By Robert Cesafsky

In the past two years, the professional services industry has spent billions of dollars on AI. The return on investment for most organizations remains elusive. Leaders are frustrated. Boards are skeptical. And the dominant explanation—that AI just needs more time to mature—is starting to wear thin.

Here’s the harder truth: Most organizations are failing at AI because they’ve misdiagnosed which problem they’re trying to solve.

Across virtually every professional services organization I’ve met with in the past year, I see the same error playing out: Leaders treat AI as a single capability to be deployed across the business. In reality, their business requires two fundamentally different kinds of AI—and conflating them is killing their returns.

The Two Operations Hiding Inside Every Services Organization

Every professional services business is running two distinct operations simultaneously.

One is services delivery—the work your organization sells to clients: consulting, analysis, implementation, advisory. When a team uses AI to accelerate requirements gathering, synthesize research, or draft a client-facing deliverable, that’s services delivery AI at work. The client sees the output. It defines your product.

The other is services management—the operational infrastructure that makes delivery possible: resource allocation, project margin tracking, billing, revenue recognition, renewal management. Clients never see this work, but its failure cascades immediately into everything they do see.

These problems are not the same. They require different AI architectures today and, increasingly, different agentic strategies tomorrow. But most enterprises are deploying one type and expecting both results.

Why the Two Domains Demand Different AI

The distinction matters because the nature of intelligence required by each domain differs in a critical way.

Services delivery AI works at the edge of human expertise. It augments practitioners’ skills, accelerates research, and amplifies judgment. Large language models are well suited here, as they are broad, generative, and flexible. A senior consultant may own the recommendation; AI helps the consultant reach the recommendation faster. Probabilistic outputs are acceptable because human judgment remains the final layer of validation.

Services management AI operates inside a deterministic system. There is no acceptable margin of error when calculating project profitability, enforcing billing rules, or triggering revenue recognition.

The time employees spend reviewing and correcting probabilistic AI outputs costs organizations thousands of dollars per employee per year. That’s a verification tax that quietly erases whatever efficiency gains the AI was supposed to deliver. When senior project managers and finance controllers are manually auditing AI-generated margin calculations and billing schedules, the operational promise has already unraveled.

AI positioned as an external advisor—one that observes and flags issues from outside core systems—runs into a structural ceiling in services management. Where finance, delivery, and customer commitments are tightly interlocked, operational authority requires operational integration.

The Misdiagnosis and Its Cost

The scale of the problem is now well-documented: 95% of generative AI pilots in a recent MIT study failed to deliver measurable bottom-line impact. And only about 6% of organizations qualify as high performers, seeing real earnings before interest and taxes (EBIT) while also reporting significant overall value, according to McKinsey.

The pattern is consistent enough that it has a name: pilot purgatory. And while the causes are numerous, a common thread runs through nearly every stalled implementation: An organization deployed AI as a horizontal layer across the business rather than building it differently for fundamentally different operational domains.

The SPI Research 2026 Impact of AI on Professional Services report identifies where the misdiagnosis takes root. Data quality has ranked as the greatest barrier to AI adoption for the past two years. When the underlying data infrastructure is fragmented—when finance, delivery, and customer systems don’t share a unified data model—AI has no solid foundation on which to operate. Organizations end up anchored in document-adjacent lightweight use cases rather than the operational workflows where the real value lives.

What Getting It Right Looks Like

The enterprises pulling ahead are making a deliberate architectural choice across both domains.

On the delivery side, they’re investing in AI that equips practitioners with faster research, stronger synthesis, and better client output, with human judgment anchoring every consequential decision. On the management side, they’re investing in AI agents that autonomously execute operational infrastructure: staffing workflows, project risk monitoring, billing cycles, and revenue recognition.

High-performing organizations are nearly three times as likely as their low-performing competitors to have fundamentally rebuilt their workflows around AI rather than layered AI on top of existing systems, according to McKinsey’s 2025 state of AI report. That finding reflects a different theory of what AI is actually for. It’s not a productivity layer you apply to existing workflows but an operational framework you build around. In services management, layering probabilistic AI on top of fragmented systems produces the verification tax. But rebuilding workflows around deterministic, domain-specific AI produces the margin.

The compounding effect on the management side deserves more attention than it typically gets. Reclaiming 20 hours per month per project manager reallocates your most expensive human capital toward high-value client outcomes. Multiply that across an entire delivery organization and the productivity story becomes a strategic one—a fundamental shift in how that organization deploys talent and creates value—and one that closes the gap between AI investment and AI return.

Where the Two Sides Begin to Converge

Treating delivery and management as separate domains is the right starting point. But it’s not the final destination.

The next frontier in AI for professional services is true autonomy: agentic AI that executes complex multi-step workflows without human intervention at each decision point. Reaching that frontier requires something neither delivery AI nor management AI can provide on its own: a shared context layer that draws intelligence from both sides simultaneously.

Think about what fully autonomous resource staffing actually requires. An AI agent making a staffing decision needs to understand not just who is available on the management side but also what skills the current client engagement demands, how the project is tracking against scope, and what client relationship dynamics are at play—all of which live on the delivery side. Now run the logic in reverse: An AI agent supporting client-facing delivery needs to understand capacity constraints, margin targets, and contractual commitments that exist only in the management layer.

This is why the most forward-looking enterprises are already investing in a unified data and intelligence layer that bridges the two domains. The service operations platform serves as the system of record for deterministic management AI, and this system must feed the context layer powering autonomous delivery AI. Building that connective tissue requires capturing data from both sides with equal intentionality, even as the two domains continue to operate through different workflows and demand different AI architectures in the near term.

The practical implication for leaders is straightforward. Build the two sides differently today toward an inevitable convergence. The enterprises that treat delivery and management as permanently siloed will hit an autonomy ceiling. But those that design for eventual integration now, with clean data and integrated systems on both sides, are laying the foundation for a fundamentally different kind of services operation.

Robert Cesafsky is the Chief Operating Officer at Certinia, a leading provider of AI-powered Professional Services Automation (PSA) serving global enterprises including Salesforce, Dell, and Siemens.

---

To see how leading services organizations are building the unified context layer described in this article, explore Veda: Certinia’s AI engine for services operations, combining specialist agents and intelligent actions grounded in decades of institutional knowledge to shift services organizations from reactive workflows to an agentic operating model.