What changed
Michael Rangel founded Novo in 2016 and served as its CEO for more than nine years, helping transform it from an early-stage concept into a financial platform serving small businesses. He stepped down as CEO in April 2025 but remains involved as Founder and Board Member. Novo is not a bank; banking services and FDIC-insured deposits are provided through partner Middlesex Federal Savings. The platform integrates with more than 40 business tools, including QuickBooks, Stripe, Shopify, Square, and Xero, and uses AI models from OpenAI, Anthropic, and Google for processes such as onboarding and expense categorization.
In an interview with Unite.AI, Rangel gave a grounded account of what AI has changed inside a financial platform built for independent businesses. For a business reader trying to decide what to do about AI at work, the most useful parts are not the predictions. They are the specific operational problems he says AI now solves, the accuracy threshold his team had to clear before trusting it, and the limits he still sees.
Refer & earn: 30% of a client’s first month, then 10% of every month after that, for up to 24 months per client. See the partner programme →
The problem AI is actually solving
Rangel describes the original problem Novo set out to solve as an operational one. A $1,000-a-year business, a $100,000-a-year business, and a $1 million-a-year business can create a surprisingly similar operational burden for a bank, because onboarding, compliance, fraud, security, transaction monitoring, and customer service are required regardless of how much revenue the customer generates. That meant banks had to build large operations teams to support customers who, individually, were not always that valuable to them.
Ten years ago, he says, the belief was that software could change that equation. AI is now taking that idea further. He gives concrete examples: a compliance specialist can use AI to summarize information, process documents, and draft reports or responses, allowing one person to handle more than ten times the volume they could before. Customer service teams can let AI handle routine questions so people spend their time on problems that require judgment. Fraud models can identify potential issues earlier and help recover losses faster when something goes wrong.
That is a useful frame for any business owner. The question is not whether AI is impressive in the abstract. It is whether it removes a specific operational burden that scales with volume rather than with revenue — the kind of work that consumes headcount regardless of how big the customer is.
The accuracy bar that made it usable
The most concrete example in the interview is transaction labeling. Rangel says that historically, getting these models to work well required a huge amount of training data, and even then you might get around 80% accuracy. That is not good enough when you are dealing with someone’s financial information, because you still have to check everything.
By using a smaller amount of Novo’s own data, some model distillation techniques, and frontier models when the systems are not confident, he says accuracy has been pushed well above 97% and closer to 99% recently. Once you can accurately understand what every transaction represents, he argues, you can build much more useful financial tools on top of that data — expense dashboards, basic FP&A, and helping a business separate money for things like payroll and vendor payments so it does not accidentally get spent elsewhere.
This is the part worth copying into your own evaluation process. The difference between 80% and 97% is not a marginal improvement. It is the difference between a tool that creates work (because everything must be checked) and a tool that removes work. Rangel also notes that LLMs are surprisingly good at classification and clustering, which means a business’s transactions can be turned into a personalized view of its finances without the engineering complexity that used to be required.
For a business owner, the practical takeaway is to ask vendors for the accuracy figure on the specific task you care about, and to ask what happens when the model is not confident. A system that escalates uncertain cases to a stronger model or a human is more trustworthy than one that guesses.
Where human expertise stays
Rangel is careful about the limits. He notes that AI is moving incredibly quickly, but the pace looks very different depending on where you sit. If you work in technology in San Francisco or New York, you probably have a very different experience of AI than a small-business owner who is simply trying to run their company.
He says a lot of the problems small businesses face are already within AI’s reach, but the bigger challenge is that the technology is moving faster than people can learn how to use it. That is why he is interested in what companies like Anthropic are doing with products designed to give less technical users a starting point. He also mentions April Tax as an example of a product making things that once required a lot of time and specialized knowledge easier for a small business owner. His conclusion: we are not at a point where AI can put your business on autopilot, but if you have ever lost a weekend preparing for tax filing, getting that time back matters.
That is a realistic position. The value is in reclaimed time and reduced burden, not in abdication.
What an intelligent financial platform looks like
Rangel’s description of a genuinely intelligent financial platform is worth reading closely, because it sets a test you can apply to any tool. He says the best banking experience you can have is still talking to someone at your bank who knows everything about your business — your history, what you are trying to accomplish, and what you might need next. Software has always been an enabler of that relationship, but it has not replaced it.
For him, an intelligent platform starts to feel like you have reduced the number of people you need in the room to manage your business. You should be able to ask a question, get an answer, see the right information, and receive useful guidance without having to figure out where to look for it yourself. The important part, he says, is that you trust the answer.
He adds a line that captures the competitive bar: at Novo, they sometimes say they are competing with the sidewalk. Going to a banking advisor is still a great experience — there is no app to navigate, someone knows you, and they can guide you based on your business. What AI gives them is the chance to bring some of that experience to an underserved group of businesses at a scale that was not possible before.
The virtual CFO question
Asked whether AI could eventually function as a kind of virtual Chief Financial Officer — continuously monitoring finances, identifying risks, and recommending actions — Rangel says yes, confidently, because it is something Novo is already building. Earlier this year, the company released AI Agent infrastructure specifically to move toward that vision.
But he is explicit about where the difficulty lies. The hard part is not getting a model to answer a question. It is building something people can trust with their financial lives. He says Novo learned that firsthand: the first beta cohort was good, but not good enough to say the experience would not eventually break someone’s trust. So the team kept pushing, improving how the system learns, how specialized agents work together, and how they use the information and tools available to them.
What they built is essentially a banking-specific system of agents, with different agents handling things like transactions, fraud, cards, and ACH, each with their own tools and knowledge. He says the results are now genuinely good, and that they did not want to rush something like this just because AI is having a moment, because that would be unfair to the customer.
What to check before you act
If you run a business and are deciding what to do about AI, Rangel’s account suggests a short list of questions.
- Is the task volume-driven or judgment-driven? The clearest wins he describes are high-volume, low-judgment tasks: summarizing documents, drafting responses, handling routine customer questions, labeling transactions. Judgment stays with people.
- What is the accuracy on your specific task, and what happens when the system is unsure? The 80% to 97%-plus jump is the difference between a tool that adds checking work and one that removes it.
- Can you trust the answer? Rangel’s own test for an intelligent platform is that you can ask a question and trust the response. If you cannot, the tool is not ready for your finances.
- Does it reduce the number of people needed in the room? That is his practical definition of value, and it is a better measure than feature lists.
- Is the provider willing to wait? Novo held back its first beta because it was not good enough. A vendor that ships because AI is having a moment is a different kind of partner.
None of this requires a large budget or a technical team. It requires knowing which operational burden you want removed, and holding the tool to an accuracy standard that makes it usable rather than merely interesting.
Disclosure
Faida Workforce advises organizations on workforce and technology decisions. This article is editorial and is not a recommendation of Novo or any other vendor. We have no relationship with Novo to disclose beyond covering it as a subject.
Sources
By Faida Workforce Editorial. AI tools assist in production; a human editor is accountable for the published article.
30% of their first month, then 10% every month after that
Refer a business to Faida Workforce and you earn 30% of the client’s first paid month, then 10% of every month for up to 24 months per client — rising to 15% from the accrual after you pass 10 active clients. Paid by PayPal, batched on the 1st.
10 Pro clients = $2,598.70 over 24 months — $3,748.13 at the 15% tier.
Figures are potential earnings worked from the published rates, assuming every referred client pays for the whole cap — not a promise, not an average. The $1 trial is never commissionable.
