There’s a question that should be keeping every IT services CEO up at night:
When your clients ask about AI — and they’re all asking — who are you sending them to?
If the answer is a Big Four consultancy, a boutique AI firm, or a vague “we’re working on it,” then revenue is walking out the door. And it’s not coming back.
The irony is that IT services companies — managed services providers, outsourcing firms, technology consultancies — are better positioned to deliver AI value than almost anyone in the market. They already have the client relationships. The infrastructure access. The data adjacency. The trust.
They just haven’t built the practice yet.
The capability layer, not the new business unit
The most common misconception about launching an AI practice inside an IT services company is that it requires hiring a team of data scientists and building a separate business unit with its own P&L, its own sales motion, and its own brand.
That model fails more often than it succeeds. It’s expensive to build, slow to generate revenue, and disconnected from the very asset that makes an IT services company valuable: the existing client relationship.
The better model is a capability layer — AI expertise embedded inside the services you already deliver.
A helpdesk team that uses AI to classify and route tickets, reducing resolution time and volume — that’s not a new product. It’s a billable service improvement on an existing contract.
An infrastructure team that deploys predictive analytics to identify potential outages before they happen — that’s not a separate AI engagement. It’s a premium SLA tier that commands higher margin on a relationship that already exists.
A client success team that uses AI-driven analysis to identify expansion opportunities — upsell triggers, underserved workflows, unaddressed pain points — that’s not a new sales channel. It’s an intelligence layer on top of a book of business that’s already generating revenue.
Each of these lives inside existing contracts. Existing relationships. Existing trust. The AI practice doesn’t replace the managed services business. It makes it more valuable, more differentiated, and harder to displace.
The pricing trap
Pricing AI services is where most new practices stumble, because the instinct is to package AI engagements the way the company already packages everything else.
Fixed-fee engagements sound clean and client-friendly. But launching a fixed-fee AI practice before understanding actual delivery costs is a margin trap. The first three engagements will teach more about scope, complexity, and client readiness than any planning exercise ever could.
The smarter sequence: start with time-and-materials. Treat the first engagements as learning investments — for both the provider and the client. Measure actual effort. Document actual outcomes. Build an internal cost model based on reality, not projections.
Then, once the delivery data exists, transition selectively to outcome-based pricing. “We reduced your ticket volume by 35% — here’s what that’s worth” is a fundamentally different conversation than “here’s our hourly rate for AI consulting.” One sells effort. The other sells impact.
The companies that get to outcome-based pricing fastest will own the market. But the ones that skip the learning phase to get there will lose money on every engagement until they figure out what they should have measured from the start.
The talent question everyone overthinks
“We need to hire 20 data scientists.”
No. Not on day one. Probably not on day 200.
What an AI practice needs at launch is one senior leader who sits at the intersection of technology and P&L. Someone who can train the existing team to deliver AI-enhanced services. Someone who can build a repeatable methodology. Someone who can walk into a client conversation and speak both the language of the technology and the language of the business outcome.
Critically, that leader also needs the judgment to say no. Not every client is ready for AI. Not every use case delivers ROI. Not every enthusiastic pilot deserves a production deployment. The ability to decline an engagement that won’t deliver is what separates a sustainable practice from a hype cycle.
The existing team — the engineers, the service desk analysts, the infrastructure specialists — already understand the client environment better than any outside AI consultant ever will. They know where the data lives. They know which processes are manual. They know which workarounds have been in place for years. That institutional knowledge, combined with AI methodology and guided by senior leadership, is the talent model that works.
Hiring data scientists becomes relevant later, once the practice has proven its model with pilot clients and needs to scale delivery. Building a team before building demand is the most expensive mistake in professional services.
The complacency differentiator
There’s a temptation in AI services to sell the magic. The demo is impressive. The client is excited. The contract gets signed on enthusiasm.
And then the deployment meets reality.
The most durable AI practices aren’t built on excitement. They’re built on trust. And trust in AI services comes from an unexpected place: teaching clients to question the AI.
This connects to a theme that’s been running through the broader AI leadership conversation — algorithmic complacency, the uncritical acceptance of AI output that erodes human judgment over time. In a client-service relationship, complacency is a retention risk disguised as satisfaction. The client who accepts every AI output without questioning it is the client who blames the provider when the model eventually gets something wrong — because nobody taught them what the model’s limitations were.
The IT services company that builds critical AI literacy into its delivery model — that trains clients to understand when to trust the output and when to challenge it — earns a fundamentally different kind of loyalty. Not the loyalty of dependence. The loyalty of respect.
That client doesn’t churn when a competitor offers a flashier demo. They stay because they trust the judgment behind the service, not just the technology in front of it.
Hype acquires clients. Trust retains them.
The sequencing that works
The temptation is to launch big. Press release. Dedicated landing page. Conference keynote. “We are now an AI company.”
That model has a failure rate that should embarrass the industry.
What works:
Start with two pilot clients. Choose existing clients where the relationship is strong and the data environment is understood. Pick use cases that are unglamorous and high-ROI — document processing, ticket classification, predictive alerting. Prove delivery. Measure outcomes. Document everything.
Build the methodology from reality, not theory. The pilot engagements will reveal what the delivery model actually looks like — resourcing, timelines, client readiness requirements, governance needs. Codify those lessons into a repeatable framework before scaling.
Then scale with proof. The third client conversation is fundamentally different when it starts with “here’s what we delivered for two organizations like yours” instead of “here’s what we think we can do.”
The IT services companies that figure this out in the next 12 months will own a new market segment. They’ll expand existing contracts, increase margins, and build a retention moat that pure managed services can’t match.
The ones that don’t will watch their clients buy AI from someone else. And eventually, those clients will buy their managed services from that someone else too.
The clock is running.
Gabor Szentivanyi is a Transformational CIO and Certified AI Consultant with 25+ years of enterprise IT leadership across manufacturing, higher education, consumer goods, and life sciences. He works at the intersection of AI strategy, IT governance, and organizational change across three continents. This article is part of an ongoing series on AI leadership and algorithmic complacency.