AI at Dancerace: Playing a Different Game

It’s taken me time and experimentation to develop some clear thinking. So here are my thoughts about the smart way for a business like ours to approach AI – and the practices and processes we’ll apply at Dancerace to get the most from the technology.

1. A game-changer, but not in the way many see it

Working with AI tools is not new to us. We’ve been doing it for years on a practical, day-to-day level in our development systems, service desk platforms and so on. But it was agentic AI that really turned our heads. That’s the genuine step change for the kind of work we do.

For much of the business world, LLMs have been transformative for administrative work, advanced chat, legal review, design, communications. Great – but that’s not us. For us, the most impactful thing agentic AI can do is supplement and accelerate the real-world software engineering and service tasks our people perform every day. Always with our people guiding it.

That’s a game-changer. But it’s a different game from the one most of the market is describing.

Will we throw away learned experience and knowledge around security, resilience and robustness? No. Will we do everything we can to use AI to optimise how we work? Absolutely.

And is this transformational? Yes. But not in the sense that we’ll completely change our business. I don’t want AI to transform Dancerace into something it wasn’t before. But enhance, accelerate and liberate Dancerace – please, yes.  

The real impact agentic AI can have in a business like ours is amplifying what already works.

Note what this implies. We are keeping what’s good. We want to do more of it. We’re using AI to free us to do more of the human things that we have built our success on – making those judgements that no AI can do for us.

Interestingly, this is the commercially smart approach. For us, given our role, in our industry. Overreacting one way or the other would be the failure, and the risk.

2. What we’re not doing: overreacting with blind enthusiasm or fear

We were never guinea pigs for an unmanaged, barely understood technology in the past, and I’m not volunteering us now. We’ve always gained knowledge about a technology first. I like being on the bleeding edge – but only the bleeding edge of things we understand, not things we don’t.

We wanted to see what governments and the industry would do to protect users and businesses from the downsides of AI. We’re now seeing that discussion rapidly become more urgent.

So we were never going to race in and embrace every aspect of AI with heedless enthusiasm. A large amount of how AI works remains opaque, and to me that’s a reason for moderation. But importantly, it’s not a reason for paralysis. The new possibilities are too interesting.

Equally, we have never avoided AI on principle, as if it should always be shunned for fear of what it might do. That’s not the way we operate as a business and as people. We have incorporated numerous AI tools into our workflows, from general business admin tasks to code reviews and future product specifications.

Neither blind enthusiasm nor fearful paralysis is viable. Nor desirable.

Our clients are banks, lenders and business borrowers. They value judicious judgement and continuity, and so do we. Accuracy and security are critically important to us. We take our responsibilities seriously because, frankly, we have to. We work for banks and lenders who operate in a regulated space, and their business continuity is vitally important, not least to their thousands of customers – businesses who rely on a dependable flow of finance available exactly when they need it.  

Using AI well requires human judgement at every stage.

3. What we are doing: focusing on what matters most

When something important changes, we pay attention. AI continues to change and we keep watching. It was agentic AI that really caught our attention.

Traditional AI – if you can call it that without sounding ridiculous – required user input and struggled to generalise and adapt. It responded to a particular input and generated new content from that. Agentic AI processes information and acts autonomously. In software development, for example, instead of seeing AI as a code-completion tool, it can be used to create a new feature by itself. The human in the loop then has to assess and improve the solution and the code used. But the point is, agentic AI has the capabilities to do tasks like this.

That’s important.

We’ve been adopting intelligent tooling for years as a matter of sensible business practice. Machine learning and AI features have arrived inside our development systems, our knowledge bases, our ticketing and service desk platforms. As they landed, we embraced them fully. Agentic AI has impelled us to be that much more conscious about our decisions and the strength of our embrace.

In short, we’re being serious about AI. That cuts both ways.

The fact is, these AI features and tools are not what we built our business on. We have 50+ people carrying 30+ years of banking and lending experience. That’s where our success comes from. The AI tools? They save us time and allow us to explore new ways of doing things – but without changing the formula.

We focus on that critical knowledge. It’s what matters most, and no AI tool can replace it. What an AI can do is help us get more out of that knowledge. Much more.

4. The risks and costs are real

AI will never be as cheap as it is today.

AI today is massively subsidised by investors, to the point where the costs we pay don’t remotely reflect the costs of the infrastructure and the compute. It’s financially unsustainable. Which is of course a serious risk to businesses incorporating AI into their workflows.

If we build deep dependency on AI at today’s prices, we’re exposed to the moment when pricing inevitably catches up with the real economics. That’s not just an awkward internal disruption; it’s a commercial risk. A vendor whose AI cost base blows up is a vendor whose pricing blows up.  

So our strategy is long-term: use AI as well as we possibly can without locking ourselves into tools and ecosystems that are financially unsustainable.

We therefore deliberately keep our options open. We are not architecting ourselves into a single provider’s ecosystem, and we are actively planning for a future where the economics of AI look different from today’s.

Supplier lock-in, supplier risk and the AI bubble are all named concerns internally. We’ve priced that risk in rather than ignored it.

Beyond the risks are the costs. Embracing cheap AI and running with it across every part of the business may drive rapid scaling, but rapid scaling has its costs, both business and human. There’s the cost to the business, which then needs to be recovered, and a cost to the time and efficiency of our existing people, absorbed into training and ongoing support. Plus a broad psychological impact from rapid upheaval and disruption. We will avoid this particular temptation.

AI itself has its own costs beyond the financial. Everyone now knows just how extraordinarily energy-hungry data centres are. Why? Because the way LLMs work is to crunch quite unbelievable amounts of data, and that takes vast amounts of compute, and compute requires vast amounts of energy. And cooling. There’s a direct environmental impact, not only from the emissions much of that energy generation releases but also the water that data centres need and the land they occupy.

There’s a further human cost to using AI. When people rely too much on AI, their skills can atrophy, and so can their desire to collaborate. Accepting skills loss to gain some efficiencies sounds like a poor strategy to me.  

These costs and risks are unresolved. We’re going to see big changes from today’s status quo and it’s not clear what the landscape will be like when the dust settles some years from now. We’re not betting the farm on a particular outcome.

Our restraint is a feature of our clients’ supply chain rather than a lack of ambition on our part. It’s the type of approach they rely on us to take. And I believe it’s the commercially savvy one.

5. Where we can win

AI should amplify the goodness of what already works.

Fifty people. Deep knowledge. At Dancerace, the aim is to use those same people, with that same knowledge, and amplify their efficiency. Let’s get more out of what people already know. The sequence matters: first build knowledge within people, then use AI to expedite the time-consuming tasks around it. Not the other way around. I feel strongly about this.

A word about time. There’s an important difference between time-consuming tasks and tasks you spend time on. Time-consuming implies something takes you away from what you’d rather be doing and actively eats up the time you have available to do what is important. Spending time implies agency and intent.

Both tasks are necessary. AI should be eating the time-consuming kind. That’s a wonderful prospect.

What does it mean in practice? It means the areas where AI is better. People are not well suited to sitting and staring at data points, monitoring dashboards and alert streams. AI is. I want it intelligently notifying us about frequently occurring issues and making suggestions about tuning alerts – so our people are applying judgement rather than watching screens.

We will do less of things humans aren’t good at, where AI wins. And we will do more of what humans are better at. Judgement. Applying that deep knowledge.

There’s clear opportunity here to do more with what our people have, to optimise our learned experience. AI is a gift. It can remove the distractions, the inefficient tasks, the time sinks and the barriers.

At Dancerace we’re a small team. When you’re a small team, something’s always got to give. This is the opportunity to stop that being true. We will use AI to secure our product roadmap, not double it. We will buy back enough time to improve the resilience, robustness and scalability of what we do.

What does more capacity actually mean to our clients? A significant uplift in what our team get done, real reduction in workload strain, time returned to doing things properly in every area.

We’ve always had more ambition than capacity to deliver. Changing that is the point.

We want to achieve our ambition level, not our resource level. That’s what I want out of this, more than raw output.

6. Our playbook for using AI at Dancerace

We’ve established an approach to AI that we’re excited to be implementing.

First, AI is coin-operated. It needs a person to feed it the pennies. In other words, there’s always a person in the loop. A human reviews and understands the inputs and outputs of the AI when performing a task of any type. Without that, AI needlessly does things that nobody actually needs, simply because it can. That way it becomes destructive.

Second, people are accountable for the consequences of the decisions they make using AI. You can never blame the AI.

Third, we will use AI efficiently. Not only to reduce the potential future costs, but to reduce the environmental impact. We will avoid unnecessary use. We will also prefer providers demonstrating genuine ESG effort.

Fourth, we will regularly review the impact. We’ll run an annual ESG and environmental impact review so we can see the gains and the losses. At the same time we will review our tools and their impact on our people and a proper work cadence.

Fifth, we will choose to use AI in three tiers of tooling:

1. A core AI system for chat, work improvement and agents.
2. AI built into our workflow systems – service desk, development project management, documentation and knowledge bases, meeting transcription and note-taking, project planning.
3. AI integrated into specialist engineering systems – source control, pipelines, security and monitoring, infrastructure as code, development environments.

We're also training and guiding each team on use. This wasn't switched on and left alone.

Sixth, we’ll continue to be on the bleeding edge, but only of things we understand. That includes understanding the unintended consequences and unanticipated side effects.

We’re taking this approach because it’s the commercially smarter long game. It also happens to be the responsible one. AI adoption will be a good thing for businesses if they do it properly. It’ll be a terrible thing for businesses if they do it badly. By adopting the approach outlined in this paper, we seek to make Dancerace better than before – even if it means playing a different game from other software providers.