On why headcount-centric resourcing is over — and what comes next for regulated operations.
I used to propose multi-year resourcing solutions where the cost never reduced. I can't stand by that anymore. The right answer today is capacity now, with a commitment to materially reduce it within six months — starting immediately.
I've been inside reg reporting and reconciliation functions for over twenty years. Every wave of change — MiFID, EMIR, SFTR, T+1, now EMIR Refit Phase 2 — has landed the same way: more fields, more regimes, more volume — and inevitably more people. The resourcing model tracked the pressure. Body counts grew in lockstep with drivers. That was the deal.
The deal has broken. Every resourcing partner now has an "AI-enabled" slide — but clients aren't seeing anything change. Two prospects told us the same thing in the same week: "our resourcing partners should be doing this. They talk about it, but we haven't seen any improvement land in reality" — the body count hasn't moved, the cost hasn't fallen, everything looks the same on a Tuesday.
That's not a product problem — it's a model problem. Selling capacity on a multi-year contract makes the provider's revenue a function of the client's cost — and that cost scales with every new wave. Fixing that requires unwinding the provider's own commercial interest. Very few providers will.
I've run Oxygen for four years, and I've also sat where our buyers sit. What we're building is the opposite commercial shape: capacity now, with a contractual commitment to reduce it.
A practitioner — embedded in the function from day one. Working the same break queue the team already has, but with AI doing the early lifting before they log on.
Take an EMIR Phase 2 reconciliation break queue on a Monday morning. Under the old model, a specialist opens the queue, figures out where the biggest problem is, picks the first break, pulls the underlying trade, walks the lineage, hunts the root cause and solution. The practitioner's clock starts at the break. So does everyone else's — on thousands of breaks.
Under our model, the AI has already worked. Every break surfaces instantly, before the queue builds — across the full expanded Phase 2 field set, not just the old one. Every fix the team made last week is encoded, so the same break never costs twice. Every investigation the practitioner opens starts with root cause surfaced and a resolution suggested. Morning one.
The practitioner still owns the work, but the work starts further up the value curve. Their hour is elevated to judgement — ambiguous breaks, counterparty calls, regulator narratives — not processing. What used to fill the day takes minutes. What used to get minutes fills the day.
This sits inside a broader frame. Operational excellence has always been a five-layer problem — culture, low-friction product design, technology, process, people. What changes now is that the technology layer has a new entrant that reaches further than anything before it. Agentic AI unlocks a final mile that deterministic tooling — RPA, workflow engines, rules — never could.
Less manual, more elevated.
Every client we speak to has a "never-AI" list — work they won't trust to a machine. Judgement on ambiguous breaks. Counterparty conversations. Regulator narratives. Trading. Discretionary corporate actions analysis. Final research report proofing. Something is always on the list.
The list is often shorter than clients think, and it's shrinking. Partly defensiveness, partly first-cycle skepticism — in any technology cycle, the boundary gets drawn before the trust is built. As the tools prove out, the boundary moves. We respect the list and flex the model when the client is ready.
As the technology earns more scope, the team gains it back as judgement time. Fewer hands needed to do the processing; the same hands doing more of the work that actually drives outcomes and needs their deep expertise. The part that's high-value and under-resourced — the part the regulator is scrutinising.
Of course, there are new responsibilities that running AI creates — governance, curation, assurance. We run them as part of the engagement, within a joint oversight framework, so the function doesn't have to staff them. AI-enabled isn't AI-replaced. The client's core team holds and develops; what comes down is the resourcing layer that was there to scale with pressure that the tech now absorbs.
By shortening the horizon for results. Capacity today, with a commitment to materially reduce it within six months, starting immediately. The how is our problem.
That shape is the opposite of the model most resourcing providers sell. A headcount-based contract makes cost a function of pressure. More regulation, more volume, more complexity — more headcount. That's been true for twenty years. It's also what boards have been telling ops leaders to stop doing. "Do more with less" has been the lived directive, year after year. There hasn't been a good answer.
Until now. This is the first answer that actually reshapes the deal. I can sit in front of an ops leader who's been asked to do more with less, and offer something that isn't another headcount-based contract — and doesn't take a year to cut costs. The commitment is specific, the timeline near, the accountability ours.
Clients aren't signing up to manage headcount into the future. They're signing up to watch it fall. The practitioners are in the function on Monday and the reduction lands in months.
What we're pointing at is where regulated operations is heading, and we're building the model that lands there.
Two weeks of discovery, no commitment. The client's priority issues mapped, root causes surfaced, projected savings on the table. If the picture's right: capacity in the function next month, with a contractual commitment to reduce it within six months.
Founder & CEO of Oxygen Consulting. Twenty-plus years inside regulated post-trade operations — most recently Managing Director at MarkitSERV (OSTTRA). Oxygen partners with Gentek.ai to build AI-First Operating Models for capital-markets firms. A different shape for capital-markets operations.