In brief
Ally Bank's technology chief argues that a digital-only bank earns trust through security, consistency and simple experiences. Ally built its own AI platform early, tied to a clear goal: every customer should feel like the bank's only customer. It combined cloud data, embedded compliance staff and several AI models to move fast without losing customer confidence.
The central tension is speed versus safety. Ally moved quickly on generative AI after ChatGPT appeared, yet built a layer that strips personal data before it reaches outside models. Sathish also warns that small fintechs can now build quickly with AI. He says Ally must keep re-examining its own business, and that experimentation on problems others call too hard is how it stays ahead.
Questions
How did Ally use generative AI without exposing customer data?
Ally built its own platform, Ally AI, that sits between the bank and outside language models. Sathish says it removes personally identifiable information, tracks all traffic, and restores the details afterwards so the response still fits the customer.
Why does Ally use more than one AI model?
Different models are better at different tasks, so Ally picks the best fit for each use case. Sathish says it uses different ones for content creation, code generation and risk assessment, and can combine answers from several.
What does Ally mean by wanting every customer to feel like its only customer?
It is the bank's stated goal, according to Sathish. Customers compare banks with the best experience they have had anywhere, such as Amazon or Uber, so Ally uses cloud data and applications to personalize experiences in real time.
How does Ally handle compliance without slowing innovation?
It involves compliance, risk and legal staff at the start, when requirements and designs are written, rather than after the build. Sathish says this works through Ally's technology operating model, called Adam.
What did Ally learn from its COVID-era auto loan deferral offer?
Ally let customers defer auto loan payments for six months through a fully digital process that rewrote the contract. Sathish says customers who took the deferral still kept making payments, which built loyalty.
Facts
Stated by the speakers. Each quote matches the transcript, which shows what was said, not that it is true.
Sathish Muthukrishnan says Ally was started right after the 2008 financial crisis and is the largest direct-to-consumer digital bank in the US.
“Ally was started right after the worst financial crisis the country has ever seen in 2008. We are the largest direct-to-consumer digital bank in the US.”
0:02:01 · Sathish Muthukrishnan · Company history · 2008
Sathish Muthukrishnan says that about five years before the interview Ally still ran on mainframes, used hardware-based networking and had six different apps.
“it would be hard to believe that five years ago, we were still running on mainframes. We still had network that was plug and play, hardware enabled. We had six different apps.”
0:03:13 · Sathish Muthukrishnan · Legacy technology
Sathish Muthukrishnan says about 75 percent of Ally's applications run in the cloud.
“Having all our applications on cloud, now we have about 75 % of our applications running on cloud.”
0:08:37 · Sathish Muthukrishnan · Cloud migration
Sathish Muthukrishnan says about 95 percent of Ally's enterprise data is in a single cloud-enabled data warehouse.
“Having all our data in a single data warehouse that is cloud enabled, I have about 95 % of the enterprise data in the cloud.”
0:09:09 · Sathish Muthukrishnan · Data warehouse
During the peak of COVID, Ally offered auto loan customers a six-month payment deferral.
“So anybody that had an auto loan, we offered them to defer their payments for six months.”
0:10:52 · Sathish Muthukrishnan · COVID relief · 2020
Sathish Muthukrishnan says customers who accepted Ally's COVID auto loan deferral kept making their payments anyway.
“the consumers that took the deferral never deferred their payments.”
0:11:34 · Sathish Muthukrishnan · COVID relief · 2020
Sathish Muthukrishnan says Ally originates about $45 billion in auto loans a year and works with about 21,000 US car dealerships.
“We do about $45 billion a year, have about 21 ,000 U.S. car dealerships that we have a relationship with.”
0:14:48 · Sathish Muthukrishnan · Auto lending
Sathish Muthukrishnan says 90 percent of Ally's Invest customers come from its bank deposit customers.
“And 90 % of our invest customers come from our bank deposit customers.”
0:15:55 · Sathish Muthukrishnan · Ally Invest
Ally's one-click Invest account opening required three pages because of FINRA regulations, according to Sathish Muthukrishnan.
“We said one click, but for FINRA regulations, we had to go through three pages.”
0:15:55 · Sathish Muthukrishnan · Account opening
Sathish Muthukrishnan says Ally created its own technology operating model, nicknamed Adam, that embeds compliance staff in requirement writing.
“we created our own technology operating model. We called it Allies Technology Operating Model.”
0:17:34 · Sathish Muthukrishnan · Operating model
Ally has an internal technology lab that tests new technology with a business outcome and a stakeholder from outside technology.
“we also have what we call an allied technology lab.”
0:22:52 · Sathish Muthukrishnan · Innovation lab
Sathish Muthukrishnan says an Ally board member asked him about ChatGPT on the morning of a board presentation, before he had heard of it.
“One of the board members literally that morning is like, did you hear about chat GPT?”
0:22:23 · Sathish Muthukrishnan · ChatGPT · 2022
Sathish Muthukrishnan says Ally's leadership team spent three days in Seattle meeting Amazon and Microsoft after Microsoft's investment in OpenAI.
“we need to go sit in Seattle. So my entire leadership team and I, in November 2020, not the ideal time to visit Seattle when it's really cold.”
0:23:55 · Sathish Muthukrishnan · Microsoft partnership · 2022
Ally AI removes personally identifiable information, logs transactions and restores the information afterwards when connecting to an external large language model.
“It would remove PII. It would keep track of all of the transactions going in and out. The ability to rehydrate PII so it'll have the context for Ally consumers that are using it.”
0:25:07 · Sathish Muthukrishnan · Ally AI · 2023
Sathish Muthukrishnan says Ally built its Ally AI platform in two months.
“Very simple. We built it in two months.”
0:25:07 · Sathish Muthukrishnan · Ally AI · 2023
By early 2023 Ally had a use case running on Ally AI using GPT-3.5 on its own private VPN on Microsoft Azure.
“by beginning of 2023, we had a use case that was working on Ally AI using 3.5 GPT running on our own private VPN on Azure.”
0:25:41 · Sathish Muthukrishnan · GPT-3.5 deployment · 2023
Sathish Muthukrishnan says Ally uses different large language models for content creation, code generation and risk assessment.
“Our content creation LLM is different from what I use for code generation. and that is different for what I use for risk assessment.”
0:27:26 · Sathish Muthukrishnan · Multi-model AI
Sathish Muthukrishnan says a company with fewer than 20 people can quickly reach $100 million in annual recurring revenue.
“a company with less than 20 people can quickly build something and can very quickly have a $100 million ARR”
0:29:19 · Sathish Muthukrishnan · Fintech competition
Sathish Muthukrishnan says Ally customers with savings buckets save twice as much as they normally do.
“people save twice as much as they normally do when they have a savings bucket”
0:33:17 · Sathish Muthukrishnan · Savings buckets
Ally uses generative AI to create five-to-six-line summaries of its financial education articles on its Conversationally platform, which increased engagement.
“we simply took that entire article and we created a five to six line bullet of this is what the article is talking about. And it was done by generative AI.”
0:36:04 · Sathish Muthukrishnan · AI summaries
Ally has a product called Plans that shows customers how to pay down debt across multiple accounts.
“We have rolled out a product that is called Plans that essentially tells you how you can pay down your debt across your multiple accounts”
0:35:24 · Sathish Muthukrishnan · Debt tools
Sathish Muthukrishnan paraphrased a blog by the founder of Anthropic as saying a world of geniuses in a data center can be accessed through an API.
“you now have access to a world of geniuses in a data center that you can enable through an API.”
0:38:22 · Sathish Muthukrishnan · AI outlook