Available for orders from 15 Feb 2026.
Available for orders from 15 Feb 2026.
Let me start with a confession: I’ve sat in enough “AI transformation” strategy sessions to recognize the pattern. Someone puts up a slide with a pyramid on it. Another slide shows a robot next to a human, usually smiling at each other. A third slide announces that the firm is “embracing AI” and has “launched several pilots.” Everyone nods. The meeting ends. Nothing much changes.
That’s not transformation. That’s theater.
And the firms that mistake one for the other are going to be in serious trouble.
Because the future of service companies won’t be decided by whether they “adopt AI” — that phrase is already too weak for the moment. It’ll be decided by whether they’re willing to rebuild themselves around an entirely new theory of what they’re actually selling.
The old theory and why it’s running out of road
The old theory was elegant in its simplicity. Clients bought talent, scale, process, and reliability. Service firms organized effort better than the client could organize it internally. Revenue tracked headcount, programs, tickets, projects, and seats. The more ambitious firms layered in consulting and platforms, but underneath every P&L, the engine still ran on labor structures.
Think about what a mid-sized IT services engagement looked like in 2010. A bank outsources its application support. The vendor builds a delivery pyramid: a handful of senior architects onshore, a larger layer of project managers in the middle, and a wide base of developers and support staff offshore. The bank pays per resource, per hour, per ticket. The vendor makes money by optimizing that pyramid — keeping utilization high, attrition low, and billing rates defensible. It was a good model. For a long time, it worked.
But here’s the problem. When the fundamental economics of the work change — when a large fraction of that base layer can be handled by agents that don’t sleep, don’t need onboarding, and cost a fraction per task — the pyramid doesn’t just shrink. It loses its commercial logic entirely. You can’t take the same model, hire fewer people, and call it innovation. You have to rethink what you’re actually selling.
The new theory is harsher in some ways, but genuinely more promising: as raw intelligence gets cheaper, the scarce premium migrates. It moves into judgment, orchestration, trust, governance, enterprise context, and commercial accountability. In other words, value shifts from labor supplied to consequence arranged.
This is where the framework I lay out in The Agentic Advantage becomes genuinely useful — not as branding, but as migration logic. The book argues that service firms don’t transform in a single leap. They move through three distinct phases: Initiate, Adopt, and Excel. Each phase has its own logic, its own commercial design, and its own failure modes. Understanding which phase you’re in — and what it actually demands — is the difference between real transformation and expensive theater.
Phase one — Initiate: Choose your wedge carefully
I’ve seen firms try to boil the ocean. Launch fifty pilots. Explore every use case. Build an internal AI center of excellence that produces white papers nobody reads. It feels like momentum. It isn’t.
The Initiate phase, as The Agentic Advantage frames it, is not about experimentation at scale. It’s about wedge selection — finding the specific category of work where AI can transform economics fastest, with the least organizational friction, and the clearest commercial path.
The right wedge has a few reliable characteristics. The work is already outsourced, so the budget line already exists. The activity is repetitive and intelligence-heavy enough for AI to make a real difference. And crucially, the buyer is already purchasing an outcome rather than managing internal headcount — which means the conversation is about vendor substitution, not organizational redesign.
Think service desks. Finance operations. Procurement workflows. Claims handling. Compliance documentation. Application support. Knowledge operations.
Here’s a concrete example. A global insurance company outsources first-line claims triage to a services vendor — roughly 400 people reviewing incoming claims, categorizing them, flagging anomalies, routing them to the right adjusters. It’s important work. It’s also overwhelmingly pattern-recognition work. An AI-native delivery model could handle a large portion of that with agents: ingesting claims, extracting data, checking against policy rules, flagging outliers, and routing — faster, at a fraction of the cost, with consistent audit trails.
The critical distinction The Agentic Advantage draws here is between vendor substitution and organizational redesign. Replacing an outsourcing contract with an AI-native service is a procurement decision. Replacing internal headcount with AI is a political, cultural, and structural battle. The former moves much faster. That’s your wedge. That’s where the Initiate phase begins.
Phase two — Adopt: Don’t mistake decoration for redesign
This is where most firms will stumble — and honestly, it’s where I have the most sympathy, because the temptation is enormous and the pressure is real.
Leadership wants to show the board that AI is happening. So the firm deploys a few copilots. Productivity ticks up 15%. A case study gets written. The quarterly update sounds good. And underneath, nothing structurally changes.
That is not the Adopt phase. That’s putting new wallpaper on an old building.
Real adoption — what The Agentic Advantage describes as the core challenge of phase two — means redesigning the delivery model itself. Not augmenting it. Redesigning it. The agent retrieves, classifies, drafts, tests, monitors, reconciles, triages, and escalates. The human interprets, decides, intervenes, authorizes, and — critically — reassures. The workflow stops being “human work aided by software” and becomes “machine-led intelligence governed by human judgment.”
That is a different company. Different metrics, different management, different muscle memory.
A delivery leader at a mid-sized services firm recently tole me in plain terms : “We thought we were adding AI to our delivery. Then we realized AI was exposing every process we’d never bothered to document properly.” That’s the hidden cost of the Adopt phase — and also its hidden gift. The act of orchestrating agents forces you to understand your own delivery in ways that years of offshore optimization never required. You can’t orchestrate what you haven’t understood.
The other thing the Adopt phase reveals, consistently, is that AI doesn’t arrive in isolation. It pulls on data infrastructure, process design, operating model, and workforce architecture all at once. Firms that treat AI deployment as a standalone capability build tend to discover — expensively — that the real work is everything around the model: the context it can access, the controls that govern it, the humans who supervise it, and the telemetry that tells you when it’s drifting.
The Agentic Advantage is explicit about this: you are not deploying a tool. You are redesigning a production system.
Phase three — Excel: Sell the flying hour, not the engine
Here’s where commercial imagination starts to matter as much as technical capability — and where the best analogy I know comes from a company that makes jet engines.
In 1962, Rolls-Royce invented something called “Power-by-the-Hour.” The idea was simple and radical: instead of selling an airline an engine and then billing separately for every repair, Rolls-Royce would charge a fixed fee per flying hour. The engine, the maintenance, the replacement parts — all bundled into a single performance-based rate. You pay for the hours the engine actually flies.
The brilliance wasn’t the pricing mechanism. It was the alignment it created. Rolls-Royce and the airline now had exactly the same interest: keep the engine flying. Downtime was everyone’s problem. Performance was everyone’s incentive. The manufacturer stopped being a vendor and became a partner in operational reality.
That is precisely the shift the Excel phase demands of service firms. Stop billing for effort. Start standing behind throughput, reliability, cycle time, quality, and governed automation. Stop selling the engine. Start selling the flying hour.
Otis Elevator tells a similar story from a different angle. In its 2024 annual report, the company noted that services represented roughly 60% of sales but more than 90% of operating profit. Ninety percent. Once an installed base becomes mission-critical infrastructure, the economics don’t sit in the first sale. They sit in maintenance, optimization, and continuous performance. Enterprise AI will almost certainly follow the same pattern. The first wave of excitement is in licenses, pilots, and proofs of concept. The durable value pool will be in operating AI-enabled enterprises safely, continuously, and accountably — year after year.
The Agentic Advantage calls this the consequence layer: the point at which a service firm stops being evaluated on activity metrics and starts being evaluated on business outcomes the client actually cares about. Cycle time reduced. Error rates eliminated. Decisions accelerated. Exceptions governed. Risks surfaced before they become losses. That’s the Excel phase — and it’s also where the commercial relationship becomes genuinely hard to displace.
The execution gap is where the money is
Here’s a number worth sitting with: Gartner warned in June 2025 that more than 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value, or inadequate risk controls. Nearly half.
The reason, as Gartner’s analysts put it, is that companies are treating AI as spectacle rather than operating discipline. They underestimate what it actually takes to deploy agents at scale in a real enterprise — with real data governance, real compliance requirements, real exception rates, and real humans who need to trust the outputs before they’ll act on them.
I find this clarifying rather than discouraging. A market that is strategically important but operationally immature doesn’t reward the loudest evangelists. It rewards the disciplined operators. The firms that can make a volatile technology feel, in the best possible sense, boring — reliable, repeatable, observable, auditable, governable.
This is the through-line of The Agentic Advantage: the firms that will win this era are not necessarily the ones with the most advanced AI. They’re the ones with the most advanced operating discipline around AI. The ones who’ve figured out how to move through all three phases — Initiate, Adopt, Excel — without getting stuck performing transformation for an audience rather than actually delivering it.
Think about what disciplined execution actually requires. Someone has to build the context layer — the enterprise knowledge graph that tells the agent what matters in this organization, with these policies, in this industry. Someone has to design the controls — what the agent can do autonomously, what it must escalate, and what it must never touch. Someone has to instrument the process — so that when something goes wrong at 2am on a Friday, the right human gets the right alert with the right information to make the right call. Someone has to structure the commercial relationship so that the buyer is paying for outcomes, not activities, and the incentives stay aligned over time.
That is not software work. That is services work — reimagined, but recognizably services.
The real moat is orchestration
Here’s what I keep coming back to: in the agentic era, models will proliferate. Tools will multiply. Agents will become table stakes. Every firm will have access to roughly the same AI capabilities at the foundation layer.
The scarce layer — the one that generates defensible margin — is orchestration. Where do agents act? What can they see? When do they escalate? How are they evaluated? How are they retrained when they drift? How do humans intervene without breaking the flow? How is the result tied back to something the buyer actually cares about?
There’s a useful analogy in how Formula 1 teams operate. Every team on the grid has access to similar engines, similar regulations, similar tire compounds. The differentiation comes from the pit crew choreography, the race strategy, the telemetry systems, the judgment call made by the strategist on lap 34 about whether to pit now or stay out. The team that wins isn’t the one with the fastest individual component. It’s the one with the best orchestrated system — and the best instincts about when to override it.
The Agentic Advantage makes a similar argument about service firms: the competitive moat in the agentic era is not the model you access, it’s the orchestration you’ve built — the reusable assets, domain playbooks, policy layers, context graphs, assurance systems, and telemetry that make your delivery superior to anything a client could assemble on their own. That’s the asset. That’s what compounds over time. And crucially, that’s what a competitor — whether incumbent or AI-native — can’t easily replicate.
Service firms that internalize this will stop thinking of AI as something that reduces their role. They’ll start thinking of it as something that makes the orchestration role more valuable, not less.
Two futures, clearly visible from here
The firms that fail will use AI defensively. They’ll talk about productivity, trim some effort from the bottom of the pyramid, protect near-term margins, and call the job done. Their demos will get better. Their fundamentals will quietly erode. Over time, they’ll get squeezed from both sides: clients demanding lower costs, and AI-native challengers attacking narrow service lanes with software economics.
I’ve watched this movie play out in adjacent industries. Think of the travel agencies that added a website in 2001 and thought they were digital. Or the newspapers that launched apps in 2010 and thought they were platform companies. The form changed. The model didn’t. And the market noticed — slowly, then all at once.
The firms that win will move differently. They’ll use the Initiate-Adopt-Excel arc not as a PowerPoint framework but as an actual operating roadmap — knowing which wedges to enter, how to redesign delivery at each stage, and when to shift commercial models from effort to consequence. They’ll build reusable assets — domain playbooks, policy layers, context graphs, assurance systems, telemetry — into the heart of how they deliver. They’ll retrain their managers not just to supervise people, but to supervise systems of people and agents. They’ll shift their commercials from effort metrics toward consequence metrics. They’ll turn trust into infrastructure. And when they do, something quietly remarkable happens. They stop looking like service vendors in the traditional sense. They start looking like the invisible operating architecture of the enterprise itself. The thing that makes everything else work.
That’s the real opportunity here. Not decline. Not commoditization. Not a long retreat before the software wave.
A reinvention — in which the best firms become the machinery through which enterprise intent becomes reality. Where they take the chaos of transition, the abundance of machine intelligence, the stubbornness of institutional process, and the fragility of real-world execution, and compose them into something dependable.
That’s what the three phases of The Agentic Advantage are ultimately in service of. That’s what selling orchestrated business consequences actually means.
And that is how service firms — the ones willing to rebuild, not just redecorate — become indispensable in this era.
Part 1 of the article is here.
”Sadagopan Singam is a global business and technology leader and the author of Agentic Advantage. He advises boards and executive teams on GenAI-driven transformation and autonomous enterprise models.”
