Five Times the Output: A Year of AI-Assisted Advisory in Practice

Open my calendar on an ordinary week in May, and the first thing you notice is how dense it is. Building sessions, accelerator tracks, trainings, and communities of practice stacked from early morning to evening, spread across firms in law, finance, wealth management, software, and industrial technology. A year ago a calendar like that would have looked like serious overcommitment. Today it is simply the shape of the work. Since January the demand for AI has gone vertical. The questions coming in now are concrete, they are urgent, and they are owned right at the top of the house.

In a normal advisory and training business, all of this would be staffed by a small team with a project manager holding the parts together. With me it runs from one practice. Using AI across the whole of it, I now produce roughly five times the revenue a single person could carry in the old model, and every deliverable goes through a gated review before it leaves my desk.

What the portfolio actually holds

The breadth is the whole point, so let me put it in one view. 12 AI Accelerator tracks run at the same time. Some one to one, some in small groups, with law firm partners, CEOs, and several one-to-one tracks with board chairs. Inside a software company owned by Goldman Sachs I have my hands on the build itself, plus a team with a VP, their AI Lead, and me meets every week and reviews together. For a private wealth management boutique I build the AI solutions myself, from the workflow up. On top of that, trainings run in other organisations, among them a fintech platform with twelve people doing live online training twice a week. And underneath it all, seven communities of practice I run for F5 carry on at their own rhythm.

No two of them ask the same thing of me. A board chair learning to think with a model needs something very different from the twelve people on the fintech platform, and they need something different again from the software firm wiring AI into its own product. The thread is simple, though. In thirty percent of the cases I am building the actual solution alongside the client, hands on the work, and I never hand over a slide deck describing a solution someone else has to build later. In the rest I am either training board chairs and CEOs one to one, seven of them right now, or training inside the organisation itself, five of those at the moment. Every one of these trainings runs to twenty-four sessions, built on an Accelerator concept I developed myself.

What multiplies is judgement

You might assume AI multiplies output by writing faster. Speed is the smallest part of it. What really moves the needle is that the model now carries the production work that used to eat the hours between the thinking. The first draft of a governance model. The rewrite of a decision memo. The second version of a workshop design, after the first one has been tried in the room. I bring the judgement, the client context, and the standard. The model takes the volume.

You get a finished draft, a critique of that draft, and a corrected version in a single pass. So the work that reaches a client has already survived its own review before I have read a word of it. The hours that used to vanish into producing now go into deciding. And that is exactly the part a client is paying for.

How the quality holds when more than fifteen tracks run at once

At this volume, quality cannot depend on me reading every line one more time at midnight. It rests on a gated review that every deliverable passes before it leaves my desk. Facts and sources checked. A structured draft. A diagnosis of the draft's own weaknesses. A revised version that answers each one.

Each gate carries a single question. Is every claim sourced. Is the language unambiguous. Can the client act on this tomorrow morning. Would it hold up to an audit. Those gates do the work a second pair of eyes does in a bigger firm, and they do it the same way on a tired Thursday evening as on a fresh Monday morning. You get a consistency that no longer rises and falls with how much I slept.

The Danish word for what holds all of this together is ordentlighed. The quiet standard of doing the work properly, so it can be handed on without apology and without a footnote of excuses. AI did nothing to lower that standard. What it did was let one person hold it steady across a full portfolio at once.

What this means inside your own organisation

The lesson reaches well past one adviser carrying a team's load. The same mechanism works inside a company of any size. A controller, a recruiter, an account manager. Each of them can carry more if three things hold together. The judgement stays human. The production is shared with the model. And a gate protects whatever goes out the door.

That is the pattern behind the Wharton finding I wrote about earlier this year. Organisations keep funding the tools while underfunding the people who have to turn those tools into decisions. The limiting factor is rarely the model (and please remember, right now you have the worst model you will ever get). What matters is whether the people working alongside it have a standard to hold it to, and the protected time to practise that standard on real work that matters.

The practice I would build again

A year ago this calendar would have needed a team. Today it needs a standard and a machine that respects it. More than fifteen tracks differ in almost everything that matters to a client. And still they run on one principle. The human decides. The model produces. A gate stands between the work and the world. That is the practice. After a spring this full, it is the one I would build again.