I bumped into a "nutter" last week who is on a mission to build a wealth platform from scratch using Claude. He may turn out to be a modern-day Icarus — but this moonshot ambition is not as crazy as it sounds.

For twenty years the mantra in the wealth management industry has been "buy, not build". Industrial-scale software development is hard, regulated software development is harder, and platform software is hardest of all. That conventional wisdom is now being challenged by the widespread adoption of AI frontier models, the most notable being Claude Code. From a standing start in February 2025, the growth curve went vertical in early 2026: past a $2.5bn run-rate and, by mid-2026, more than two million weekly users.

The cost of building has dramatically reduced while the cost of buying has not, and that gap is both an opportunity and a potential killing field for the unwary. The question is no longer buy versus build. It is where do you draw the line and how do you build?

And there is a deeper reason to care. Licensed software makes you identical to every competitor who bought the same thing. When everyone runs the same platform, the platform is not your edge: it is table stakes. Differentiation lives in what you build and own.

Escaping the legacy shackles: the incumbents' own economics are straining

Legacy wealth platforms — FNZ, Bravura, GBST and their peers — built durable, valuable systems on the dominant architecture of their era. But they are expensive to license and maintain. The version-upgrade treadmill is punitive: you pay to stay on a supported release, your customisations make every upgrade painful, and the gap between the version you run and the version they ship widens with each cycle you defer.

The dominant providers' own economics are now under visible pressure. FNZ, valued at around twenty billion dollars only a few years ago, is fighting a multi-billion-dollar shareholder dispute over the value of its own equity. Bravura's market value has fallen to a fraction of its former level. If you are betting your firm's differentiation on a platform whose pricing model, upgrade cadence and roadmap priorities are not yours, and whose own finances are strained, you are exposed to decisions you do not control.

The strain shows in the migrations. Quilter's multi-year move onto FNZ ran well over time and budget and the firm is now building its own adviser back-office, with FNZ reportedly paying Quilter some £20m a year toward it. When your largest clients start building around you, the moat is already leaking.

The hype and the reality

The build-not-buy community will not have it all their own way. Some firms, especially those that are PE-owned, are pushing for an "AI story" to drive their valuations. Others are far more sanguine, and point to the lack of use cases that demonstrate an EBITDA uplift. The spend on AI is immediate; confidence in the margin improvement can be hard to pin down. The business case can look like asking for a blank cheque.

Wealth platforms are also genuinely complex. Custody, settlement, tax, corporate actions, reconciliation, regulatory reporting and the transactional core are hard to build and dangerous to get wrong. It is far easier, and far safer, to build the surrounding layer than to rebuild the engine room. That is not a reason to do nothing. It is a reason to have a strong architectural design, a build/transition plan that reflects the complexity and the risk, and a pragmatic mindset about which hard-to-replace ecosystem components you surround and supplement rather than rebuild.

This is not theoretical: the entrants are already here

If this still sounds like a thought experiment, look at who is already building. Obsidian, a UK start-up, is building an adviser platform from the ground up on its own core. Its AI practice-management layer — meeting notes, CRM, portfolio aggregation, document digitisation, auto-drafted suitability reports — is live and offered free, with custody and execution being built into the core rather than bolted around it.

Others are close behind. Wealth Vector is signalling an AI-native operating platform with a feature set built for the rapidly evolving demands of an AI-transformed wealth industry.

The "nutter" I bumped into last week is part of a growing community who fundamentally believe — as we do — that the technology underpinning the wealth management industry is a constraint, not an enabler. Re-tooling has never been easier, and for the first time the build option looks viable.

When Steve Jobs returned to Apple in 1997, he launched the "Think Different" campaign. The narrative ran:

Here's to the crazy ones. The misfits. The rebels. The troublemakers. The round pegs in the square holes. The ones who see things differently... While some may see them as the crazy ones, we see genius. Because the people who are crazy enough to think they can change the world, are the ones who do.

Steve celebrated the nutters. So do we.

How to go about building

The first step is to prove the business case — honestly. I remember, in my early days at the Pru, trying to replace a couple of tightly coupled legacy systems. A big, bold idea turned into a colossal failure the moment we had to come clean to the board that the new systems would have to sit on top of the legacy, not replace it. That was on the Tuesday. By the Friday the programme was cancelled, and circa £20m was down the toilet. The lesson has aged well: build around the core, and don't claim you have replaced it until you actually have. Architectural honesty is the difference between a moonshot and a write-off.

These are the essential components that must be present to underpin the business case:

  • The solution architecture, including the data substrate that will drive and control the AI deployment. This must be a clean, layered model with plug-and-playable UI, orchestration and data management, designed to eliminate vendor lock-in.
  • A granular cost-to-serve model, based on workload volumes and "should-take" times per task.
  • A deployment and governance model: the piece most firms underestimate.
  • Clarity about benefits realisation: an honest assessment of headcount savings versus the value of retaining people where automation is only partial.

Governing what the machine writes

Frontier models make it cheap to write regulated software. The hard part is governing what they write so it stays safe, auditable and maintainable for years. That is the job of our Specification-Governed Development (SGD) framework. SGD makes the specification — not the code — the source of truth: every capability is a versioned spec with testable acceptance criteria, every line of AI-generated code is traced back to the spec that authorised it, and an automated merge-gate blocks anything that drifts. Think of it as the operating system for the build. It governs design, build and ongoing operation — which is exactly what turns a clever prototype into something a board, and a regulator, will sign off.

Treat it as an operating-model change, not an IT project, and your team can build high-integrity systems in unprecedented timescales, with the controls and guardrails built in from the start. No army of consultants required.

My prediction is that more wealth management firms will follow Fidelity's lead, when in 2024 they paid £29m for a licence to operate and develop Bravura's Sonata platform with their in-house team, taking control of their own platform destiny rather than waiting in a vendor's queue and the ching-ching of the return-to-core version costs. It wouldn't surprise me if another household name, with deep pockets and currently powered by FNZ, becomes the next to take the bold step and commit to a build-not-buy manifesto.

Let's talk

The conventional wisdom was right for its time, and its time is ending. Buying (for now) still wins out for the regulated core. Building now wins for the rest. If you want a view on where that line sits for your firm — and what you could own rather than rent — drop me a line at dave.howard@wealthtechpros.com.