Fintech 🧠 Food - The Co-Pilot Revolution: How ChatGPT Changes Fintech.
Plus leaked stripe results, India and Singapore link payment systems and why you should never outsource your ledger to your payments processor.
Hey everyone 👋, thanks for coming back to Brainfood, where I take the week's biggest events and try to get under the skin of what's happening in Fintech. If you're reading this and haven't signed up, join the 27,487 others by clicking below, and to the regular readers, thank you. 🙏
Hey Fintech Nerds 👋
These days people love to beat up on Tech and Fintech (unless it's Generative AI, of course). The narrative has firmly shifted. But the truth is always more complicated.
Tech companies are being punished for having grown incredibly during the pandemic and now having to reset back to the pre-pandemic averages. But they're still growing, and some are growing earnings.
The Stripe vs. FIS example (I cover more in Things to Know) is a perfect microcosm. Stripe makes the headlines for its short-term adjustment back to baseline, but FIS has to adjust to a longer-term trend, being disrupted.
There's a lot of uncertainty in markets, and we're in a period of flux. If only we could ask an AI what happens next.
Speaking of Generative AI. It's taken me a while to gather my thoughts on its impact on Fintech, primarily because I couldn't escape the conclusion that the answer was "the same as every other industry." To some extent, that's true, but the double click is more interesting. The Fin in Fintech is a complicated beast I enjoyed exploring.
This essay took two weeks, so I hope you enjoy.
Here's this week's Brainfood in summary
📣 Rant: The Co-Pilot revolution; How ChatGPT changes Fintech. Generative AI works best as a co-pilot for humans to manage complexity rapidly and reduce admin. Operators need to learn new skills to get the most out of it, and there are opportunities in the Fintech domain in compliance, marketing, and of course, engineering. But what excites me is if we could train a model on finance data and use ChatGPT as a UI.
💸 4 Fintech Companies:
Kennek - Credit workflow automation (EU)
Uplinq - Alternative credit scoring for SMBs
Mistho - Payroll API for Europe
Dotfile - Onboarding workflow
👀 Things to Know:
1. Leaked Stripe results paint an interesting story for the industry. Stripe is reported to have 2022 gross revenues of $14.3bn, $3.4bn in net, and on track for $100m in EBITDA next year. 🤔 These are solid numbers, and they're slowly becoming a more rounded business. They could graduate from their difficult teenage phase very soon.
2. India and Singapore link real-time payment systems. This link enables real-time cross-border payments for low fees for the $1bn sent annually. 🤔 As the US adds, FedNow could a similar link be on the horizon for the 4.5m ethnic Indians that live in the US?
📚 Good Read: The match should always add up (or why you should never outsource your ledger to a payments processor). Trupti makes the case that relying on bank account statements and payment processors as a source of truth for cash flow is dangerous. 🤔 Businesses that scale use many payment processors, which is a key opportunity for the many "payment automation" startups that have appeared.
Weekly Rant 📣
The Co-Pilot Revolution: How Chat GPT Changes Fintech
Generative AI will become a critical co-pilot for Fintech companies and operators in financial services. But for that to happen, we must understand its opportunities and limitations for a regulated industry. We can't have bots giving bad financial advice!
To unlock its potential, we need to be able to access financial services data, and for extra credit, we need language models that can partner with more finance-specific models. We need AI that can talk to other AIs.
But let’s step back.
Tech predictions are hard.
The hardest thing about new technologies is predicting what life will be like with them. We tend to be directionally correct, but the form factor is always rooted in our current context and understanding of the world.
The Star Trek communicator predicted the mobile phone, but we held it in front of our faces, and it had no video.
The personal digital assistants (PDAs) from the 90s had email and the internet but didn't imagine TikTok.
Generative AI is the hottest thing in Tech.
First, it was Dall-E and Stable Diffusion changing creative industries.
The pilot always has an intent, and the computer is the co-pilot.
ChatCPT could finally make Chat the most efficient interface between humans and computers. Alexa and Siri were good toys but mostly useless.
ChatGPT is useful, but its Achilles heel is that it is often wrong.
It would not make a good financial advisor 👇
The problem? ChatGPT is trained on the internet, and the internet is weird.
Finance as an industry craves certainty.
An occasionally hallucinating bot could give the wrong advice and land a regulated business in hot water.
ChatGPT's ability to be wrong could be catastrophic for a regulated industry like financial services. Just look at how the press is covering Microsoft's "creepy new" AI to see that once a large institution goes near AI, the "reputational risk" can explode. If a Chat AI gave bad financial advice or were rude to a customer, that would make headlines. No bank or Fintech company CEO wants to make headlines for the wrong reasons.
Because where the headlines go, the regulators (and fines) follow.
No wonder then that JP Morgan is banning staff from ChatGPT
Prompts matter. And the prompter (human) is the pilot.
Generative AI’s ability to create efficiency when properly supervised cannot be ignored. It is like hiring a college-educated all-rounder with little or no domain knowledge but god-like efficiency for $20 a month.
Like some hires at the beginning of their career, it delivers messages confidently but can be wrong. Your job as a human is to curate that as a good manager should.
ChatGPT works best as a co-pilot to a human, handling the time-consuming admin, providing research, and being helpful. Ask any software engineer who's used OpenAI co-pilot for writing code, and they'll tell you it's a huge unlock for efficiency vs. traditional approaches.
Use cases in Fintech will be similar and are already starting. Navan (the artist formerly known as TripActions) is already using ChatGPT to help its travel managers make recommendations for trips.
How does that change Fintech?
Fintech operators need to get good at prompting.
Support function use cases like marketing and engineering are emerging, where ChatGPT is proving to be a powerful co-pilot. This is where the biggest impact will be in the short term. (I go deeper in section 4 of this essay).
But I’m looking for something that also reaches the industry's core. Not just using the tool to be more efficient at what is already done but to change the game. This essay covers both.
That's all Tech. But where's the Fin in Fintech?
We need to understand the nuance of financial services data to see how Generative AI uniquely changes the industry. We need to put it in the context of other AI and data efforts, and then we can build a roadmap.
Finance data is hard to use. It is regulated, siloed, and of poor quality.
Fintech companies have already done a lot with AI, like cash-flow underwriting, data cleansing, fraud, AML, and compliance tech. ChatGPT itself was trained on the entire internet. Training a model on all financial services data is a massive challenge.
Early Generative AI use cases apply to any sector. Fintech companies will use it as a tool operationally. The compliance, engineering, and marketing co-pilot is a potential massive upgrade in efficiency, however. But there is a much bigger opportunity to disrupt the industry's core profit centers.
Generative AI use cases that work for Fintech today. Co-Pilots for departments and Chat as a UI to abstract complexity and simplify.
Building a GPT-like model requires more data access at the individual customer and industry-scale. We could consider numerous ways to achieve this (like open finance, synthetic data, and federated learning).
Finance data would unlock new customer-facing opportunities in Fintech. The holy grail is a private banker AI. Having a team of people who can help you or your business be more efficient and achieve better financial outcomes.
Finance data would unlock new infrastructure opportunities in Fintech. With accurate data and access, Fintech companies have a staff of unlimited analysts. Spreadsheets run the world (especially in finance), but Chat is the UI that can finally dislodge them with more convenient dashboards and answer complex questions quickly.
Don't forget the regulators. Supervisory Tech (Suptech) has enormous potential. If compliance teams are understaffed and deal with too much complexity, pour one out for people in Government with a day job.
What happens next? Generative AI will be a transformational tool for financial services, especially as GPT-4 arrives, promising more accuracy. Short-term, expect co-pilots and chat as a UI for complexity to dominate.
1. Finance data is hard. 😫
Financial services data is hard to use.
The problem of building a model large enough to compete with ChatGPT for a domain like finance is getting enough data. You can't train a model on data you can't use, and most financial services data is:
C) Poorly quality
A) Regulated data is hard to use: While every industry wrestles with privacy rules like GDPR and PII, financial services layer a whole other level of regulation and law. Most crime involves money, and most governments require the financial services industry to be the police of money. As noted in "KYC is still broken"
To do almost anything in financial services, you need to Know Your Customer (KYC). It's the law. The fine for not doing KYC is a minimum of $500,000 and three years in jail for executives.
Since 2008, regulators have issued over $403bn in penalties for KYC and AML violations.
KYC is just one set of regulations; thousands of them exist.
Every company in financial services wrestles with using customer data because they have to manage the spider web of regulations that might be triggered before they can.
B) Finance data is siloed within organizations and between organizations.
Data is siloed when you look at it from individual customer and financial institution perspectives.
Consumers have many accounts for many products and different institutions. Americans have, on average, 3 credit cards and at least 3 Fintech apps installed. That is before you consider mortgages, savings, or investments. For a large business, this multiplies as they cross multiple geographies.
Many large banks have silos created through M&A, historic procurement choices, and outsourcing. A bank's family history is like a family's DNA history. Banks inherit new customers, technology, databases, and vendors with each acquisition. Some will attempt to rationalize this all onto modern technology stacks. However, the reality is that most don't, or if they do, the underlying data is still siloed in various ways, making it hard to access or combine easily.
Newer (born post-2016) Consumer and SMB-facing Fintech companies fare better. They might have a modern data tech stack and data vault like Basis Theory or Skyflow handling customer PII. Older Fintech companies might be managing this more themselves but have less of a silo problem. The biggest problem facing Fintech companies is they're often a small slice of their customer's life which limits the potential of their models to be customer facing.
There's likely no single Fintech company or bank alone that could train a model at the scale of data GPT-3 had been trained on.
And even if there was, there's the data quality problem.
C) Most financial data is of poor quality.
I think there are three reasons
Old data structures persist: Every time you use a card, a small file is sent somewhere behind the scenes (think of it as a .txt attachment to an email). They were never designed to be consumed by the public or developers building apps. One example is that Fintech operators are fond of citing ISO8583 files used in card payments. When the industry "digitized paper," it became a PDF and used the underlying data structures they always had.
Data governance wasn't a discipline when most data structures were designed. Despite spending billions on data governance programs, most large banks have not transformed their core data structures (with notable exceptions like Capital One and JPMC). Data scientists are often not at the table, as consultancies and executive leadership attempt to build new business processes to fix a data quality problem
Government standards for “regulatory reporting” are narrow and antiquated. Government agencies receive regulatory reports, usually in the form of a PDF or CSV. The standards for these reports are not modern, and various agencies want that data sliced and diced differently. Most governments and agencies lack sophisticated tools to extract that data and find risks. As new regulations get added, agencies add FTE. These regulatory reports ("reg reporting") often come from a large bank's legacy system. Banks don’t want to upgrade their legacy core tech, which is painful, expensive, and risky.
So yeah, Finance data is hard.
But damn, it's valuable.
2. Fintech companies have transformed finance with AI. 💱
The tech part of Fintech companies brought a generation of data scientists, engineers, and talent into the sleepy financial services industry. The advent of infinite data storage and compute in the cloud, and modern data tooling from Amazon, Microsoft, and Google means some past constraints don't apply to companies born in the past decade.
Below are a few examples of how better data, data science, and tooling have impacted the industry.
Data aggregation: Open finance allows customers and 3rd parties to see real-time checking, savings, investing, and loan data.
Data cleansing: A wave of companies now cleanse, normalize and enrich that data, so it's ready to use (this alone is a massive uplift)
Cashflow-based underwriting: Using a customer's willingness to pay bills and not just their income or previous credit performance created new lending categories. It is slowly becoming an industry standard, with even big banks adopting it.
Fraud prevention: Sardine*, Seon, Riskified, and many others apply modern ML techniques to catch more fraud than historically possible. During onboarding, companies will also look for liveness detection (detecting you're an actual human).
Data vaults: Fintech (and non-finance) companies can outsource the complexity of GDPR and manage PII almost entirely. Data vaults (like VGS, Skyflow, and Basis Theory) can store customers' data during onboarding and manage it from then on.
Chatbots: The "OG" use case to quickly manage customer queries. These much-maligned experiences exploded in popularity during the pandemic and have been effective. But for the most part, they triage an issue and deflect the most common queries. They could be much, much better.
(These are mostly consumer examples, I could go on, but we'd be here all week)
All of those use cases have something in common.
They're not generative AI.
3. Early ChatGPT use cases work for all sectors 🚀
(but could be low-key game changers for Fintech too)
Three areas immediately emerge when I think of text Generative AI like ChatGPT.
Engineering: Engineers using Co-Pilot from OpenAI rave about how quickly it can perform repetitive tasks, identify bugs and remove a lot of busy work. Fintech companies (and banks that haven't outsourced it all) have engineering talent; they can use ChatGPT as the co-pilot.
Research: ChatGPT will find things that wouldn't appear in a search result given the right prompts and summarize them. Ask it which regulations apply to lending to consumers in the United States or what the most common types of fraud are, and explain those typologies. It will do an above-average job. It works particularly well when there are massive amounts of information to sift through and summarize.
Marketing: Marketers can grind out copy with ChatGPT, Jasper, and Lex, but the best are adopting it for idea generation. For example, try asking about compliance officers' fears and goals and then suggest 5 counterintuitive ways to achieve those goals.
The theme here is time-saving stuff you would have had to do anyway and doing it quite well. It is much stronger on deterministic responses where the internet has a great data set (like engineering): short term, ChatGPT is a great co-pilot.
The key thing to remember using ChatGPT is that it is designed to give the most likely output to a prompt or the next set of words. It is an adversary to creativity or expertise unless you prompt it not to do that.
4. Generative AI for Fintech today
Co-pilots for departments: What would a compliance co-pilot do? How would an underwriting co-pilot support you? What might a fraud analyst co-pilot spot that you might not think about?
Chat as an interface for complexity: How do business reports and dashboards change if you can ask them questions?
Co-Pilots for Financial Services Operators: Have you read Dodd-Frank or the Fair Lending Act from cover to cover? Me neither, and I read a lot. I often research passages or rules, but when it comes to if this particular transaction, customer, or event is allowed, I have relied on compliance superheroes to keep me honest. Those heroes could use help as spending on fraud and compliance continues to skyrocket.
When you read a rule or regulation, it's written in legalese. It is often designed to be accurate and comprehensive, and there are thousands of them. Simply prompting "Which regulations apply in xyz scenario" would save much scrambling and busy work. It might not give a comprehensive answer, but it does help massively.
Chat as an Interface for Complexity: Answering a question can involve countless inputs from countless sources, but provided you can copy + paste + prompt, ChatGPT and AI might help make sense of it. A question like "Help me translate these Chinese and Spanish regulations to see if they will impact our cost of funding next year." What might take teams in different departments collaborating today could be streamlined by a series of prompts.
These use cases are valuable and could be significant time savers for individuals and teams. Saving time is saving money. Saving time is one less hire or more productivity. We need that.
I expect to see a swathe of Fintech companies focussing on vertical-specific use cases. But doing that with a generic model is only a part of the opportunity.
These use cases apply to any regulated industry.
Where's the Fin in this Tech?
5. How we get GPT-like finance AI 🤖
There are two problems to solve
Training the model given the constraints of #1
Data access would be solvable if we have a working open finance infrastructure and the data can be sufficiently cleansed. This would enable the existing ChatGPT model to be applied to individual customers' account data. That alone could be huge (and not that hard to do).
Imagine prompting, "If I want to go on holiday this year and save for a new rental deposit, build me a personalized financial plan. Recommend accounts I should close, new ones I should open, and list my next actions."
But we'd still have the problem of ChatGPT occasionally hallucinating. And would you want to be the provider of the service that recommended a consumer close their savings account and take on more debt and have to explain why to a regulator?
We need to train a new model on an industry-scale data set.
It would look something like this
As Francisco says on his Chaos Engineering substack (image also from there)
The conclusion is simple: ChatGPT will learn superficial stuff about finance, but in order for it to become a financial expert, it needs to be explicitly trained to be.
But how do you get the data?
How do we access it?
Open finance isn't a solution because it relies on individual customers allowing access to their data. It's not a way to view data at an industry scale (that said, imagine if a Government mandated the use of open finance plumbing to build an industry-scale data set).
There are two more options if we can't access the data through a single API.
Training a model using synthetic data. The idea is to create a realistic "twin" of financial services data sets without the challenges of missing data, poor quality, or it is regulated. Fintech companies like Hazy do this, and JP Morgan even has a blog about their process for doing so. But it's hard to get at the industry-scale (again, unless someone mandated doing so)
The limitation of synthetic data is will often produce wrong results (for models trying to detect fraud, for instance). The approach could be useful, but we can't rely on it.
Training a model using federated learning. Federated learning is a way to prepare a shared model without moving the underlying data from where it lives today. Cloud vendors have started offering this technique, which could be powerful if we gain enough adoption.
We could build a more use-case relevant set of models if we had a mix of data access, synthetic data, and federated learning.
6. Customer-facing use cases 💻
Let's assume two scenarios for a “Finance-GPT” model in Fintech.
For ChatGPT with access to customer account data
For a GPT model trained on industry-scale data
The low-hanging fruit productizing GPT as it is today + data access + context.
For FinanceGPT with access: The best financial health practices for most consumers are well understood. With access to customer data, they could account for nuances like your current situation and context. A business might be better off not raising debt on paper, but all its competitors are. A GPT-like model could (if prompted correctly) consider this.
(In his blog post, Francisco shares how he got ChatGPT to do this)
Productizing prompts for consumers also helps limit the risk surface. Just as Microsoft has now prevented its GPT model from talking about itself, vertical-GPT companies can provide guard rails above the more general model where danger lives.
Personal Finance Management (PFM) apps have built this type of AI for over a decade. With ChatGPT and their data access, we may get closer to the private banker in your pocket for the mass market. (PFM-GPT?)
What does that look like for accounting? Customer service? Heck, even the humble chatbot?
There's a lot to unpack here. Take a picture of a company's value chain and its customer touchpoints, and then ask what happens if GPT models have access.
For a model trained at industry-scale: Suddenly, you can ask interesting questions like "What do people on my income do financially that I don't, that helps them be better off?" The example I gave earlier of "help me find the best financial products for my financial situation" might be possible if the model understood regulation, customer behavior, and outcomes over time.
A model trained this way wouldn't guarantee a lack of hallucinations, but combining data access with a model trained on industry-scale data becomes less likely.
But I can't escape that it would require an ecosystem of vertical-specific companies in the early going to productize (and keep it away from danger).
7. Operational use cases ⚙
What happens if you remix GPT + department activity?
Let's assume for this section, a GPT model has
Access to data within an institution
Is trained with industry-scale data
At the institutional level, compliance departments can start to prompt things like "Find customers where KYC information is out of date and send them reminders to update this information."
Or "Assess the areas where new regulation ABC impacts our current customer base, and identify next steps we should take."
Underwriting teams might prompt things like, "if we changed our deposit pricing to X and our lending pricing to Y, how would that change our overall cost of funds and profitability?"
There are much simpler examples too.
But what would you do in your Fintech day job if you had access to a GPT model that could see all of your data?
8. Don't forget the regulators 👩⚖️
What happens if you remix supervision + GPT?
The reporting regulators get low quality and must make tough calls based on it. Ultimately, it's the same data sitting somewhere. It just can't be queried.
Governments have a hard time procuring anything new but assuming you could solve that. Imagine being able to prompt, "How well implemented is new regulation X based on the industry's data today?"
9. What happens next? 🤔
Chat GPT won't disrupt Fintech.
It will disrupt everything.
At first, Fintech is just one of those things.
Financial services is a complex industry, but it craves accuracy. The numbers must always add up. We could build something novel with new technologies like federated learning if we can train Generative AI on finance data sets.
So here are my themes.
Generative AI is a co-pilot today. It needs a skilled pilot to know what they want to achieve. But what if that changes?
Chat is the ultimate UI to compress complexity. Needs a business person to know what questions to ask. But what if that changes?
Models need context, and today that context is provided by the person prompting.
Industry or customer-level data sets provide more context.
Open banking and just copying + pasting a regulatory PDF document passage could be quick wins. But I'm in love with the idea of Chat working with other ML models over the long term
It's going to be a fun decade for the curious.
You better learn how to prompt and get yourself a license to GPT.
Damn, this was a hard rant to get done.
But it was fun.
Until next time.
PS. Reply to this email with your thoughts. In this essay, I almost dismissed the ~5-year opportunity in favor of federated learning. Where should I explore next?
PPS. Federated learning is an excellent and under-explored technology for training models on industry-wide data sets. And if anyone is doing that. I'd love to talk to you.
4 Fintech Companies 💸
1. Kennek - Credit workflow automation (EU)
Kennek helps lenders manage a loan from origination to maturity with its workflow tool. Kennek has built a set of default workflows, or lenders can build different flows to suit their needs. On the back end, there's also a funding marketplace for credit investors to buy the loans.
🤔 Lending is hot, workflow automation is hot, and I haven't seen a lending workflow platform for the EU yet. Kennek's timing is excellent. Their likely target clients will be smaller and upstart lenders, but they could do well with larger organizations. Traditional lending software providers have proven hard to dislodge because of their ability to handle the workflows and complexity involved in Lending. And there's a lot. Building a desk that can sell the loans and fund the balance sheet is one of the hardest things to manage. Abstracting that painful process is powerful. You can tell that the people building this are deep lending experts.
2. 401 Financial - Wealth advice for minorities and PoC
401 Financial is a financial advisor with a flat fee structure ($500 per month) instead of the traditional 1% of AUM. 401 doesn't have a minimum account size or hidden charges. Users can access the 401 Financial resources and create a tailored financial plan for that fee.
🤔 Flat fees are epic for consumers but leave a lot of money on the table. If you ever do the math on retirement planning, switching to a flat fee 401k or advisor can near double your retirement fund over 30 years. So this is an option for the financially savvy and upwardly mobile, but that $500 per month fee is a decent family car. The messaging will need to be strong to cut through that, but I believe there is a minority mass-affluent market that doesn't have a good advisor option that's the smarter economic choice. 100k customers would be a $50m monthly run rate, but they'll need to diversify. How many folks would pay $500 once for the plan and never return?
3. Mistho - Payroll API for Europe
Mistho is a payroll API that connects up to 20m UK employees (70% of the workforce) via its API for use cases like income verification or direct deposit (current account) switching. Use cases include applications for a mortgage, car finance
🤔 Payroll company APIs are not a thing in the UK or Europe, whereas it's an entire startup category in the US. Part of that is because payroll companies don't have nearly the same coverage. Many largest employers (like Mcdonald's, Tesco, or the NHS) have a proprietary portal with no API access. However, 70+% Like Atomic and Pinwheel in the US, the primary competitor to Payroll APIs is likely the credit ratings agencies (Experian, Equifax, etc.). Mistho could be as transformational for the UK, but remember; the UK has a population the size of California. For Mistho to be big, they need to win in Europe too.
4. Dotfile - Onboarding workflow (EU)
Dotfile aggregates dozens of KYC, KYB, and AML providers and lets them quickly build workflows. Dotfile has handled the contracts and the integrations focussed primarily on European providers.
🤔 Europe doesn't really have an Alloy competitor (although Alloy just launched in the UK last week). So Dotfile could have good timing. Dotfile is also focussed on Europe more broadly (they're based in Paris). And as I often remind US-based folks, "Europe" is a highly fragmented Fintech market. The value of one contract and one API reduces operational overhead and allows smaller companies to scale. But as those companies get bigger, performance matters much more, and there's much more to life than onboarding.
Things to know 👀
According to the Information, Stripe is pitching a story to investors that it is growing faster than much of the market, has promising new sources of revenue, and is getting new customers focusing on AI. its leaked results paint a compelling picture—gross revenues of $14.3bn, $3.4bn in net, and on track for $100m in EBITDA next year.
🤔 Stripe indexes heavily to e-commerce which is correcting but will resume historical growth patterns. Like many Tech companies, Stripe had to ramp up hiring to cope with demand and growth during the "new normal" phase of the pandemic. As the world re-opened, e-commerce growth slowed but did not vanish. Once things re-balance, I expect normal service to resume. (I went much deeper in Stripe's difficult teenage phase here)
🤔 The bigger story is incumbent disruption. Incumbents like Fiserv and FIS (Worldpay) grew closer to 5% YoY. Incumbents have lost significant market share to disrupters in the past decade.
🤔 Perhaps the best example is FIS. Meanwhile, incumbent FIS will divest Worldpay fresh from a round of layoffs. Incumbent acquirers have eroded significant market share to "disrupters" like Adyen, Square, Stripe, and PayPal in the past decade
India and Singapore have linked their payment systems for real-time cross-border money transfers. Estimates place the cross-border payments volume to be $1bn per year. The goal is to reduce the cost and time it takes to send money for remittances or business use cases.
🤔 This is a logical starting point for connecting domestic payment systems. Ethnic Indians are around 10% of the Singaporean population, and both countries have real-time consumer payment systems (UPI and PayNow, respectively).
🤔 An interesting next step after Singapore could be the UK. The UK has a domestic instant payment system and 1.8m ethnic Indians from the 2021 census, which is 10x more than Singapore has). Annual remittance flows are ~$4bn. That said, the UK is upgrading its RTGS, so this might not happen soon.
🤔 If the US had a real-time system, the potential cross-border remittance volume could be substantial. The US is home to 4.5m ethnic Indians who remit ~$12bn annually and is the 2nd largest corridor. FedNow is set for launch in the coming years. Connecting to UPI may take some time, but that feels inevitable. The largest remittance corridor is the UAE at $~14bn annually. The UAE also launches its real-time payment system in 2023.
Good Reads 📚
1. The match should always add up (or why you should never outsource your ledger to a payments processor)
Trupti makes the case that relying on bank account statements and payment processors as a source of truth for cash flow is dangerous. If merchants have any pricing offers, sales disputes, or order amendments that will impact their cash flow. If those events are not logged at the end of the year, the accounts will not reconcile, and your accountant will have to be able to explain that to tax authorities or auditors.
🤔 Making sure accounts add up (reconciliation or recs) is an afterthought for most early-stage companies, but it is life for at-scale merchants and marketplaces. Determining what happened from bank statements is often challenging because the payment amount or dates may not match. Finance teams become detectives very quickly.
🤔 This is why many ledger-as-a-service companies like Payable*, Segment, and Nilus have started to gain traction. The initial value these companies provide is scale and automation, but longer term, they're experts in ledgering and recs.
🤔 Most accounting systems were not built to help developers build a source of truth. Quickbooks, Xero, and Netsuite all allow outputs and have APIs, but they're getting older as products. What would it look like to re-imagine that developer-first?
2. 🤓 Extra Credit:
This excellent piece by Alex Johnson on scams vs. fraud is a must-read.
Tweets of the week 🕊
That's all, folks. 👋
Remember, if you're enjoying this content, please do tell all your fintech friends to check it out and hit the subscribe button :)
Disclosures: (1) All content and views expressed here are the authors' personal opinions and do not reflect the views of any of their employers or employees. (2) All companies or assets mentioned by the author in which the author has a personal and/or financial interest are denoted with a * (3) Any companies mentioned in Rants are top of mind and used for illustrative purposes only.
If you read all of this, hit me with a reply and prove it.