Meet Andrei Smirnov, Head of AI

 · 5 min read

Behind every receipt ANNA matches and every tax nudge you get, there's Andrei Smirnov. As our Head of AI, he's the reason ANNA gets smarter – and the first to admit when it doesn't. We sat down with him to talk about where AI is headed, why humans are harder to fix than models, and what a self-running business might actually look like.

Andrei Smirnov, Head of AI
Phone ANNA app with a credit card
Open a business account with ANNA and get your taxes sorted
With ANNA you get a debit card, automated bookkeeping, a personal payment link, up to 40% cashback and 24/7 customer support

What’s your role at ANNA, and how does your job make a difference to customers?

I’m responsible for our data, machine learning, and AI applications. The part that’s directly visible to the customers is related to AI: whenever ANNA seamlessly solves one of your admin tasks (or misses the point and gives you a not-that-smart reply in chat), you can blame me. We’re currently working on a major update of ANNA – by introducing LLM-powered multi-step reasoning, we aim to significantly improve ANNA’s capabilities and contextual awareness. That update will definitely not go unnoticed.

Another part of my responsibilities is to make sure that our business intelligence is accurate and served on time. This part might not be visible to the customers, but it ensures that all the product improvements and experiments are measured properly, and the product is moving in the right direction.

What’s the most surprising thing you’ve learned about how people talk to AI?

That people are generally unhappy with the current level of AI capabilities. Tolerance for errors is much lower for AI than it is for human mistakes. That may be because humans can understand why other humans make mistakes: they might be tired, overworked, have lost concentration, or be distracted. AI doesn’t have the same excuses.

What’s one everyday admin task you’re secretly proud that AI has taken off customers’ plates?

I like how the bookkeeping documents – receipts and invoices – are reconciled with payments. This was one of the first AI features we introduced, and it was much more fascinating seven years ago than it is now, when LLMs can do this for you, but it still feels like a magical moment. And it’s heavily used for tax calculations.

What are the most common misconceptions about AI?

It might not be a misconception per se, but I think it’s important to remember that many of AI’s flaws are amplified versions of flaws in its training data and, ultimately, amplified flaws of humans. This is true both of biases in AI’s judgments and of its critical thinking.

The good news is that if you’re aware of these flaws, you can quickly assess their severity and address them where needed. Changing humans’ behavioural patterns usually takes much more time.

If you could give the AI one ‘superpower’ it doesn’t have yet, what would it be?

To be honest, I don’t think AI needs any help from me in that regard. Give it ten years and it will, on its own, develop all the superpowers it wants.

How do you make sure the AI feels helpful and friendly, not robotic or intimidating?

Tone of voice matters, and we apply the same guidelines to AI’s tone of voice that we apply to our human customer support. That said, for business-related queries, understanding customers’ business needs and the context of their questions is more important for reducing the feeling that ANNA is robotic. Whatever the tone of voice, if you forget what a customer told you yesterday or a week ago, the experience won’t feel great.

If you fast-forward a few years, how do you imagine AI changing the relationship between small businesses and their business accounts?

The business account will evolve into a capable business assistant and eventually into a self-running business. All the friction involved in running the business will disappear. The business owner’s role will be to decide what they want to build and do and to guide strategic decisions; the rest will be handled by AI.

What did you do before ANNA?

I studied theoretical and mathematical physics and wanted to become a scientist. But then I got really excited about the fascinating advances in AI and NLP (Natural Language Processing) in particular, and decided to leave my PhD studies in the final year to get hands-on experience with the new technology.

Over the next four years, I worked as a research scientist on Natural Language Processing and Automatic Speech Recognition, improving the quality of machine learning models used for customer support automation. And then I joined ANNA.

What’s your favourite thing about working at ANNA?

Freedom to do anything you want to do, and massive support from colleagues in achieving your goals.

What has surprised you most about the job?

The absence of hierarchy and the fact that the company has maintained a flat structure throughout the years I’ve worked here.

Cats or dogs?

Dogs!

Open a business account in minutes

Take the load off with ANNA, the business current account that sorts your invoices and expenses.
Get a business account and a debit card that miaows
We create, send and chase up your invoices
We snap and sort your business expenses
Never miss a deadline, with handy tax reminders