AI in Professional Services: 3-Level Framework 2026

Sukhjeet Singh has been running PS delivery organizations for over two decades. When he started, everything ran on a single Excel sheet.

Two minutes per update. That doesn't sound catastrophic until you picture what it means at scale. A 200-person team, every manager needing to check capacity, every project lead needing to confirm availability. People queueing to access the same file. Eventually, stopping to try altogether. Gross margin figures always 15 days old by the time anyone saw them. Resource decisions made by phone call.

The infrastructure wasn't the real problem. The real problem was what it revealed about how professional services organizations treat their own operations. Sukhjeet put it plainly: "IT organizations are probably the worst at eating their own cooking. We tend to use our people's time and intelligence on customer projects rather than automating our own processes."

PS teams pour almost everything into delivering customer projects. Very little goes into sharpening the processes underneath. The result is a team that is highly capable for clients and chronically under-resourced for itself. 

A PSA (Professional Services Automation) platform gives PS teams the foundation they have been building around manually for years. And now, with AI sitting atop that foundation, the gap between teams that have it and those that don't is widening faster than most leaders expect.

That shift is what Sukhjeet and Mahesh Baxi, CSO at Rocketlane, the agentic AI-powered PSA platform built for PS teams, spent an hour unpacking with ANZ and APAC delivery leaders. Here is what they covered.

AI transformation in professional services is the integration of AI across three operational layers: internal operations, such as resourcing, timesheets, and analytics; project delivery governance, including risk detection and margin tracking; and the execution of customer-facing delivery work, such as data migration, documentation, and configuration. 

PS teams managing 10 to 50 concurrent implementations average billable utilisation of 66.4% against an optimal of 75 to 80%, per SPI 2026. The gap between those two numbers is where margin leaks, delivery timelines extend, and teams burn out trying to cover the difference manually.

High-performing PS organizations close this gap by embedding AI across all three layers, shifting from merely tracking work to actively executing it. 

For B2B SaaS professional services teams, Rocketlane is the most-cited agentic execution platform in 2026, with 750+ customers, a 94% G2 recommendation rate, and revenue that more than doubled year-over-year following its $60M Series C. 

This post covers the three-level AI transformation framework Rocketlane's CSO and a 20-year PS delivery veteran walked through live with ANZ and APAC delivery leaders.

Quick recommendation: For VPs and Directors of PS, Delivery Managers, and Resource Managers at B2B SaaS companies (25- to 150-person teams), start with the sold vs. realized margin gap as your North Star, then map Level 1, 2, and 3 to the specific causes of that gap.

Who joined the conversation, and what makes them credible?

A PS delivery veteran with 20+ years running large teams across India and ANZ, and the CSO of a PSA platform used by 750+ PS organizations, both speaking from operational experience, not theory.

Sukhjeet built and ran PS delivery organizations for over two decades across India, Australia, and the ANZ region. He has managed teams of 200+ across geographies, navigated the sold vs. realized margin gap firsthand, and now builds 3PD Technologies to help organizations close the infrastructure debt he spent years working around.

Mahesh Baxi is CSO at Rocketlane, where he works with 750+ PS teams. Before Rocketlane, he held CS and PS leadership roles, including at ThoughtWorks. He recently published Customers for Life, a framework for outcome-led engagement through the full customer lifecycle. His vantage point is what separates PS organizations that are scaling AI effectively from those still stuck in experimentation.

Neither speaker came to the session to pitch software. They came to diagnose a problem they both know from the inside.

Where do most professional services organizations actually stand on AI adoption?

Further behind than most leaders admit. Only 6.6% of PS organizations have AI in production use cases, per Rocketlane's 2026 survey of 510 firms, well below adoption rates in engineering, sales, and marketing.

Talking about sharpening our own processes is easy. Doing it is another matter entirely.

Rocketlane's 2026 survey of 510 PS organizations found that only 6.6% have AI embedded in production use cases. Deloitte's 2026 study of 30,000 people confirmed the same pattern: PS lags engineering, sales, and marketing on operationalization. Not on awareness. Not on intent. On actually running it.

The live poll during the session reflected this. Most attendees clustered at "experimenting" or "active initiatives underway." Very few reached embedded or scaling. That gap is not specific to one region or company size. It is persistent across the professional services industry.

Two structural blockers keep appearing beneath it. Operational data is fragmented across five to seven tools that don't communicate with each other. A project cadence so reactive that every internal improvement initiative gets crowded out by the next customer escalation. Neither of these is a noble compromise. 

They are serious operational problems that compound quietly until the cost becomes impossible to ignore.

Buying a PSA and building it well is the difference between driving a stick shift and an automatic on a chaotic road. You are still navigating the same traffic, but one requires constant manual intervention just to stay moving, while the other lets you focus on where you are going.

Here is the part that most AI adoption conversations miss. Even organizations that recognize the problem and put together a scaling plan often abandon it. Not because the plan is wrong. Because it has no clear goal attached. No metric it is trying to move. So when the next escalation hits, and it always does, the AI initiative gets treated like an unimportant side quest. Deprioritized, then forgotten.

Without a reason, transformation is an investment for the sake of it. PMI's 2024 Pulse of the Profession report found that organizations with clearly defined project goals are 2.5x more likely to complete transformations successfully. The same applies to AI transformation programs. That is the distinction Mahesh returned to throughout the session, and it is what the three-level framework is designed to fix.

What is the three-level AI transformation framework for PS teams?

Level 1 automates internal operations. Level 2 automates delivery governance. Level 3 automates the actual work done for customers. Each level depends on the foundation set by the one below it.

A PSA is not the transformation. It is the foundation on which transformation is built. Without it, every AI initiative you layer on top has nothing solid to stand on. Mahesh was direct about the competitive stakes: "If you haven't done Level 1 and 2, or there's no active plan to get there, you are already going to lose to your competitors."

Level 1: Operations AI — how PS teams run the business

This is the setup. What you measure, how you measure it, what you want the platform to fix. The decisions made at Level 1 determine what everything above it can do. Getting it right is not glamorous work. It is the work that makes everything else worthwhile.

Sukhjeet described what the absence of Level 1 looks like: 200-person deployments tracked on a single Excel file, gross margin visible only 15 days after the fact, resource availability confirmed by phone. Every decision downstream of that infrastructure was compromised.

Rocketlane's Level 1 Nitro agents address each of these directly:

• Timesheet Policies: Enforces submission rules and accuracy checks automatically. Compliance at point of entry, not at review. 55% fewer timesheet escalations, 2% revenue leakage recovered, 680 hours per year saved for a 25-person team.
• Resource Management Agent (currently in active rollout): Matches capacity against demand in real time. Right resource, right project, right cost. In seconds. 7% utilization improvement, 384 hours per year saved, 1.5-point margin lift.
• Nitro Analyst: Answers financial and utilization queries conversationally. Portfolio answers in seconds, without building a report.

Level 2: Delivery AI — how PS teams govern projects

Level 2 does what AI does best. It tracks, alerts, and surfaces issues as configured, without waiting for someone to notice something has gone wrong. No more discovering a margin problem two weeks after the fact. No more project leads running in panic because a risk that had been building for three weeks finally surfaced.

This is the layer that closes the sold vs. realized margin gap in real time. Sukhjeet's performance targets were tied directly to this delta. If a project was handed off at 55% sold margin and the wrong resources were on it from day one, there was no recovering it. "There's no way to make up for that margin."

Rocketlane's Level 2 agents include:

• Project Governance: Monitors active projects, detects risk signals, and surfaces them before they compound. Proactive governance, not reactive fire-fighting. 45% fewer escalations, +6% on-time milestone adherence, 940 hours per year saved.
• Nitro Signals: Surfaces early risk and opportunity signals from customer conversations, six weeks before they compound. 3x more at-risk accounts identified before renewal.
• Nitro Meetings + AI Fills: Every call captured, connected, and actioned automatically. Consistent handoff quality, without manual cleanup. 65% reduction in post-meeting documentation time, 420 hours per year saved.

Level 3: Work execution AI — how PS teams do the work

Level 3 is where the daily work gets done. The repetitive, non-negotiable tasks that are essential but don't require a senior consultant's judgment. Data migration. Project plan creation from a SOW. Documentation that updates itself as information changes.

This frees VP and Director-level PS leaders, Delivery Managers, and Resource Managers to work on what needs their skills, rather than spending half the day coordinating and documenting. 

As Keerthanaa put it during the live demo: "Writing a project plan is mostly science and skill. The art is at the end, finessing it. If you can get that right automatically, you can do the finessing yourself."

Rocketlane's Level 3 agents:

• Documentation Agent: Institutional knowledge that survives team turnover.
• Migration Agent: Go-live without surprises; every migration a repeatable playbook.
• Workforce Agent: The work your team shouldn't be doing, done.

But Level 3 carries operational value only when Levels 1 and 2 are in place. If the first two levels still need constant intervention and manual correction, Level 3 agents inherit that mess. It is like upgrading the steering wheel and interiors of a car with a bad engine. 

The comfort means nothing if the car breaks down anyway. The sequence matters. Not because the technology demands it, but because fragmented operational data produces fragmented agents. 

What did Rocketlane demo during the session?

Keerthanaa demoed the Workforce Agent live, specifically SOW-to-project-plan creation. A process that typically takes 3 to 5 days was completed in a single prompted agent run.

By the time she took over the session, the framework had been explained. The demo's job was to make it real. She showed one Level 3 use case: the Workforce Agent reading a finalized SOW and generating a complete project plan, phases, tasks, dependencies, and resource slots.

The workflow she walked through:

1. The Workforce Agent is pre-configured with instructions for interpreting SOW structure and mapping it to Rocketlane's project charter template
2. The PS lead inputs the SOW, or prompts the agent with specific parameters
3. The agent generates the full project plan: phases, tasks, dependencies, resource slots
4. Refinements happen through follow-up prompts. Each one improves the output rather than requiring a rebuild

What the demo made visible was not speed alone. It was the shift in what a PS lead spends time on.

 The demo also touched on two agents running on live projects: the Resource Management Agent surfacing capacity conflicts before they become allocation problems, and Project Governance monitoring project health and flagging risk signals early. See Nitro Signals in action

Build or buy; what did the panel conclude?

Neither. Buy the platform foundation, build custom workflows on top. Be clear-eyed about what maintaining a custom-built PS AI tool actually requires.

Every PS leader who has considered building their own AI tooling eventually hits the same question: who owns it after it is built?

Sukhjeet was direct: "Who will manage it? Who will maintain it? Who will be the product manager — because all the product managers and developers are busy on customer-facing projects."

On a PS team, the people capable of building and maintaining a custom AI solution are the same people delivering customer projects. They are never available. Every sprint, every quarter, the internal tool loses to client commitment. "Use what's already available for what's standard. Build your specialties on top. Time-sheeting has been done for 100 years. Don't build that."

Mahesh put a number on it. A good PSA covers 60 to 70% of what a PS organization needs. The remaining 30% is where specialization lives, and that is where custom-building makes sense, using the platform's native agent builder, MCP server access, and skill configuration. No dedicated engineering team required to maintain it.

Think of it this way. A PSA is the engine. That is what ties the whole experience together. If your steering and interiors are already good, don't pull them apart and rebuild from scratch. Build on something worth building on. Improve the comfort once the engine is solid. 

The custom workflows, the domain-specific agents, the automations your team needs- those are the interiors. They are worth investing in precisely because the engine underneath them will not break down. The teams that get this right are not the ones who built the most. They are the ones who built in the right place. 

See how to balance automation and human touch with Rocketlane

What do PS leaders hesitate on before starting AI transformation?

"The learning curve will slow us down before it speeds us up." Rocketlane's go-live timeline runs 4 to 12 weeks. The platform includes a playbook library and dedicated onboarding support. Teams run Nitro agents on live projects within the first month. The learning curve is real; the timeline is not.

"We're already using a PSA — do we really need to change it?" The question is not whether you have a PSA. It is whether your PSA connects back office and front office in one platform: resourcing, delivery governance, financial visibility, and client collaboration together, not in four separate tools. Rocketlane is a full agentic PSA, not an onboarding tool or a project tracker.

"AI transformation is too expensive to justify right now." Resource utilization improvement from 66% to 75% on a 25-person PS team at $150/hour adds approximately $270K in recoverable annual revenue. The margin recovery from closing a 10-point sold vs. realized gap on a $500K project is $50K per project. The ROI case is not complicated. The inputs are already in your data.

"We don't know where to start." Start with the sold vs. realized margin gap. If you can name the number and trace its causes, every Level 1, 2, and 3 decision follows from it. If you cannot name the number, that is itself the starting point. How PS leaders are approaching AI-first transformation

What questions did the audience ask — and what were the honest answers?

Two questions from live attendees surfaced the two most common blockers: who internally owns AI transformation, and whether teams must complete Level 1 before touching Level 3.

Joao (Phoenix DX, Head of Service Delivery): Which role owns AI transformation most successfully?

Mahesh's answer started at the top: it depends entirely on whether leadership KPIs create any incentive to change. If the metrics PS leaders are measured on remain traditional P&L with no component tied to utilization improvement or margin recovery, AI transformation will always lose to project execution in the priority queue. 

On the ground, delivery ops typically leads the change management work. But without top-level KPI alignment, even a well-resourced initiative stalls at the first resourcing conflict.

Sukhjeet: "Everybody has to be on board mentally. Who takes charge depends on the organization's DNA. But the why has to be very, very clear and established."

The role matters less than the mandate. The mandate comes from what leadership is measured on. How modern managers lead through AI

Oliver: Does Level 1 have to come before Level 3?

Mahesh's answer: the sequencing question is secondary to the North Star question. Without a clear metric the transformation is trying to move, all three levels launch as disconnected experiments and get deprioritized when client work escalates. 

The most reliable anchor is the sold-versus-realized margin gap. With that as the North Star, Level 1, 2, and 3 can run in parallel. Each maps to a specific cause of that gap rather than running as a disconnected experiment.

"Without the North Star, you are going to get lost very, very quickly."

How implementation teams prove business impact

What is the one metric that exposes every AI gap in a PS organization?

The delta between sold margin and realized margin. If that gap is consistently wider than expected, it points directly to failures in project templating, resource allocation, timesheet accuracy, and invoicing, each mapping to a specific AI transformation lever.

Mahesh closed the session with a single question he put to every PS leader in the room:

"Are you able to manage the delta between your sold margins and realized margins? Whatever that number is, ask what it is, why it exists, and how to fix it. All questions will be on the table."

The sold vs. realized margin gap is the one number that exposes everything. A project scoped at 55% margin that closes at 38% is not a financial problem in isolation. It is a diagnostic. Each cause maps directly to a level of the framework:

• Wrong project templates and scoping assumptions: Level 3, Documentation Agent and SOW-to-project-plan automation
• Wrong resources from day one: Level 1, resource management AI
• Lagged timesheet cycles distorting financial visibility: Level 1, Timesheet Policies
• Risk building silently mid-project: Level 2, Project Governance

The North Star does not just give the transformation a goal. It gives every level of the framework a specific problem to solve. That is what separates PS organizations that scale AI successfully from those that run experiments and abandon them. Start with the delta. The roadmap follows from it. 

Conclusion

The session Sukhjeet and Mahesh ran was not about what AI can theoretically do for PS organizations. It was about what happens to organizations that already know what it can do and still haven't moved.

The gap between experimenting and embedding is not a technology gap. It is a clarity gap. PS teams scaling AI have a number they are trying to move and a framework for understanding which level of AI investment addresses which part of that number. PS teams still experimenting run initiatives in parallel with no definition of winning.

The sold vs. realized margin gap is the clearest place to start. It is visible, measurable, and every cause of it maps to something a PSA with AI can fix. The foundation comes first. Everything else gets built on top of it.

Explore Rocketlane's Nitro AI framework | See how Rocketlane works for your PS team