When AI Gets Its First Real Proving Ground

Last week’s note laid out the Three Pillars – the structure that separates AI-as-theatre from AI-as-infrastructure. Many wrote back with the same question: “What does this look like when it’s actually put to work?”

This week is about that: three 90-day arcs, each from a different sector, each with a tightly scoped transformation. Small surface area. High-leverage workflows. A clear line from Pillar 1 to Pillar 3.

Logistics: When ETA Stops Being an Apology

In last-mile operations, the question “Where is my order?” is not a customer query; it is an operational indictment. Most teams have maps, telemetry, planning systems – yet end up negotiating with uncertainty.

In Table 1 we show what happens when the data already flowing through the fleet is grounded, modelled, and given an autonomous action path.

Once the structure is in place, drivers aren’t left improvising their own updates, dispatch avoids the whack-a-mole of manual corrections, customers get clarity instead of vague promises and SLA credits stop being a monthly ritual.

One ETA. One notification loop. One visible behavioural shift inside the operation. A 90-day win that earns the right for the next one.

Utilities: When a Transformer Warns You Before Physics Does

Utilities often know which assets are old, stressed, or overloaded – but rarely which ones are quietly drifting towards failure. Thresholds tell you what has already crossed the line. Patterns tell you what will.

What changes when real-time streams and historical signatures are treated as a single system rather than scattered telemetry points is illustrated in Table 2.

When Pillar 1 → Pillar 2 → Pillar 3 align, a transformer’s subtle deviations become meaningfully detectable, a crew is dispatched before customers call about flickering lights, maintenance shifts from reactive firefighting to controlled intervention and reliability stops depending on luck.

It’s a narrow slice: one asset class, one alerting path. But in 90 days, it makes the case for expanding to the rest of the grid.

Healthcare: When Prior Authorization Stops Owning Your Week

In hospitals, the bottlenecks that slow the revenue cycle rarely arrive with alarms; they arrive as paperwork. Over and over. PDFs, faxes, scanned notes – all demanding manual review, data entry, cross-checking of policies, and form submission.

The workflow in Table 3 shows what happens when this process is reframed through the Three Pillars.

The shift is immediate: Documents become structured signals instead of objects to open, policy rules become computable, not interpretive, routine form-filling moves to an autonomous Agent, leaving humans only the complex cases and authorizations stop slipping by days simply because someone couldn’t get through a stack.

Hospitals don’t need a monolithic RCM overhaul. They need one high-volume workflow that stops consuming their mornings. A 90-day relief valve.

A Repeating Pattern Across All Three

Despite the sectors having almost nothing in common – delivery vans, transformers, clinical documents – the same architecture emerges:

1. One narrow surface area that avoids organisational overwhelm
   

2. Data that already exists, organized instead of expanded
   

3. Intelligence that refreshes itself, not a science project in a corner
   

4. An Agent that performs one surgical action, not a grand reinvention
   

5. A 90-day window that produces trust instead of asking for it
   

This is the quiet truth about Enterprise AI: scale begins with constraint, not ambition.

What Happens in Week 13

Once a small win lands, the operation immediately sees the next fracture point. In logistics: “Can we correct upstream ETA drift?” In utilities: “Can we bring more asset classes into the same pipeline?” In healthcare: “Can we predict denials before triage?”

That is the point. The first 90 days aren’t the transformation. They are the permission slips.

Planning a Q1 2026 Win?

Pick a process that matters. Pick one integration path. Respect the Three Pillars. Let the 90 days do the work.

Enjoy your Sunday!

About the day a dataset disagreed with him – and lost – is the one colleagues still like to rib the author Shrikanth Jagannathan, Chief Data Scientist and co-founder, Moative about.

A few that aged well, in our vault