In the late 2000s, we replaced IG's manual back-office settlement process with straight-through processing. Trades that had required a reconciliation clerk to manually match and confirm each end of the transaction started flowing from client to market without human intervention. A process that had employed a significant back-office team became a process that ran on rules, queues, and exception handlers. The people who had been doing the reconciliations moved into higher-value roles in risk, client services, and operations oversight. The business grew without linear growth in back-office headcount.
I've watched this pattern repeat across a long list of functions. Client onboarding. KYC and KYB. Appropriateness assessments. Market data entitlement. FX desk exposure management. Hedging logic. Risk monitoring. Marketing attribution. Algorithmic bid management across Google AdWords and every major paid channel. At Capital.com, we built on everything we had learned at IG and added global buying optimisation, sophisticated affiliate management, automated client acquisition across multiple markets, and an AI-powered trading bias detection platform that identified psychological patterns in trading behaviour — loss aversion, confirmation bias, recency bias — and delivered real-time education to clients to help them make better decisions.
Half the organisation at Capital.com was technology. We ran a cash-generative global business with a very small operational team. And the reason I'm writing this piece is that everyone is now discovering what that operating model feels like, at the same time, in every industry — and the conversation about it is being dominated by people who haven't been through one of these transformations from the inside.
Twenty-five years of building automated fintech businesses produces four observations that the current AI discourse mostly misses. The most important one is that the intangible asset portfolio decides whether the transformation compounds or delivers a one-time saving.
Here is what I know from a quarter century of building heavily automated businesses that I don't hear in most of the current AI discourse.
When we automated STP at IG, the reconciliation team didn't vanish. They became the risk operations team, the client operations team, and the oversight team. Their expertise — the pattern recognition they had built up doing manual reconciliations — became the training data and the exception-handling logic for the automated system. The best of them ended up running the automated system. The weakest found their roles contracted or eliminated.
The pattern is consistent across every automation project I've seen: the people who could articulate what they were doing and why became the builders and operators of the next generation of the capability. The people who couldn't, found the ground shifting under them.
This is what is happening now in knowledge work, at scale. The lawyers who can articulate what makes a contract clause robust will become the architects of AI-assisted legal work. The ones who can't will find their roles automated. The consultants who can explain their methodology will license it, productise it, and deploy it at leverage. The ones running pattern-matched intuition will find that pattern-matched intuition is now reasonably cheap.
This surprises people who assume regulation slows automation. The opposite is usually true. Regulation forces documentation. Documentation is the raw material for automation. The reason KYC, AML, appropriateness, and trade reporting automated so thoroughly in fintech is that every step of the process had to be written down, justified, and auditable. The written process was then translatable into code. Unregulated knowledge work — consulting, strategy, creative — has automated more slowly precisely because the methodology was rarely documented to the standard regulated work demanded.
The AI wave is changing this, because large language models are unusually good at extracting implicit process from examples. A team that has never documented its methodology can now produce a reasonable first version of it by running an LLM over the last six months of its output. The result is that previously unautomatable knowledge work is becoming automatable — not because anyone wrote it down, but because AI can now infer the pattern from the artefacts.
The lean companies I've built and invested in don't run lean by cutting cost. They run lean by building operational leverage. Every role is scoped so that it has a throughput multiplier — one person, with the right tools, does the work that would have taken ten. The tooling, the data, the processes, and the culture all support this. You can't achieve this by taking a traditional organisation and cutting it. You can only achieve it by designing the operating model for leverage from the ground up.
Companies executing AI-driven layoffs are in a difficult middle ground. They are trying to retrofit an operating model that was designed for one scale of headcount into an operating model optimised for a smaller one. The ones that will succeed are the ones that use the moment to genuinely redesign how work is done. The ones that will struggle are the ones that treat it as a cost exercise and expect the remaining team to do the same work with less help.
This is the insight I've come to over the last couple of years that I didn't have when I was running operational automation at fintech. Every automation investment we made at IG and Capital.com was simultaneously a build of intangible capital. The automated KYC system was a technology asset and a data asset and an organisational capital asset, all at once. The algorithmic bidding platform was a technology asset and a data asset. The AI trading bias detection platform was — and remains — a remarkable intangible asset combining technology, data, human capital (the researchers who defined the bias patterns), and customer capital (it genuinely made clients better traders and built loyalty as a result).
We didn't have a structured way to measure any of that at the time. We knew the projects worked. We knew they generated returns. We didn't have a view of the intangible asset portfolio we were building. Looking back, I can see that the projects that created the most value were the ones that built multiple intangible drivers simultaneously. The ones that only built technology, without the accompanying data, organisational capital, or customer capital changes, were the ones that delivered less durable returns.
This is the lesson I've been carrying into Opagio. The companies that will compound advantage from AI are the ones that build intangible portfolios deliberately. They don't deploy AI to cut cost. They deploy AI to build technology capability, data assets, organisational capability, customer capital, and brand equity simultaneously. They measure the movement of each driver. They allocate their subsequent investment based on what's working.
If you are currently considering an AI-driven transformation, the questions that matter are not the ones most vendors are answering. The vendor questions are about model selection, integration architecture, and implementation timelines. Those questions matter, but they are not the questions that determine whether the transformation creates lasting value.
The questions that determine the outcome are different. What is the current intangible asset position of the business, driver by driver? Which drivers does this AI investment strengthen? Which does it put at risk? What is the plan for capturing the knowledge that sits with the workforce today before it leaves? How will the transformation be measured across the full intangible portfolio, not just the cost line? Which of the twelve drivers will be the decisive one for whether this compounds or delivers a one-time saving?
These are measurement questions. They require the kind of framework that existed in fintech for operational domains — scoped, specific, trackable, with clear connection to enterprise value — but applied to the full intangible asset base. That framework didn't exist when I was automating fintech. It exists now, and it's the reason I'm doing what I'm doing.
A mid-sized regulated financial services firm in 2026 deploys an AI compliance review layer. The vendor case focuses on time savings per file. The board's question is different: which of the twelve drivers will this build, in what proportion, and how will we measure it? The team produces a one-page projection showing expected movement across Technology, Data, Organisational Capital, Human Capital, and Customer Capital. The Technology line is large but not the largest. Data and Organisational Capital together account for more of the projected enterprise value contribution than headcount savings. The investment committee approves on the strength of that composition — not the cost line.
Twenty-five years ago, I was pushing settlement automation into a business that hadn't yet understood the implications. The automation worked. The business compounded the returns. The back-office team was transformed, not eliminated. The companies that failed to automate spent the next decade being outperformed by the companies that did.
The same thing is happening now in knowledge work, across every industry. The automation will happen. The question is which companies will use it to build durable intangible capital that compounds, and which will use it to deliver one-time cost reductions that their competitors will match. The difference between the two outcomes is measurement — and the willingness to treat intangible assets as the portfolio they actually are.
This is the fifth piece in a series of seven on AI and intangible capital measurement. Companion pieces: Block's 40% workforce cut, the measurement layer the AI economy is missing, Salesforce and organisational capital, the seven drivers that move when you deploy AI, and the real ROI of AI as intangible capital formation.
The pattern is the same every time. A new wave of automation arrives. The companies that engage with it as a portfolio problem — what assets do we build, what assets do we deplete, how do they compound — pull away. The companies that engage with it as a cost problem run the same business with fewer people. Both can look fine on the P&L. Only one builds an enterprise that's worth more in three years than it is today.
The instrument to tell them apart already exists. It's a question of whether the CFO and the CEO are willing to pick it up.
Most AI investment cases are evaluated on a single dimension. The companies that compound advantage evaluate them across twelve. Two ways to start:
The companies that compound the AI decade will be the ones that treated it as capital formation from the start. The instrument exists. The question is which side of that line you want to be on.
When a CFO reviews an AI investment, the dashboard shows cost savings, implementation budget, and a productivity metric. What it almost never shows is the movement of the intangible asset base — seven drivers typically move simultaneously. Most companies track one.
Read more →
Salesforce reduced its customer support team from 9,000 to 5,000. The operational story is clear. The intangible story — what happened to 18 months of learned context per departing engineer — has almost nowhere to live on a standard dashboard. It deserves its own line.
Read more →R&D spending creates intangible capital that rarely appears on the balance sheet. This article examines the gap between what companies invest in R&D and what their financial statements show, with practical guidance on mapping R&D to intangible asset categories under both US and UK accounting frameworks.
Read more →Get the latest insights on intangible asset growth and productivity delivered to your inbox.
Book a free consultation to see how Opagio Intangibles can help your business.