Is Your AI Strategy Missing the Unglamorous Work? | Aniket Desai, Revecore

Aniket Desai’s career has taken him across networking, cybersecurity, cloud, machine learning and now enterprise AI. Having recently moved into healthcare as Senior Director of AI Systems at Revecore, he is building the company’s AI team from scratch, with a focus on delivering AI solutions at scale that are well governed, simple, predictable and explainable. It is the second time he has done this. 

At Capital One, he built AI capability within an internal organisation focused on global finance, a function that previously had no clear AI vision or strategy. That experience, together with his time on Capital One’s executive leadership development programme, has shaped a practical view of AI: it is a powerful tool, but only as strong as the data, feedback loops and business understanding underneath it.

Why is revenue cycle management ready for AI?

Aniket Desai explains how the complexity of US healthcare billing makes the industry a natural fit for AI, and how he chooses which problems AI should solve.

“Healthcare in the US generates huge amounts of documentation. Every symptom, treatment and recommendation has a code, and a three-day stay can produce a medical record of 300 pages. The hospital then has to bill the insurer, which will often deny or underpay, and this happens at the scale of millions of claims. Revenue cycle management companies exist to help hospitals recover that money, but until now they have hired a tonne of people, often overseas, to work through the messy documentation. If you have millions of claims, how many people do you hire, and how long does it take?

That is where AI comes in. This industry is made for AI, and I like the mission: hospitals recovering the money they are owed so that they can provide better healthcare. When choosing problems for AI, I look for two things at once. The problem has to be useful to society, and it has to lend itself to being solved by AI. Not all problems are solvable by AI, and not all should be. A student should not use it to write their homework. I am now building our entire AI team from scratch, delivering solutions at scale that are well governed, simple, predictable and explainable.”

Why do most AI initiatives fail to create business value?

Aniket Desai argues that organisations misunderstand what AI actually requires. He also suggests that the unglamorous work is where success is decided.

“AI is basically another technology stack, like big data, serverless and cloud computing before it. A lot of people approach it as the answer to everything, or as a typical toolbox, and it is not. What is specific to AI is that it requires constant feedback. It is not only tools and algorithms but data, and not just any data. You need labelled data. To teach an algorithm the difference between a cat and a dog, you feed it thousands or even millions of images, and every one has to carry a label. It is much like how we teach our children, constantly labelling things for them.

If you do not have processes where people constantly label things and give feedback on your AI outputs, it just will not work. Everybody is excited about the technology, but this is the grunt work that nobody wants to do. That is why I have read that 94% of AI initiatives really do not go anywhere. The money is simply wasted. It also takes data sophistication. Big tech solved data collection 15 or 20 years ago, while a lot of companies have not even started. If big tech are the PhDs, some companies are in kindergarten.”

What changes when AI moves from prototype to daily use?

Aniket Desai explains that daily reliance on AI changes user behaviour, and why blind trust is as risky as blanket scepticism.

“People’s behaviour changes drastically. You can take away the cognitive overload of documenting, using Control F and doing calculations in Excel. The downside is that AI will eventually hallucinate and will not be accurate. Even a tiny error rate adds up at a scale of 10,000 per day. Sometimes that is fine. If one of Netflix’s 50 movie recommendations is wrong, no harm done. Sometimes it is not okay. It depends on what is being predicted and what happens if that prediction goes wrong.

That is why active participation is necessary. We should not have blind faith in what AI tells us. It is a statistical machine making statistical guesses, and it will eventually be wrong during your workday or your week. Some people are completely sceptical, while others treat AI as the ultimate god, and talk of superintelligence is a lot of hype. It is a great tool, but we have to use it responsibly, checking outputs and giving constant feedback.”

Where does agentic autonomy end and human accountability begin?

Drawing on driverless cars, Aniket Desai argues that the real question is the nature and fidelity of each decision, and that accountability remains unresolved.

“I saw Waymos driving all over San Francisco this summer, and that is autonomous AI. At every point, it is making a decision. But when people say they have an agentic system, we have to understand what that means. You are letting your AI autonomously execute decisions on your behalf. If a research agent gets a report wrong, my report is wrong. If an agent gets a $100,000 transaction wrong, I am taking a big risk. The issue is fidelity. If a customer withdraws $100, the machine has to give them $100. That is not an AI problem. That is deterministic software, and software solved it 40 years ago. The only real example of autonomous AI I have seen is the driverless car.

Accountability is a huge problem. On Capital One’s executive leadership development programme this year, we looked at a scenario where a car chooses to hit a person who jumps into the street rather than swerve and kill its passenger. With a human being, you can hold the person accountable. With AI, can you hold an algorithm accountable, or the company that produced it? Even an open-sourced algorithm still solves for some utility function. I do not think that problem has been solved, and we do not have a framework to handle it right now.”

What does it take to be a genuinely AI-ready company?

Aniket Desai describes the foundations companies need, the skills engineers should build, and why the biggest question for senior leaders has not changed.

“I do not think anybody knows what is going to happen in the next three months, let alone three years. But an AI-ready company should understand that AI is a tool, and the tip of the pyramid. For the pyramid to stand, you need a strong base of data engineering, process and a constant feedback loop, plus active users who know where AI is likely to make mistakes and report them faithfully. Then apply it responsibly to the right problem, because otherwise it will just burn a lot of token usage. 

I also believe in owning your internal AI tools. Specialised pieces such as document OCR can be outsourced, but outsourcing the tool means outsourcing data, and then you lose your intellectual property.

The most important skill for engineers is the ability to build an end-to-end system and connect it to the business, and the trait I look for is curiosity. 

What is one question every leader should ask themselves today?

Aniket Desai argues that the question every senior leader should ask has not changed with AI.

As for the big question every senior leader should be asking, it has always been the same: what is my business, how am I going to get my leverage, and what is my strategy? AI is not going to change any of that. It is a boring answer, but not every senior leader has that strategy-level focus.”

Join the Conversation: The Ortus Club’s Executive Network

As Aniket Desai has shown, challenges like data foundations, feedback loops and AI accountability are rarely solved in isolation. Leaders at this level increasingly rely on peer-level dialogue to compare how others are governing, scaling and questioning AI in practice.

His view that AI is a tool built on a strong base, rather than a shortcut, reflects a broader reality: leaders cannot rely on hype or a one-size-fits-all toolbox. The most effective executives get into the details and test their thinking with peers.

At The Ortus Club, we host curated executive roundtables that bring together senior leaders facing these exact challenges. Step away from the hype and join the kind of open, high-value conversations that connect AI ambition to the foundations needed to deliver it.

Frequently Asked Questions

Q: Why do most AI initiatives fail to deliver business value?

A: Aniket notes that organisations often treat AI as the answer to everything rather than as another technology stack. AI also needs labelled data and constant feedback from the business. Without processes for both, initiatives stall, and he cites a figure that 94% of AI initiatives do not go anywhere.

Q: What is labelled data and why does AI need it?

A: Labelled data is data that carries a description of what it shows, such as an image marked as a cat running. Algorithms learn by being fed thousands or millions of labelled examples, and the labelling does not go away for any company.

Q: When does agentic autonomy make sense?

A: It depends on the nature of the decision and the cost of getting it wrong. An agent preparing a research report carries little risk, while one executing a $100,000 transaction carries a lot. Where a 100% accurate outcome is required, deterministic software is the right tool.

Q: Should companies build or outsource their AI tools?

A: Aniket advocates owning internal AI tools. Specialised pieces such as document OCR can be outsourced, but document intelligence processing should stay in-house because it is tied to the company’s own data. Outsourcing the tool means outsourcing data and losing intellectual property.

Q: Which skills matter most for engineers as AI takes on more technical work?

A: The ability to build an end-to-end system and connect it to the business, supported by strong foundations in computer science, distributed computing, networking and security. Curiosity and a willingness to learn and experiment are the traits Aniket looks for.