AI transformation is easy when you’re looking at optimising individuals. The harder transition, and the one that will decide who is still competitive in five years, is transforming the entire organisation. This is part one of a 2-part series about AI transformation within our scale-up, Firmable. What we've learnt so far about the process and how we’re thinking about leveraging AI across our business.

This post is based on a presentation we gave to our executive team at a recent offsite.

We were recently comparing notes on AI in the organisation with a few startup founders. One founder said they’re using token spend as a signal of AI adoption in their organisation.

To us, token spend indicates who is using the tools. It tells you nothing about whether or how the company has changed.

There's a lot of noise right now about using AI to create 10x individual contributors. The best people, given these tools, really are producing work at a level that would have been implausible two years ago.

But an organisation full of 10x individuals is not a 10x organisation. It's the same company with a few faster operators.

The challenge ahead for the most ambitious companies is to transition from an organisation with the occasional 10x individual to a 10x organisation.

This is a much harder transition, but there’s gold at the end of the rainbow. A company that gets there has a cost advantage and an agility advantage. In the long run, they’ll be able to outcompete, outprocess and beat their competitors.

What AI transformation is right for you and your company is likely to change depending on how well you understand the technology. AI-native founders speak about their AI operations at work very differently from large corporates.

We felt it would be useful to be specific about what AI transformation means for any business and what we’re aiming for at Firmable. We describe the journey as a ladder of five levels:

Level three is our target for this financial year.

In our experience, you can buy your way into the first two levels with new tooling. Level three is the transition where you’ll need to be building.

A 10x has built their prompts and projects, wired up their own MCP connections within the apps they use, and developed a feel for what good output looks like. Their individual throughput has increased severalfold.

It's also fragile. The gain is attached to a person, their setup, and their individual responsibilities.

It's likely undocumented, so nobody else can pick it up. It walks out the door when they do. Two people in the same role solve the same problem in completely different ways, with different prompts and AI point solutions.

A 10x system, on the other hand, has the same capability but is transferable, reusable, governable, and measurable. Anyone in the role can run it and get the same result, and it scales with the team. A new starter inherits the capability on day one instead of spending six months rebuilding it.

The shorthand we use internally is apps, not artefacts.

When someone shares a clever setup they've built (like an artefact), that's still a personal productivity thing they're handing you.

An app is something the company owns, maintains and improves. At Firmable, we built an internal environment we call Labs specifically so that people have somewhere to test, share, and publish their applications. It’s the act of publishing to the organisation that moves me toward a 10x organisation because it’s shared, tested, and governed.

We approached our AI org transformation from first principles, which told us that AI needs two things to operate well inside any business:

So first, we brought the core datasets together. This included sales and deal information from the CRM, in our case, HubSpot; subscriptions and billing on Stripe; product usage and adoption; and user information. Even the accounting stack and the HR software.

That gave us a picture of how all the bits were moving inside the organisation.

At that point, we could ask, "Build me a dashboard for our sales last month," and get a live dashboard back. But our leadership team still had to prompt, and prompt again, to get at what they actually meant. It was automated, but often wrong, which is worse than slow.

We learned pretty quickly that connecting the systems was the easy half.

Let’s go deeper into that dashboard example.

Say you want to know the value of deals sold last month. In the old world, you ask your revenue operations lead. She comes back with a number, and because she’s great at her job, the report is infused with a dozen judgment calls nobody asked her to make but that are expected. She converted every currency into Australian dollars because she knew that's what you meant. She excluded renewals because although we manage renewals as deals in the CRM, she knows that's not what you were asking about. She's done it before. She personally carries the context and knows what the management team was looking for.

An AI doesn’t have that context yet. So the real work was sitting down to define our operational context. What counts as a deal? Which currency? What's in and what's out. And then exposing those definitions to every tool and model that touches the data. It’s a meaty job, but it helps you start to think about your business as a system or a set of jobs to be done.

So where we are now is a data lake that brings together all the moving bits inside Firmable, and a context layer we’re constantly improving, which knows how our leadership, managers, and individual contributors want work delivered.

Every company we've seen successfully integrating AI has an internal AI architect. While each business is different, the closest comparison would be to a chief operating officer. However, the primary responsibility of this role is to implement AI into the business and maintain its functionality.

It isn't a common title yet, but it will be. It isn't necessarily the most technical person in the building.

The requirement is someone commercial, with real breadth across the business, technical enough to understand what's possible, and with the judgement to sequence it and the authority to implement it.

It’s also not someone you can bring in temporarily, because it is never a single decision that’s set-and-forget. First, you need the data layer. Then the right tools. Then a company-wide context layer, because agents can't operate well without one. Then somewhere to publish what people have built, like custom automations. Then shared apps, so work compounds instead of being rebuilt.

This is non-trivial, and we suspect very few organisations have this person. We'd go further, slightly controversially: it's changing what the chief operating officer job is. Organisational architecture used to be something you did function by function. It can't be anymore.

The organisation layer is the part you can’t skip and can’t delegate to a vendor. Operational data, organisational context, a place to publish what gets built so the whole organisation can benefit, and one person whose job it is to sequence all three.

Get that wrong, and every clever thing your people build stays theirs, not yours.

We’re not there yet. Level three is our target by the end of this FY, and some functions are a long way off. But the question we ask has changed, and it’s the one we’d pass on to any founder: is our best AI capability something the company owns and can hand to the next person, or something a few talented AI-pilled people are carrying?

If it’s the second, you have 10x the people. Enjoy them. They’re great, but they’re not the same thing as a 10x company.

Which brings us to the harder half. You can build the whole organisation layer and still not be a 10x organisation because the people inside the organisation haven’t changed how they work.

Stay tuned for part two, where we’ll talk about the organisational change required to 10x your organisation.