AI's Impact On Professional Services: An Industry Veteran's Perspective
An interview with Neil Fletcher, an international tax expert and consulting industry leader.
Neil Fletcher has over 30 years of experience in M&A and international corporate tax advisory, including 16 years as a Tax Partner at PwC and latterly as a Managing Director at Alvarez and Marsal. His recent work includes advising on multi-billion dollar acquisitions and working in the Fintech and cryptocurrency sectors.
How is AI infiltrating the consulting sector today? Is this mainly via formal adoption of Enterprise AI tools, or through use of public models like ChatGPT?
In both ways. The public models are rapidly increasing in capability, and becoming seriously useful. Does that mean people are feeding confidential information into open source models? I don’t think so.
Personally, I’m using both public and custom models. When I upload the OECD Pillar 2 (the international global minimum corporate tax framework) documents into a general LLM, the accuracy of answer is usually less than fifty percent initially. I then increase the level of instruction until eventually it finds the answer. Even then I don’t necessarily trust it, as it does beg the question as to whether the model is only reflecting back the specificity I am entering into it.
Overall, it's very useful in research and generation, though human expertise is still vital.
Recently, people have come back to me suggesting this or that tool designed more specifically for what I want to do. You can see the benefits of these tools. It's beginning to demonstrate the investment case more clearly.
What does the business case for AI look like for a large scale consulting firm?
The easy answer is, as the accuracy goes up then costs go down. Any professional services firm is in danger of being on the outside looking in, always chasing the AI tail as the technology marches on. The current focus of Agentic AI seems to be on relatively routine workflows.
For consultants, it's probably a case of identifying use cases today to demonstrate where it can go in the future.
It may be worth reflecting on what professional services means in an AI context. A lot of people see the ‘secret sauce’ of professional services as being specialist knowledge. However, this may develop into leveraging enormous data sets to innovate new products and services. Pure knowledge management and data retrieval could be self-limiting.
How do you see AI impacting the professional services industry going forwards?
The number one question from junior people in professional services organisations is, Am I still going to have a job? I can see how A2A (Agent To Agent) workflows are going to have an impact. There are some products and services that will either no longer be needed, or will be performed by more relevant organisations.
As a professional of any level of experience you want to be there to shape what comes next. That will only happen if you’ve acquired the foundational knowledge on how these tools work.
Lawyers, accountants, engineers et cetera all need training on how to work with AI, but I don’t yet see a very structured way of building capability. There doesn't seem to be a better time to invest in acquiring these skills than today.
You recently commented on an article on LinkedIn describing AI as an existential threat to the Big 4 consulting giants including PwC, where you were a partner for 16 years. The article concludes “Do you still need a Big 4 firm to compete, or do you need better prompts?" Your response was “If there are no Big 4 Partners doing the Big 4 Partner job then how can “Act as a Big 4 partner” prompt work?” Can you expand on that?
What’s underpinning my comment is, they're perhaps diminishing what an experienced professional, such as a Big 4 partner does.
The linear process of how to do things faster or cheaper is not even debatable. The point is, how do you move forward so you can realise the value of your work for clients? This is a forward-looking aspect that a backward-looking prompt that says “Act as a Big 4 partner [has historically acted]” can never capture.
The delivery model doesn’t determine the outcome. Why would you just produce the same reports quicker, now you can do all these new things?
I’ve been reading about the concept of ‘operating rails’, whereby AI platforms offer to take over certain categories of decisions on customers’ behalf with bounded liability, based on their ability to predict outcomes using big data. Is that the sort of thing you have in mind?
Absolutely. For instance, risk transfer is fundamental to the financial services industry, and it can only be done if you can put numeric values on things. The pensions industry, to take one example, became widely accessible with the advent of data on morbidity, and so on.
Pattern recognition might have been done by a human super-expert in the past. Now it’s done with AI. Professional services firms have this huge flow of information going through their people and systems. The data itself is the real source of value, particularly when it’s curated by domain experts.
To take it to the next level, we need to be looking to create new, trusted tools using this data. You can only do that by bringing together people from different disciplines, which is why I find the current debate on LinkedIn so valuable.
What value do you personally get from LinkedIn?
I’m steeped in the international tax environment and I’ve written articles on LinkedIn that dive into the topic in some depth. I also participate in ongoing dialogue on international tax fora.
The responses I've had to technical articles on LinkedIn have predominantly been from other domain experts such as in-house tax, tax advisors and academics. It tends to be similar people that respond every time.
These days I’m more likely to participate in broader discussions, albeit from a taxation perspective. LinkedIn is doing a great job of enabling a public discourse about AI. It’s a huge topic that a lot of people have on their minds. People with very different backgrounds and agendas are coming together and having respectful, informed discussions in a way that goes beyond trying to sell your services. A lot of high quality content is being shared.
I’m in an ongoing dialogue with a variety of people on LinkedIn, including in the US and Asia. Where else would I be able to meet and have one-to-one conversations with former leaders at Google, for example?
You tend to engage on LinkedIn at least once a week. How much of this is premeditated, versus spontaneous engagement for the joy of joining the conversation?
It’s a mixture. I’m not generally thinking about impressing potential clients—my engagements tend to be bespoke and relationship driven.
In my experience, on LinkedIn you are able to get much broader inputs and a wider perspective. There’s a willingness to listen to people outside one’s own domain, with less emphasis on protecting a particular patch. AI is coming at a frightening rate of knots. Participating in the conversation builds your confidence, as many people are facing similar challenges, and are open to sharing them.
Which types of professional services firms do you think will win, or at least survive, the race for AI?
One route is about compliance, productivity and workflow automation. The other is about product and service innovation. It’s about moving forward with a mindset that’s open to both.
I believe the traditional professional services model still has a vital place, based on these firms’ proficiency with data. It’s not purely about defensive action. For example, AI will enable niche consultancies to invent new product categories and become the market leader, without needing to be the biggest firm on the planet.
What’s your advice for those working in the consulting industry today, or looking to enter it?
These new products and services will need people to build them. Learn the tools now, so you’re around to take it forward.
There will always be a need for the human element. Notwithstanding the speed of developments, human oversight of the models is essential as there will always be exceptions and imperfections in the data.