DISCLAIMER: this wasn't written by AI. Please excuse the lack of em-dashes.
The joy of portfolio life
I have the privilege of sitting in the CFO seat for a number of high-growth companies across a range of sectors. It's a complete blast - every day is different, and I have the pleasure of working with and learning from such a broad and diverse set of people across the companies. I've been CFO for a single company several times before, and each of those experiences was fantastic. But, it's the portfolio role where I get the greatest kick!
Working across a portfolio of companies means my focus and responsibilities vary significantly across the different roles. For some companies I'm the entire finance function, and for others I manage a team. With some we use Xero and with others we don't. With one company we use HubSpot, but with a couple of companies the teams manage their sales pipelines and CRMs in Google Sheets. Every single company is massively different. However, one thing is consistent across them all: automation of workflows and key processes is no longer just a nice-to-have - it's a core strategy.
AAA…! (AI, Automation, Agents)
Automation is certainly the buzzword of the decade for the finance function, and I'd wager that most finance professionals often find themselves paralysed by the vast amount of materials and content out there on the subject. And the confusing picture isn't helped by the never-ending list of tools and services that are seemingly now able to automate absolutely everything. As a finance leader, you can easily find yourself worrying that you're perhaps the only finance team on the planet not geared up to the teeth with AI and managing to close month-end by 00:01 on WD1.
I've been experimenting with AI and agents for some time, using the likes of Claude Code and Codex for automating processes, modelling and reporting. It's been an amazing, eye-opening journey. I'm sure most people that have dabbled can now see the huge potential AI presents for the finance function of the future. But the problem now for a lot of people/teams is… where to start? If you're not already underway with your AI & automation journey, or if you're still struggling to know how to move past the Claude Excel plug-in - perhaps I can offer a bit of guidance. And for me personally, the penny really dropped when I started to explore the digital twin concept…
What is a digital twin?
The digital twin idea isn't a new one. In fact, the concept of the digital twin - a virtual representation of a physical system - can be traced right back to the start of the space age¹. Since then, the industrial sector has been the main adopter of such systems, and with great effect. But, with AI rapidly finding its way in some form into every type of business, the digital twin concept is relevant everywhere.
At the start of this year, it was an alien concept to me as well, but a chance conversation with Archie - CEO and founder of Mercury Labs - gave me the thrilling opportunity to get set up and testing the digital twin for each of my portfolio companies - a portfolio that quickly grew to include Mercury Labs once I saw the potential of what Archie and the team were working on.
Today, I'm running the MLX digital twin across 4 very different but equally exciting companies. With every system connected and each company 'world' defined through playbooks and context layers (both set up and managed like skills in Claude/ChatGPT), I'm able to monitor and quality control a large estate of processes and automated workflows (month-end close, year-end management, R&D claims prep, financial model builds, board reports, investor DD, VAT reviews etc. - the list goes on).
I like to think about the 5 C's: Connectivity, Context, Collaboration, Control, Cost. We'll get to cost later on, but the first four are hopefully now clear. With everything in one place - securely managed, with adequate read-write controls, and with users and teams able to collaborate effectively - the potential for process design and automation is near-limitless. And with a twin able to integrate to virtually any system or feed, it's exciting (for me, at least) to think that legacy systems don't need to be constantly switched out. Automation can sit in the twin with legacy systems remaining where they are, doing their core job.
And one of my favourite things about the twin? Its versatility. Roughly, I split my time using the twin 50/50 between the web app and then Claude Code or Codex. With some thoughtful setup of systems and devices, I'm able to achieve huge amounts on the go.
Automation is important, but quality is more importanter
With all of the above, it's really important to note that I don't just throw the keys to AI and head off to the beach every morning. Quite the opposite in fact. Gatekeeping and quality control are more important than ever. The twin and the agents I use are there to work to my specifications, and so, more than ever in my career, I have to think deeply about process design, modelling logic, reporting narrative etc. AI helps with the heavy lifting, but I still hold the pen and sign off everything. The main difference is that now - thanks to the twin-enabled automation - I have more time to craft the narrative and double-check the results.
I still hold the pen and sign off everything.
And the result? I estimate I'm now saving at least 2-3 hours a day across the portfolio, which is more and more becoming time that I'm spending on the qualitative and creative side of things. The benefit truly begins to compound.
The future is a blank canvas
Vibe-coding is another great buzz term, and one that probably strikes fear into the hearts of many finance leaders. It's genuinely epic that we can all now build software at a rate of knots, but very few of us know how to properly test, deploy and maintain these tools. It's why so many of the ERP incumbents are still holding strong today - it takes a particularly plucky finance team to think that they can build and maintain their own bookkeeping system! I'm yet to meet one that has had any real success.
And that's why the crowded marketplace for finance automation is now so intimidating. There are some seriously cool new players on the scene: Ramp, Rillet and Light to name just a few of those shouting particularly loudly about their many and varied features. One of them even boldly claiming that Zero-Day Close starts here… But, it's really hard to know where to start, because at the core of it (and putting payments and banking integrations to one side for a second), really what most finance teams want are clean, easy-to-use ledgers and slick reporting capabilities. It's why most teams that I meet are sticking to their legacy ERP system and using Claude to build out their FP&A.
With the digital twin, however, it's quite easy to see where the future state is: finance teams fully in control of automation and not relying on a third party to ship the next feature, or a particular model provider to provide the AI. The blank canvas.
The twin at the core of that is the unlock. Not just a context layer, the twin provides the canvas for automation with its just-in-time software approach. Want to merge Xero and HubSpot data to build a CRM that reconciles payments and credit notes? The twin can do that instantly. Want to create a custom dashboard that reports on inventory that's tracked on a Google Sheet? Just ask the twin. Need a custom cap table and a kanban to manage a fundraise close? The twin comes with templates for this, or you just create your own. It's genuinely as easy as that.
Keeping in control
Perhaps unsurprisingly, every finance leader I speak to on this subject brings up the same three key points very early on in the conversation:
- Does a twin become the source of truth?
- What about out-of-system records (our 3-tab CRM tool built in G-sheets)?
- How do you manage version control and user access?
I hit these three same points very early on in my own twin journey. It was therefore a real thrill to be so involved in the development of the twin early doors, working with the team to get these bits right.
- No - the twin is a… twin! It gives teams access to every system and its data with recs and controls to ensure what's queried is accurate. However, playbooks* - created and managed within the twin - quickly become the source of truth for context and know-how.
- This is a key one. Being able to define (for example) a Google Sheet or a SharePoint drive location as a resource then brings it into the context ecosystem just as if it was Xero or Notion. Functions, files and outputs within the resource can be defined and even warehoused within the twin meaning the data can be queried alongside system data.
- A team can manage version control and user access how they like. Playbooks can be set to just certain users; data can be limited to only certain user groups; channels can be assigned owners so that changes and additions have to be approved. The twin offers full flexibility.
*Playbooks - simple markdown files, just like skills in Claude etc - really are key to everything. It's a bit like a code base but for company processes and key information. Finance teams have been writing process notes since the dawn of time. However, with playbooks embedded in the twin and properly managed with authoring and version controls, using AI becomes so much more effective and controlled. The penny dropped when I first started using the twin. The penny truly smashed a hole through the floor when I started to use playbooks.
The penny truly smashed a hole through the floor.
Thinking about cyber, lock-in, cost
The obvious question at this stage is: surely I can do all this with Claude or Codex and their plugins?
And yes, to a degree that's a valid question, and thousands of people and teams are doing just this. But if you stop to think about such a setup for a moment you quickly start to think a little deeper about a few key points:
- Cyber security & data. Is this secure, and where is my data going?
- Provider lock-in. What if Anthropic 10x their prices overnight?
- Cost. How can I manage cost across a large team of users?
Lots of things won me over when first using the twin, but it's with how it tackles these very points that I've been fully sold.
As per the graphic, the twin gives me and my companies complete ownership. Our data and workflows are hosted on secure EU- or UK-based servers, and we have complete control over the models we use and where/how we send our data. Three of the companies I work for are regulated in some way, and so the recent addition to the twin of UK-hosted servers and models has been a major unlock.
Closing note
It's now very hard for me to think of my role as possible to do without the help of a digital twin. In the right hands and with the right training & guidance, I can really see how bright the future is for finance teams. This isn't about automating everyone out of a job - it's about delivering the highest quality work in timeframes previously thought impossible. And the blank canvas analogy is incredibly important to me: the finance function can now start to build their own automations and workflows and not just rely on the next shiny product to come along. That's super exciting. Automating and improving the processes of today, and then building for the future.
References
- Digital twins: past, present and future. Scientific Reports 16, 10510 (27 March 2026). https://www.nature.com/articles/s41598-026-45272-z