Get your entire signal rod underneath you?
- 2:24Expert: That's the real risk, and it's where most people underestimate the work.
- 2:28Expert: This is model drift.
- 2:30Expert: When the world moves and your predictions quietly get worse, while the dashboard still
- 2:34Expert: looks green.
- 2:36Expert: Tala has to retrain constantly because a signal that worked in Nairobi in 2022 might
- 2:41Expert: be noise by 2026.
- 2:44Expert: The lending decision is the easy part.
- 2:46Expert: Keeping the pipeline honest as behavior mutates.
- 2:50Expert: It's the actual job.
- 2:51Host: How do you even measure whether the data's still good?
- 2:55Host: Everyone claims their pipeline is clean.
- 2:57Expert: You test it, ruthlessly.
- 2:58Expert: DBT Labs, and I'll flag they sell the data transformation tooling, so crosscheck
- 3:03Expert: against O'Reilly's independent surveys, publishes numbers on how many teams still
- 3:08Expert: don't run automated tests on their data.
- 3:11Expert: It's uncomfortably high.
- 3:12Expert: O'Reilly's data engineering work backs the direction.
- 3:16Expert: Research failures aren't the model.
- 3:18Expert: They're broken inputs nobody checked.
- 3:20Expert: A lender running on phone data with no freshness tests isn't doing analytics.
- 3:25Expert: It's doing astrology.
- 3:27Host: So what stops a copycat from just cloning this?
- 3:30Host: If the mode is data, someone with a bigger balance sheet buys their way in.
- 3:34Expert: They try, but the data compounds.
- 3:37Expert: Every lone Tala issues and collects teaches the model something a newcomer doesn't
- 3:41Expert: know yet.
- 3:42Expert: That's the part balance sheets can't shortcut.
- 3:44Expert: You can buy servers.
- 3:46Expert: You can't buy repayment history you don't have.
- 3:48Expert: The invisible borrower stopped being invisible the moment Tala started keeping receipts.
- 3:53Host: Give me the uncomfortable limit.
- 3:55Host: Where does this model genuinely fail?
- 3:58Expert: When the proxy stops meaning what you think.
- 4:01Expert: If everyone games airtime top-ups because they've learned it boosts their score,
- 4:05Expert: your signal is now theater.
- 4:07Expert: Behavioral data is only predictive while people aren't performing for the algorithm.
- 4:12Expert: The day they are, you're back to lending blind.
- 4:14Expert: Wando, Asa Kasti, and possibly to the most sophisticated fraudsters, not the honest poor.
- 4:21Host: So what's the one thing a FinTech data leader should actually do with all this?
- 4:25Expert: Stop obsessing over the model and audit your inputs first.
- 4:29Expert: Pick your three most important data sources and ask one question.
- 4:33Expert: How would I know this week if they went stale?
- 4:36Expert: If you can't answer, that's your project, not a bigger algorithm.
- 4:40Host: Use for today's episode.
- 4:41Host: O'Reilly Radar, Katie Nuggets, The New Stack, Towards Data Science, MIT Sloan Management
- 4:47Host: Review, DPT Labs, Vendor, Data Tooling.
- 4:51Host: Done.
- 4:52Host: Want to know where you actually stand?
- 4:54Host: Take the CDO self-assessment at MBA-training.com.