Stop Buying AI Tools. Redesign How Work Gets Done.

The uncomfortable truth about AI in services

If you run a service or BPO business, the last 18 months have felt like a blur of AI pitches.

Every vendor promises 30–50% savings. Your teams experiment with copilots. You approve a few pilots. Dashboards light up with “AI usage.”

And yet, when you look at EBITDA, customer experience, or headcount, not much has changed.

You’re not alone.

• PwC’s 2026 Global CEO Survey finds that only 12% of CEOs say AI has delivered both cost and revenue gains; 56% report no gain at all, despite heavy investment.
• An operator guide from InflectionCX, synthesizing S&P, MIT and Gartner data, reports that 42% of AI initiatives are abandoned before production; MIT estimates 95% of pilots have zero measurable P&L impact.
• Capgemini’s research in property & casualty insurance shows that “trailblazers” who treat AI as a core operating capability see up to 21% higher revenue growth and ~51% more share‑price increase—while 42% of insurers track no AI metrics at all, and 55% say it’s unclear who even owns AI.

Across industries, the pattern is the same: the technology mostly works. The operating model around it does not.

For a service CEO or COO, this is the signal to stop thinking about “which AI tools to buy” and start asking a harder question:

What would our operating model look like if we designed it assuming AI was available and reliable from day one?

The companies who answer that question are the 12%. Everyone else is stuck in what one PE operator calls the “AI license trap.”

The AI license trap: motion without impact

FTI Consulting’s private-equity radar, summarized by Zenon, names a pattern most service leaders will recognize:

“Buy licenses. Hand them out. Hope productivity shows up in EBITDA. That is fine for experimentation. It is not a transformation strategy.”

In other words: we’ve treated AI like another SaaS rollout.

That approach creates three predictable failure modes in service companies:

1. Activity replaces outcomes.

   • Leaders track AI usage, prompts, or “tickets touched,” not resolution, margin, or repeat-contact avoidance.
   • InflectionCX calls this the “Cobra Effect” metric trap: if you reward containment (shorter handle time or deflection) instead of resolution, you get conversations with zero customer value.

2. Pilots lie.

   • Sayali Patil, an AI leader at Cisco/Splunk, puts it bluntly: “The pilot often works because someone cleaned the data before the demo, but once leadership approves the production rollout, that manual cleanup does not follow.”
   • Gartner expects 60% of AI projects to be abandoned by 2027 for lack of AI‑ready data. Mid‑market firms typically have “three to five core systems that have never been properly reconciled.”

3. Governance is an afterthought.

   • Shadow AI is now normal: IBM estimates shadow‑AI breaches cost $670K more and take 10 extra days to contain; BlackFog finds 49% of weekly AI users are on unsanctioned tools.
   • Boards are waking up: Fortune/Sedgwick report 70% of boards now have AI risk committees, but only 14% say they’re fully ready. Regulators are moving too—the EU AI Act carries penalties up to 7% of global revenue for certain violations.

The common thread: none of these problems are model problems. They’re operating‑model, data, and governance problems.

If you’re approving AI line items without changing how work actually flows—or who is accountable for outcomes—you are funding motion, not impact.

Redesign, not retrofit: what the data says actually works

The most valuable insight from the recent research is also the simplest:

Organizations that redesign end‑to‑end workflows for AI are about twice as likely to achieve financial returns.

InflectionCX’s analysis of contact center and BPO transformations shows:

• Companies that deploy AI into unchanged workflows mostly see modest efficiency (10–20%) and high failure rates.
• Those that redesign intake, triage, handoffs, and escalation around AI are 2× more likely to report real financial returns.

Capgemini’s insurance study adds more detail:

• Trailblazers treat AI as a core operating capability, aligning strategy, talent, tech foundation, and adoption simultaneously.
• They invest roughly 72% of AI spend into change, people, and operating-model work and only 28% into pure technology.
• Followers do the opposite—spending the bulk on tools—and they stall in proof‑of‑concept.

PwC distills the same point:

“Technology delivers only about 20% of an initiative’s value. The other 80% comes from redesigning work.”

That 20/80 split is the crux for service companies.

Most current AI budgets are inverted: 80–90% on licenses, pilots, and vendors; 10–20% on redesigning how work flows, how data is captured, and how governance works in practice.

If you hold that ratio, your odds of landing in the 12% are slim.

For service and BPO leaders: three structural shifts

Across contact centers, BPOs, and mid‑market services, three structural shifts show up again and again.

1. From selling seats to selling outcomes

For 40 years, BPOs and many service firms have sold a bundled product:

1. People (labor)
2. Process intelligence
3. Technology

Workato calls what’s happening now “The Great Unbundling.” Agentic AI and open standards are splitting that bundle:

• Labor becomes the least differentiated layer.
• The real value migrates into process design, orchestration, and proprietary workflow knowledge.

Gartner expects the cost‑to‑value gap on process‑centric contracts to shrink by 50% by 2027. If you are still pricing by the seat while value moves into the process layer, you are:

Charging a premium for the least differentiated part of your own offering.

Implication for operators:

• Start scoping engagements—and internal business cases—around resolved outcomes (issues resolved, days in cash cycle reduced, repeat contacts avoided), not hours.
• Treat your runbooks, SOPs, and “this is how we actually do it here” know‑how as core IP to be encoded into agents and orchestration, not as training collateral.

2. From “more tools” to a neutral orchestration layer

Most large providers have responded to AI by building proprietary platforms and piling on features.

The problem, as Workato and others point out, is that your clients live in a multi‑vendor world:

• Salesforce for CRM, SAP or Oracle for ERP, ServiceNow for ITSM, Workday for HR, industry‑specific systems, and a long tail of point solutions.
• Every “AI platform” that assumes it can own the full stack ends up adding another silo.

What’s actually missing is a vendor‑neutral orchestration layer that:

• Connects to the systems your clients already use.
• Routes work based on clear rules and escalation paths.
• Enforces one permission model and one audit trail across AI and humans.

This is architecture work, not a feature purchase. It’s also where service companies can differentiate: you already understand the messy real‑world workflows; orchestration is how you encode that advantage.

3. From activity metrics to outcome metrics by design

Multiple sources warn that the wrong metrics quietly destroy AI value.

InflectionCX gives the clearest example in contact centers:

• Rewarding “containment” (keeping customers in the bot) led to scripts that blocked escalation and annoyed customers.
• When they shifted to measuring resolution and next‑issue avoidance (no repeat contact within 48 hours), behaviors changed—and so did NPS.

Board advisors like Lutz Finger recommend a new KPI stack for AI:

• % of revenue touched by an AI surface (where is AI actually in the flow of money?).
• Cost per unit of output (salary plus tokens, per claim resolved or case closed).
• % of workflows redesigned for AI vs. retrofitted.

If you are still measuring AI with usage dashboards, you are measuring motion, not value.

Governance: steering and suspension, not a brake

A parallel thread in the latest data is governance.

The instinct in many mid‑market firms is either:

• Ignore it (“We’re too small; the rules aren’t for us yet”), or
• Freeze adoption (“Block AI tools until we have a perfect policy”).

Both approaches fail.

Mid‑market advisors at UHY put it succinctly: “Governance is steering and suspension, not a brake.”

Several data points matter here:

• Shadow AI is everywhere. Palo Alto’s Unit 42 finds an average of 66 GenAI apps per organization, with GenAI‑related data‑loss incidents up 2.5× year‑over‑year.
• IBM’s breach data shows that incidents involving AI add $670K and 10 days to the average breach.
• mlex reports 64% of vendors failed to disclose additional AI sub‑processors in their data‑processing agreements—you may already be exposed through software you think is compliant.

At the same time, governance done well is becoming a competitive advantage:

• A UK mid‑market study summarized in the digests shows governance‑mature organizations achieving AI ROI of ~85.8%, versus 20% for those with no governance—a 65.8‑point gap.
• Multiple sources emphasize that governance is what makes AI scalable, not what slows it down. Without clear permission and audit structures, teams bypass central AI and spin up their own tools.

For service companies, the practical design principle is:

Govern access and outcomes, not every internal decision step.

That means:

• Inventory: know which AI systems (including those embedded in vendors) touch your data and your clients’ data.
• Permission: ensure AI agents can only act within the same permissions as the human they’re assisting—and never exceed them.
• Audit: log who asked what, what data the AI accessed, and what it did with the result, so you can answer auditors and clients in minutes, not weeks.
• Containment: design and test kill switches and purpose limits so you can stop a misbehaving agent quickly.

Done this way, governance becomes the trust layer that lets you move faster than competitors, because you can prove control.

Why this favors incumbents—if they move

One of the quieter but most important themes in the recent research is that incumbent service firms are structurally advantaged in an AI world.

Bessemer’s analysis of AI‑enabled services in health and other verticals notes that “AI‑services‑as‑software” companies deliver service‑level outcomes with software‑level margins (70–80%+) by encoding years of domain knowledge and data into agents.

In parallel, PE and VC investors are backing firms like Pace in insurance, where roughly 25 cents of every premium dollar goes to administrative processing. As one investor put it in Forbes:

“Pace does the exact same thing for knowledge work in insurance that AWS did for cloud computing… spin up a thousand virtual, highly trained intelligent workers for a day and then scale them back down.”

The key point for existing service companies:

• You already own the workflows, the tribal knowledge, and the client relationships.
• The new entrants are raising capital to recreate what you already know—then wrap it in software economics.

Chris Taylor, whose AI‑services firm was acquired into an Anthropic/Blackstone consortium, summarizes it this way:

“For 99% of software companies — the lower your margin, the brighter your future… The newest moat is people‑driven processes.”

Non‑AI companies with strong processes and client trust are in a position to win if they codify their operating model and price for outcomes instead of hours.

So what should a service‑company CEO actually do?

The through‑line across the research is not “adopt more AI.” It is:

Make your AI posture deliberate.

In practice, that means five concrete moves over the next 6–12 months:

1. Name an owner and a metric.

   • Ask: “Who owns AI in this firm, and what single metric proves it’s working?”
   • Capgemini’s data shows that firms without a clear owner and metric overwhelmingly stay stuck in pilots.

2. Pick three workflows, not 30 tools.

   • Use the question from McKelvey’s AI‑readiness work: “Name three workflows—by exact name—where AI can change the outcome this quarter.”
   • Redesign those flows end‑to‑end (intake → triage → work → escalation → close), then pick tools that fit—not the other way around.

3. Flip your budget ratio.

   • Move toward 70–80% of AI spend on operating‑model change, data readiness, and governance, 20–30% on tools and licenses.
   • If a proposal spends more on software than on redesigning how work gets done, treat that as a red flag.

4. Install governance as an operating capability, not a binder.

   • Inventory every AI surface, including what’s embedded in vendors.
   • Implement permission‑inheritance (agents stay within the user’s rights), logging, and tested kill switches.
   • Make AI a standing agenda item at the leadership table, as Protiviti’s board research suggests high‑ROI firms already do.

5. Start unbundling your own offer.

   • Identify where you are effectively selling seats today while the value lives in your process IP.
   • Pilot at least one engagement priced on outcomes (claims resolved, days‑to‑cash, cases cleared) where AI and humans are blended behind the scenes.

None of these steps require a moonshot. All of them require management attention.

The risk of waiting

It can be tempting to sit back, wait for the technology dust to settle, and avoid “betting wrong.”

But as multiple board and governance sources warn, not making a conscious bet is itself a bet:

• Shadow AI continues to spread inside your firm, outside your policies.
• Vendors quietly add AI into products you already use, shifting your risk surface without your input.
• Competitors begin encoding their workflows and client knowledge into agents and orchestration, turning what used to be people‑only services into software‑like lines of business.

The “digital transformation” wave burned many boards; most now understand that waiting didn’t protect them—it stranded them.

AI is shaping up the same way, with one difference: this time, the data is clearer, earlier.

The companies in that 12% success band are not the ones buying the most licenses. They are the ones treating AI as a redesign of how work gets done, with governance and data foundations built in from the start.

If you run a service company, you have a choice:

• Keep buying tools and hope value shows up, or
• Redesign the operating model you already own so AI has something solid to stand on.

Only one of those paths is showing up in the 2026 numbers.