The model: advisory-native AI, built to operate

Traditional consulting sells capacity: a team, a timesheet, and a report at the end. We sell judgment — and then we stay to make it real. Advisory is how we enter: a senior partner at the table who has actually done the work, not a graduate with a framework. But unlike a consulting firm, we don’t stop at the recommendation; and unlike a software vendor, we don’t drop a tool and leave. We advise, we build, and then we run the function — and the part that runs is carried predominantly by software and agents, not by an ever-growing team of people. Advisory is the tip of the spear. The embedded, AI-run operating function is the body of the business.

Advisory is the way in — and where our edge is real

The thing we have that a frontier AI lab and a born-yesterday delivery startup do not is earned advisory judgment: founders who have sat in the room, made the call, and carried the consequence across large, regulated programs. That is the scarce input, and it is the hardest thing to fake.

What we don’t do is what the big firms are doing with it. Watch closely and the incumbents — ours included — are bolting AI onto the old advisory motion: a copilot to write the deck a little faster, a notebook to summarise the interviews. The process underneath is unchanged. The opportunity is the opposite — rebuild advisory itself around agentic workflows, so the work is done differently, not just faster. That is what we mean by advisory-native AI, and it is wide open: the firms with the judgment are too invested in the billable hour to cannibalise it, and the firms taking AI seriously don’t have the judgment.

Our buyer is a company under heavy regulatory load that is too small to staff the function properly and too small for a major firm to serve profitably — a regulated scale-up or SME operating below the big players’ floor. The regulatory load is the forcing function: it makes a serious digital and compliance function non-negotiable, exactly when the company can least afford to build one. Renewable-energy delivery entities are where that need is most acute right now (section 5), but the pattern is the segment, not the sector. This is deliberately guerrilla — we win where the incumbents can’t profitably follow.

From day one such a client gets an accountable technology executive at the board table, an architecture designed for where the company will be in three years, vendor selection run by someone who has done it at scale, and systems implemented, integrated, and operated. They don’t get a deck recommending these things — they get the things.

Then we stay — and we scale on software, not people

A recommendation a client puts out to tender is a commodity. So we don’t hand the work over at the boundary; we embed and operate the function we designed. This is where the business actually lives — the recurring, sticky, long-term layer, priced as a subscription to a working function rather than as hours.

And it runs on one hard principle: the operate layer is carried predominantly by software and agents, with the lightest human footprint we can hold. If we scale by adding people the way a traditional firm does, we have simply rebuilt the Big 4 with better marketing — the same economics, the same ceiling. We scale on software. Headcount is the thing we are deliberately engineering out of the growth curve.

We hold the degree as a goal, not a solved problem: how much of the operating layer genuinely runs on agents, versus how much still needs a hand on it, is the open question the first engagements exist to answer (section 8). But the direction is not negotiable — every part of the function we can move from people to software, we move.

We make people more capable — we don’t replace them

To be plain about what this venture is not: we are not here to walk into an organisation and cut a third of its staff. That’s deliberate, on three counts. It isn’t the company we want to build. We also don’t believe wholesale AI-for-headcount replacement is real at this scale — that story is oversold. And to the extent any organisation does want pure workforce reduction, that’s an enterprise game, served by players built for it; it isn’t ours. Our clients are the mirror image: they can’t hire the people, can’t afford the expertise, or can only do this badly or not at all. There is no one to displace — we add a function that otherwise wouldn’t exist. The segment doesn’t merely make that stance affordable to hold; it makes it true.

The same instinct runs through delivery. The operate layer is not a black box we guard — part of the job is lifting the client’s own people onto the same agentic tools and workflows, driving adoption inside their business rather than hoarding it inside ours. That compounds three ways: the client extracts more value and leans on the function more, so it grows with them; the relationship gets stickier because the workflows are ours and now run through their operation; and it keeps us on the right side of the line — we make people more capable, never quietly replace them.

The cyber-security parallel

This is not a novel shape; it is the model the best cyber-security businesses already run. They enter on advisory — a security assessment, a virtual CISO — to earn trust and learn the environment, then land the recurring managed service, where a platform and a small expert team detect and respond around the clock at software margins. The assessment is the wedge; the managed service is the business. A string of these companies have been built and sold on exactly this motion — several acquired by the big consulting firms themselves. We are running it for the digital operating function rather than the security function, with agentic systems as our platform.

The three roles

The team that does this is an inversion of the consulting pyramid — three senior-weighted roles, none of them a billing-fodder layer:

• Client Leader — the senior executive (Director/MD-calibre) who owns the relationship, sits with the board, makes the build-versus-buy calls, and puts their name on every decision. The advisory judgment, and the accountability. The scarce resource the business is measured in.
• Engagement Architect — the senior delivery technologist who designs the systems, stands up implementations and vendors, and turns decisions into working architecture across several clients at once.
• AI Facilitator — the operator of the software layer: building, directing, and quality-controlling the agentic systems that carry the production and the ongoing run — research, analysis, documentation, monitoring, compliance tracking.

A three-person practice with these roles serves five clients in the financial model (section 8); nine people serve twenty. The pyramid is gone because the work the pyramid used to do is carried by software.

The leverage equation, rewritten

Maister’s classic formula — profit per partner equals margin × productivity × leverage — defined leverage as the ratio of juniors to seniors, which is why every traditional firm’s growth plan is a hiring plan. In this model leverage is the ratio of AI-carried work to senior time. A Client Leader in a traditional firm carries two to four clients because production drags on their team; here the working assumption is six to eight, because production and run don’t touch them. This is the model’s most load-bearing assumption: section 8 shows what the economics look like if it proves to be four rather than eight, and validating it is an explicit objective of the first engagements. It is the same question as “how light can the operate layer’s human footprint really go” — asked from the financial side.

Why a better model doesn’t erase this

The obvious challenge to any AI-native firm is that the next, smarter model simply does the work itself. It doesn’t — because the defensible ground was never the model. A more capable model still doesn’t hold the client’s operating licence, sign off on the liability when the regulator calls, or own the private files, systems, and context the work runs on. That ground is private and accountable, and it is exactly where we embed: inside the client’s regulated business, with a senior human whose name is on the decision. Better models make our software layer cheaper and sharper — they don’t reach the corner we occupy. The scarce thing was never the capability; it is knowing what to point it at, and being answerable for the result.

What AI does, and what it can’t

AI carries the production and, increasingly, the run: market and vendor intelligence, architecture options, specification and documentation, compliance tracking and regulatory-change monitoring, system monitoring and alerting, board-pack preparation. This is the work consulting juniors do today, delivered at a cost measured in dollars rather than salaries (section 2 sets out that economics).

AI cannot know whether the answer is right for this client, this board, this regulator. It cannot negotiate a vendor contract, read a founder’s risk appetite, or be accountable when the market operator calls. Every engagement runs on a simple division: AI produces and runs, seniors decide, and the client always has a name on the decision.

← The shift: AI breaks consulting's production economics How organisations deploy AI — and where we sit →