Strategy & value alignment before the licence renewal arrives.

1 Business Problems

What specific business problems should AI in ITSM actually solve?

The framing trap is treating AI as a general productivity tool. It isn't. Modern ITSM AI is good at specific repeatable patterns in well-bounded data. It is bad at ambiguity, judgement calls, and anything that requires understanding context the data doesn't contain.

The four problems where AI delivers genuine value:

The fourth problem, inconsistent handling between agents, is where AI delivers its most underrated value. New starters benefit from suggestion-class AI more than experienced agents do, because suggestions encode patterns that experienced agents already know. Onboarding speed and quality consistency are realistic, measurable wins.

The problems AI cannot solve, despite vendor claims:

• Unclear processes. If your incident process is undefined, AI will accelerate undefined work. The output is faster confusion.
• Organisational silos. If teams don't share information today, AI in one team's tool won't bridge to another's. The data still doesn't flow.
• Strategic ambiguity. "Should we adopt AI" isn't a question AI can answer. It's a leadership decision about what kind of operation you want to run.
• Cultural problems. If agents don't write good resolution notes today, they won't write better ones because AI is reading them. The behaviour change has to come first.

2 ROI

How do we calculate ROI for AI in ITSM?

The ROI calculations vendors provide in sales conversations almost universally assume that every AI suggestion is acted on, every deflection avoids a full ticket, and every summary saves the full read time. None of those is true in practice.

A realistic ITSM AI ROI calculation looks like this:

Apply realistic baselines and the ROI math still works for most teams, but the payback period typically runs 9-18 months rather than the 3-6 months vendors quote. That difference matters at budget sign-off, especially when finance teams check the numbers against actual outcomes 12 months later.

Three additional cost categories that vendor ROI calculators reliably ignore:

• Data preparation. The 20-60 hours of triage, categorisation, and resolution-note enforcement that determines whether AI works at all. Hidden in vendor proposals as "your existing data is fine."
• Change management. Agents need training. Managers need new dashboards. Quality assurance processes need updating. Conservative estimate: 40 hours of senior time per 25 agents.
• Ongoing measurement. ROI cannot be claimed without measurement infrastructure. Building that infrastructure, baseline data, control groups, attribution logic, is itself a project. Most teams skip it and rely on vendor dashboards, which means they never know the actual return.

3 Quick Wins

What are the right "quick win" AI use cases to pilot first?

Quick wins matter because they create the political capital needed for the harder, higher-value AI work later. A failed first pilot kills the programme. A successful first pilot funds the second.

The three pilots that consistently work as first deployments:

What to avoid in the first pilot, regardless of vendor pitch:

• Autonomous agents that take actions without human review. The governance overhead exceeds the operational benefit until trust is established.
• AI-driven routing at the start. Routing failures produce visible, auditable consequences (SLA breaches, customer complaints). Wait until acceptance is measured on lower-risk features.
• Auto-resolution. The combination of AI confidence and irreversible action is the single highest-risk pattern in ITSM AI. Save it for after maturity.
• Multi-team rollout. Pilot with one team, one shift, one ticket type. Expand on evidence, not enthusiasm.

4 Maturity Alignment

How do we align our AI strategy with our ITSM maturity?

ITSM maturity isn't an abstract concept here, it's the specific set of operational disciplines that determine whether AI features have anything useful to work with. The three that matter most for AI readiness:

• Categorisation discipline. Tickets land in consistent categories. Generic catch-all categories ("Other," "General") are below 15% of volume. Without this, AI clustering produces incoherent groups.
• Knowledge management as a habit, not a project. KB articles are written when issues are resolved, not in a quarterly cleanup. Without this, AI knowledge surfacing has nothing to surface.
• Resolution practice. Tickets close with substantive notes describing what was actually done. Without this, AI suggested-replies cannot learn from past resolutions.

An honest maturity assessment maps to AI readiness as follows:

Most organisations sit in the middle two rows. The rare ones in the top row should be honest about it, buying AI before fixing the basics produces 18 months of vendor calls explaining why the features aren't delivering. The fix is upstream, not in a different vendor.

5 Agent Wellbeing

Can AI in ITSM actually reduce agent burnout, or does it create new problems?

The vendor narrative, "AI removes drudgery so your agents can focus on meaningful work", is half true. The full picture includes failure modes that don't appear in the sales deck.

AI reduces burnout when:

• It removes work agents already wanted off their plate. Repeated password resets, status update emails, copy-paste responses to common queries. Agents welcome the relief.
• Suggestion quality is high enough to trust. Above 30% acceptance rate means agents stop second-guessing every suggestion. Below that, they're triple-checking, which adds work.
• New starters get earlier autonomy. AI suggestions encode tribal knowledge. Junior agents reach competent independent handling weeks earlier.
• Leadership frames AI as capacity expansion. "We can take on the customer satisfaction project we never had time for" produces engagement. "We can reduce headcount" produces resistance and quiet sabotage.

AI creates new burnout when:

• Suggestions are wrong often enough to require constant filtering. Below 15% acceptance, the cognitive cost of evaluating every suggestion exceeds the benefit. Agents experience this as additional work, not less.
• Review responsibility increases without recognition. "You also need to review the AI's outputs" lands as additional unpaid work if framed as a duty rather than a productivity feature.
• Metrics shift to AI-friendly ones without context. "Tickets per agent" goes up because AI is doing the easy ones; the human-handled tickets are now disproportionately complex, but compensation and recognition don't reflect this.
• Headcount reduction is the explicit goal. Agents will undermine the AI to protect colleagues, often without realising they're doing it. Adoption collapses.

The most damaging pattern: quietly retraining the team's job description without saying so. "AI handles the easy tickets, humans handle the complex ones" sounds reasonable until you realise the easy tickets were the breaks in the day. Agents handling complex tickets back-to-back-to-back burn out faster than agents handling a mix.