Pyxl cut the time to first concept by 42%. Gross margin expanded 13 points, EBITDA margin 18.

The situation

AI was already making work faster. That did not automatically make the agency more profitable.

Like most agencies, Pyxl began using generative AI inside work that had historically taken significant manual production time. Research, exploration, concept development and other repeatable parts of delivery could happen materially faster.

But faster work creates a second question. If a project takes fewer hours to produce, what happens to the hours that disappear?

They can become capacity, margin, more senior thinking or more work with the same team. Or they can be absorbed into the economics of the business with nothing to show for it. Pyxl treated the answer as a decision rather than a by-product.

What changed

AI became part of delivery, not a side experiment.

AI moved from however individuals happened to use it into deliberate client delivery.

The objective was not to automate the agency. It was to identify the parts of the work where AI could materially reduce production time while keeping human judgment, client context and senior review in the places where they mattered.

That meant redesigning parts of the delivery process around a different assumption: the first useful version of the work could now be produced much faster.

Pyxl then adjusted delivery expectations, staffing, capacity, scope and commercial decisions around the new economics rather than the historical number of hours required to produce the work.

What happened

The work got faster. Production capacity stayed.

The most direct operating measure was the time between receiving a brief and producing the first concept. That fell 42%.

Pyxl did not respond by cutting production staff. There were no layoffs; the roles that fell away were overhead ones that were simply not refilled.

The productivity became capacity inside the existing team instead.

Reducing hours is an efficiency metric. What the business does with the recovered capacity is the commercial result.

Separately, custom agents reduced the amount of project-management and account-coordination work required to keep client teams informed and accounts on track.

What changed around the work

Custom agents reduced the project-management coordination required to run accounts.

Pyxl's AI work was not limited to producing client deliverables faster.

Pyxl Intelligence reads signals across the systems where account work already lives — including Asana, Slack, Drive and email — and surfaces client-health concerns, portfolio-level context and account information without requiring a project manager or account lead to assemble it manually.

Maybank built Pyxl Intelligence inside Pyxl as its internal implementation. Maybank Service Intelligence is the client-facing version configured for professional services firms. The three surfaces are shown at the end of this section.

That reduced the amount of coordination work required to:

• —chase status across systems
• —assemble portfolio views
• —surface account risk
• —prepare internal account context
• —relay information between teams
• —reconstruct what was happening on an engagement

The objective was not to remove client service. It was to improve it: give the people responsible for clients better information earlier while reducing the administrative coordination required to provide it.

That allowed Pyxl to reduce the need for some non-production project-management capacity while keeping production talent focused on client work.

The third place: operations

Utilization and project management stopped being a month-end exercise.

The first two places changed how the work was produced and how accounts were coordinated. The third changed how Pyxl ran itself: the operating questions every services firm answers weekly, and usually answers late.

Utilization, hours, overdue tasks, project status and budget against fee were all recoverable before. They were recoverable by a person, from several systems, on a cycle set by how long that took. Running those reads on a schedule changed the cycle, not the numbers, and a shorter cycle is what makes the decision actionable.

None of this is a claim about headcount. It is a claim about when a firm learns something it could act on. A utilization number that arrives on the eighth of the following month describes a month that is already finished.

This is also the least sector-specific of the three. The production work in place one looks different in a law firm than in an agency, and the account coordination in place two depends on how a firm is structured. Utilization, resourcing and project status look almost identical everywhere.

What reached the bottom line

Thirteen points of gross margin, eighteen points of EBITDA, and where the difference came from.

The 13-point expansion in gross margin reflects improved delivery economics. Because delivery costs sit in COGS, that improvement also flows through to EBITDA.

EBITDA margin expanded by 18 points in total, which means approximately five additional points of margin improvement occurred below gross profit through operating leverage — operating expenses either reduced, or growing more slowly than revenue.

Pyxl Intelligence — the operating-intelligence system Maybank built inside Pyxl — was designed to affect both sides of the operating model: making the work itself more efficient, and reducing the coordination and administrative friction surrounding that work. We are not claiming it produced all eighteen points, or that the incremental five came from any single expense category. What is accurate is that Pyxl changed its delivery model, its operating model and its use of AI over the period in which these results occurred.

Why Maybank exists

The difficult question comes after the efficiency appears.

Most firms begin with: where can we use AI? Pyxl learned that the more consequential question is: what changes in the business once the same work requires materially less time?

If nothing else changes, efficiency can simply create idle capacity or eventually become a lower price.

That is why Maybank works on delivery and economics together. Not because every engagement should look like Pyxl. Because every professional services firm that materially reduces the labor required to produce its work eventually has to answer the same question: who captures the value of the time that disappeared?