I recently gave a keynote talk at the Innovate UK / ESRC UK National Showcase: Next Generation Professional & Financial Services event in London, which brought together funded innovation projects in fintech, legal, accounting and insurtech with other practitioners in these sectors to discuss new ideas and opportunities.
My talk was about the evolution of services in the algorithmic era, with predictions about how the idea of programmable organisations could transform these interconnected sectors and the services ecosystem they operate within.
(You can tell the video was not AI-generated because of the mis-handling of typographical ligatures on the slide titles 🤪)
Professional services are a particular strength for the UK’s economy, and have historically played a key role in maintaining a trusted business environment; but they can also act as a force multiplier for business innovation beyond their own ecosystem.
Law, accounting rules, underwriting algorithms and financial regulations are all essentially forms of human software, and each has its own controlled vocabulary within a narrow knowledge domain. This means they are all programmable, and therefore highly amenable to the kind of agentic AI that is coming into focus right now.
But they are also very human, and reliant on trusted relationships and advice. One direction of travel from here is towards the maximalist tech-driven dream of trustless systems, zero knowledge proofs, etc. But we can also use our tech tools to orchestrate and automate the basic workings of the professions to create more time and surface area for trust, enabling the best advisers to combine agents and services to create truly unique solutions and customer experiences for their clients.
The scope for innovation in these inter-connected sectors is huge, but it will not all come from the big firms themselves, but also from startups, scale-ups, platforms and innovation projects. So it makes sense to think in terms of a connected ecosystem and the new capabilities and systems that could enhance it.
What are the missing building blocks that will unlock new value in the future, just as double entry bookkeeping did when it (re-)emerged in C13th Florence? These new capabilities could be the foundation for combinatorial innovation, where we combine services across the sectors to deliver greater business value.
We also had some fun at the event mapping funded projects and their capabilities onto a connected ecosystem canvas, and we will share more details of this future-scanning technique next week.
Professions getting into gear
There are already several key uses cases for AI being pursued in professional and financial services firms, including:
- concierge bots, co-pilots and helpers both for employees within firms and also direct to consumer to help them navigate complex processes;
- centaur teams – i.e. teams made of highly autonomous high-agency people who use various AIs like musical instruments to deliver complex work;
- knowledge synthesis – in some fields, there is too much prior art for any one person to know how it all fits together, but AIs can detect patterns and insights that are invisible to many of us;
- agents and automations that turn manual processes or models into digital service components that become part of a firm’s overall digital services platform;
- ‘vibe coding’ – if you can imagine it, you can build it, so that gives us incredible freedom to work with our clients, customers and partners to co-design the very best experiences and outcomes, and then use our service platforms and AIs to assemble it;
- smarter audit, compliance and risk management, which are still too manual today. In a platform organisation we can bake them in to the way the system works to turn these control functions into real-time features; and,
- using AIs as thinking partners to help structure your thoughts or create plans, and leaders and managers use decision support systems to query and challenge their own biases or blindspots.
A recent study in Minnesota Law, AI-Powered Lawyering: AI Reasoning Models, Retrieval Augmented Generation, and the Future of Legal Practice, hinted at the promise of combined human/AI centaur teams in handling legal process work:
“Our findings show that AI assistance significantly boosts productivity in five out of six tested legal tasks, … yielding statistically significant gains of approximately 38% to 115% and o1-preview increasing productivity by 34% to 140%, with particularly strong effects in complex tasks like drafting persuasive letters and analyzing complaints.”
And rather than just rely on publicly available LLMs, many professional services firms are already trying to build their own, more tailored and reliable AI systems.
Kennedy’s recently talked about their approach to neuro-symbolic AI that seeks to move beyond the probabilistic approach of LLMs by adding in structured, expert system models to make more reliable decision on insurance claims, for example.
Unlike pure GenAI solutions that rely solely on probabilistic outputs, Kennedys IQ SmartRisk leverages a hybrid approach, integrating Large Language Models (LLMs) with insurance knowledge modelled using Evidential Reasoning (ER), and Belief Rule Base (BRB) methodologies. Kennedys IQ SmartRisk goes beyond pure language processing, providing insurers with a structured decision-making framework.
It is encouraging to see that more professionals seem to be open to using AI in a controlled and safe way, as this survey of tax professionals in Legal Futures suggests:
According to a survey of over 350 UK tax professionals, AI adoption has climbed from 35% to 40% since May 2024. Even more striking, the number of tax professionals planning to adopt AI has surged from 32% to 49%, while those with no plans to integrate AI has plummeted from 33% to just 8%.
Even relatively conservative functions such as audit and assurance are starting to evaluate the potential of AI to transform how they deliver on their goals, albeit with a healthy degree of scepticism and caution.
And there are entire new lines of business opening up for professional services firms to advise and guide their fellow professionals on risk and compliance challenges that AI brings to the table.
Hard at Work in the Spreadsheet Mill, with apologies to LS Lowry
Agentic AI becoming more accessible
It is likely we will see specialist narrow LLMs or SLMs used in each of the professional knowledge domains, and bigger firms might use optimisation, RAG and fine-tuning to make the most of their own internal data and knowledge. But on top of this foundational AI layer, it is likely to be agents that automate and connect much of the process work that is currently undertaken manually.
The past couple of weeks have seen major announcements in the field of agentic AI.
OpenAI have launched a strategic gambit with their Agents SDK, which could have major implications for enterprise AI, according to Venture Beat:
By providing a complete stack from tools to orchestration, OpenAI is positioning itself to capture the enterprise value created atop its models. At the same time, the open-source approach with Agents SDK acknowledges that even OpenAI cannot innovate quickly enough in isolation.
More recently, Anthropic have teamed up with DataBricks for their enterprise data capabilities to offer narrow domain-specific enterprise AI agents built on top of firms’ own data.
And this week, all eyes are on the Chinese startup Manus with the launch of paid plans on its agent development platform, which looks like a potentially useful way to experiment and learn what is possible.
Tracking AI developments might seem like a full-time job right now, and rather overwhelming; but it is worth remembering that even simple AI tools and tech can enable powerful and delightful use cases.
Luke Wrobelski last week summed up two years of using his Ask LukeW AI assistant to answer 27,000 questions relating to interaction design and related topics, trained on his extensive archive. It’s a lovely tool that points to many simple but equally useful application areas for knowledge retrieval and synthesis in professional services.
Ask LukeW works by using AI to generate answers based on the thousands of text articles, hundreds of presentations, videos, and other content I’ve produced over the years. When you ask a question, AI models identify relevant concepts within my content and use them to create new answers. If the information comes from a specific article, audio file, or video, the source is cited, allowing you to explore the original material if you want to learn more.
There is also an interesting strand of enquiry emerging around how we quantify and measure AI productivity and ROI. When it comes to individual agents, some believe we should measure their output as if they are synthetic workers, as outlined in this interesting piece by Daniel Rodriguez on Measuring the Real Economic Impact of AI Agents:
If AI agents are stepping up as your new workforce, ditch the fluffy tech metrics and judge them like you would any human employee. It all starts with the endgame: Key Performance Indicators (KPIs) that are actually meaningful.
Robust data pipelines turn agent performance into hard numbers — without them, you’re flying blind, tossing cash at a black box and hoping for the best. Set the metrics that matter, then build the systems to track them religiously.
Addressing risks to learning & qualification
Whilst it is exciting to think about the potentially transformative benefits of AI to professional and financial services, this innovation will also bring a host of new risks, from hallucinating LLMs to data leaks, ethical concerns and the uncertain future of the workplace. One worst case scenario is that ageing partners stick around and rely on a combination of lower cost associates or paralegals plus AI to deliver existing outputs at lower cost, but without developing their organisations or innovating.
I am a big believer in automating the boring stuff to allow people more time to do the important stuff, but in professional services this risks taking away the learning opportunities that early-stage career grinding has traditionally played in people development. This concern is not unique to the professions, but it is worthy of attention and creative thinking nonetheless.
In the NYT a few days ago, an opinion piece by Tressie McMillan Cottom articulated the paradox as a response to the idea that AI can help people lacking education to elevate their work:
The problem is that asking the right questions requires the opposite of having zero education. You can’t just learn how to craft a prompt for an A.I. chatbot without first having the experience, exposure and, yes, education to know what the heck you are doing. The reality — and the science — is clear that learning is a messy, nonlinear human development process that resists efficiency. A.I. cannot replace it.
But A.I. is a parasite. It attaches itself to a robust learning ecosystem and speeds up some parts of the decision process. The parasite and the host can peacefully coexist as long as the parasite does not starve its host. The political problem with A.I.’s hype is that its most compelling use case is starving the host — fewer teachers, fewer degrees, fewer workers, fewer healthy information environments.
So, whilst there is an incredible transformation opportunity in professional and financial services, and the possibility that new innovation could unlock business value across the wider economy, we also need to be careful not to throw out what makes ‘the professions’ historically valuable and unique.
Thinking more broadly about the ecosystem within which professional and financial services operate could be the key to identifying missing or important capabilities that could advance these sectors in the future. But the key, as ever, is to strike the right balance between people and tech and seek to elevate and augment human capabilities, rather than just replace them wholesale.
We are looking to interview practitioners who are evaluating or implementing AI within professional & financial services for our research, so please reach out to us if you are interested in a chat