The highest-ROI AI projects in 2026 are not automating people out — they are automating the admin overhead that stops skilled professionals from doing their actual jobs.
The real AI conversation we should be having
Think about the last time you watched a skilled professional — a doctor, a teacher, an engineer — go through their day. How much of it was the work they trained for? And how much was typing into systems, chasing data, filling forms, reconciling records?
We have been having the wrong conversation about AI in the enterprise. The headlines fixate on job replacement. But the organisations getting real, measurable value from AI in 2026 are not replacing their workforce. They are removing the administrative drag that makes their workforce slow, frustrated, and expensive to retain.
Two examples keep showing up in our project pipeline, and the pattern they share is worth naming.
Where the hours actually go
In healthcare, clinicians at a typical 800-bed hospital trust spend two hours on EHR documentation for every hour of direct patient contact. AI-powered clinical note summarisation — ambient capture, structured extraction, LLM-generated draft notes reviewed and approved by the clinician — targets a 40–50% reduction in that documentation time. The doctor still reviews everything. The AI handles the transcription and structuring that nobody went to medical school to do.
In education, vocational training providers are stuck teaching last year’s skills. The job market now moves faster than annual curriculum reviews can track. AI-driven skills gap analysis ingests live job posting data, maps it against existing curricula using standardised taxonomies, and surfaces what is missing — before graduates hit the market with outdated competencies. Providers using this approach typically target a 25–35% improvement in graduate employment alignment.
Neither example is about fewer people. Both are about fewer wasted hours. And both share an architecture pattern that works across sectors.
Why admin-first AI deployments win
There is a pattern here that holds across every industry we work in. The highest-ROI AI deployments share three traits:
- They target measurable time sinks, not vague efficiency gains. Documentation hours. Curriculum review cycle days. Manual reconciliation steps. If you cannot put a number on the current waste, AI will not magically find one for you.
- They keep the human in the loop. The AI drafts, extracts, and recommends. The professional reviews, approves, and decides. This is not a limitation — it is what makes the system trustworthy enough to deploy in regulated environments where mistakes carry real consequences.
- They integrate into existing workflows. A brilliant model that lives outside the EHR or the curriculum management system is a demo, not a product. The engineering work — FHIR integration, taxonomy normalisation, API design, data residency controls — is where practical value gets built. Models are commoditising. Integration is not.
The mistake we see teams make
The temptation is to start with the most exciting AI use case — the one that looks best in a board presentation or an innovation showcase. Resist that impulse. Start with the most annoying workflow. Find the task that makes your best people groan when they sit down on Monday morning.
That is where AI earns its keep fastest, with the least organisational resistance.
Clinicians do not push back against an AI that does their typing. Teachers do not resist a system that tells them which modules need updating. Resistance shows up when AI threatens professional judgement, not when it handles administrative overhead. If your first AI project faces heavy change-management friction, you probably picked the wrong problem to solve first.
What this means for the next 12 months
The organisations that will pull ahead are not the ones with the most advanced models or the largest AI teams. They are the ones with the most disciplined approach to identifying and eliminating professional overhead — one workflow at a time, with clear metrics and human oversight built in from day one.
If your teams are spending more time on documentation than on the work they were hired to do, that is your signal. The tools exist. The architectures are proven. The question is whether the organisation is willing to start with the boring problem that costs the most.
We have been writing about both of these use cases in detail this week — links below for the full technical breakdown.