6 takeaways from ‘How to scale AI fast’ (webinar with Danilo McGarry)

Enterprise AI pilots are easier to find than monkeys in a zoo. Companies are experimenting with agents, launching proofs of concept, and automating tasks across the business. But according to AI and transformation leader Danilo McGarry, there’s a danger in moving too quickly from “we should use AI” to “let’s automate this.”

McGarry has spent 20 years working in AI, including running major AI programs for Fortune 50 companies and advising organizations on transformation. He shared what those years have taught him about moving past AI pilots and scaling AI across the enterprise in a recent webinar moderated by Celonis Lead Transformation Evangelist Kerry Brown.

A big part of advancing from pilots, he argues, is that companies are starting with technology instead of understanding the work they’re trying to improve. As McGarry put it, “There is no way for you to transform your company unless you understand the process.”

For CIOs and transformation leaders under pressure to turn AI investment into ROI, that’s a good place to start. Here are six lessons from the conversation.

→ Watch the full webinar, How to scale AI fast: Former Fortune 50 ‘Head of AI’ on making AI work.

1. Stop chasing quick wins

“For you to just go and automate low-hanging fruit, if there is any, you’re going to hit a glass wall very quickly.”

“Let’s find the low-hanging fruit” has become a broken record, says McGarry. The logic makes sense: find a few use cases, automate them, prove some value, and build from there. It sounds like the fastest way to scale AI. But McGarry warns moving fast only works if you’re scaling the AI correctly.

Sure, you might save time, make a task faster, or have an impressive ROI number to show the board. But you may also be automating work that shouldn’t exist in its current form. As McGarry put it, “You’ve automated human-heavy tasks or imperfect human tasks. You’ve just done them quicker with automation.”

He suggests taking a step back and rethinking the operating model. McGarry compared enterprise transformation to swapping the engine of a race car while it’s still racing. You can’t stop the business while you transform it, but you also can’t expect meaningful transformation to happen overnight.

2. Before AI, understand the process

“The number one thing that I always go to is process.”

Process was the thread McGarry kept coming back to throughout the conversation.

Before putting agents in places they don’t need to be, leaders need to understand how work actually moves across the business.

McGarry used the analogy of renovating a house without looking at the blueprint. Start moving walls around without knowing which ones are foundational, and you’re taking a pretty big risk. Enterprise transformation isn’t all that different. You need to understand where work happens, where it slows down, where people step in, and which dependencies sit upstream and downstream.

That’s where a living digital twin of the enterprise comes in, like the one provided through the Celonis Context Model (CCM). It gives leaders the operational clarity to understand and improve processes without pausing the business or ripping out the systems underneath them. Not every problem needs an agent, of course. Sometimes a workflow is enough, sometimes traditional automation is enough, and sometimes the process itself needs to change.

→ Find out how the Celonis Context Model (CCM) eliminates AI blind spots.

3. Assess AI maturity first

How can you scale AI if you don’t know how ready the business is for it? McGarry maps organizations against an AI maturity curve, from zero — essentially no automation or AI — through five, where work can be extensively augmented by AI.

According to McGarry, companies can’t simply skip the stages in between.

“You can never go from a one to a three, by the way, or from a one to a four where you’re really agentically helping the company.”

Companies need the foundations for efficiency first. Scripts, RPA, better data exchange, and workflow engines can all play a role. As work becomes more complex and requires greater reasoning, AI starts to make more sense.

The webinar audience offered an interesting reality check. In a live poll, 52.4% of respondents put themselves at Level 1, “Exploring,” while another 37% were at Level 2, “Operational.” Only a small proportion placed themselves at more advanced levels.

Here are the full results for the poll question “Reviewing the graph, from 0–5, how far are you on this AI Maturity Curve?”:

1 - Exploring: 52%
2 - Operational: 37%
4 - Predictive: 5%
3 - Advanced: 4%
0 - No use of AI: 1%
5 - Fully Autonomous: 1%

There’s another complication: your company probably doesn’t have one maturity level. Finance might be further ahead than procurement. One business unit could be experimenting with agents while another is still trying to automate repetitive manual work. McGarry recommends assessing maturity by business line first, so leaders can see what different parts of the organization are ready for and what they need next.

4. Get your best AI ideas out of POC purgatory

A team builds an AI proof of concept that works and saves 100 hours, maybe even 1,000. Everyone is happy with the result, but then the idea never makes it beyond the team.

McGarry calls this “POC purgatory,” and it’s one of the structural problems he sees holding AI programs back.

“A lot of companies kind of live in this POC purgatory where the POCs are working fine; it’s saving maybe 10 hours, 100, a thousand hours for a team here and there, but they forget that there needs to be a layer of approval to really truly scale that idea so that everybody can benefit from it.”

McGarry argues that departments should have people experimenting with AI and developing ideas close to the problems they’re trying to solve. But when those experiments work, there needs to be a clear path to scale them. That could be a Center of Excellence or a small committee with the authority to take successful ideas across functions, regions, or the wider enterprise.

→ Learn why a Center of Excellence (CoE) is effective for scaling Enterprise AI.

Clear executive ownership matters too. McGarry is skeptical of automatically making the CEO the executive sponsor. The CEO can champion AI, but turning AI into part of the operating model requires someone who owns the transformation day to day. Depending on the organization, that could be a Chief AI Officer, Chief Transformation Officer, CIO, or CTO.

5. Give the CFO something they can measure and marvel at

AI ROI… where art thou? McGarry’s answer is to give people measurable AI objectives tied directly to business results.

“The best objective that I found — it sounds really simple, but works really well — is by literally saying for everybody, it can be junior, senior, whichever level the employee is at, they can choose between the two.”

The two choices he recommends are using AI to create 10% more value for the team or reducing the cost base by 10%. McGarry arrived at that figure through his own work, so it isn’t necessarily a universal benchmark. The point is to give AI adoption an outcome the business can actually see and measure.

He takes a similar approach to funding transformation. Instead of asking the board for an enormous AI budget up front, McGarry recommends starting with a smaller group of high-priority projects, proving their value, and reinvesting some of the returns into the next wave. He calls it “snowballing.”

For C-level leaders competing for budget, that makes for a much more concrete conversation with the board. Each new round of investment can build on value already delivered, rather than relying on the promise of what AI might eventually deliver.

6. Prioritization might be your most important AI skill

Companies can quickly rack up hundreds of potential AI use cases. So, where do you invest?

“You can easily have a thousand AI projects, but you’re going to have to decide which are the 10, the 20 that really matter, which are really going to drive EBITDA, shareholder value, extra revenue.”

That leaves transformation leaders with some big decisions to make. Which processes have the biggest impact on business performance? Where is value being lost today? Which use cases make sense for where the organization is right now? Which projects could actually work at scale, and which ones probably won’t make much of a difference?

You need operational clarity to answer those questions. Understanding how processes actually run gives leaders the operational clarity to see where AI could have the biggest impact — and where the time and money to scale it should go.

Scaling AI fast, the right way

There’s plenty of pressure to move fast on AI. McGarry’s advice is to know how the business works first, get the right operational clarity, and use it to make better calls about where AI belongs.

Or, as he put it: “There's no other way, there's no shortcut, there's no lazy leadership. It's only through [the] process that you're going to truly redesign your company when it comes to AI.”

→ Watch the full webinar, How to scale AI fast, with Danilo McGarry and Kerry Brown.