BCG Just Published the Blueprint for the ‘AI-First Field Service Organization.’ It Left Out the Hard Part.

BCG just published “Building the AI-First Field Service Organization”. It’s a blueprint for rebuilding field service companies around AI, connected assets, and change management. It’s a good piece of work. BCG’s earlier field service research claims 10% to 15% productivity gains and 5% to 10% margin expansion from doing this well. I don’t doubt the numbers.

But read the blueprint carefully and you’ll notice something. It redesigns the org chart, the data layer, the agent stack, and the workflows. Yet it never quite touches the thing field service actually is. At bottom, field service is a person in front of a broken machine, trying to figure out what’s wrong.

What the blueprint gets right

The BCG prescription is the now-standard consulting architecture for AI-first operations: connected assets feeding predictive models, AI agents handling scheduling and dispatch, and copilots supporting technicians. To BCG’s credit, there’s also heavy emphasis on change management, because tools nobody adopts produce nothing.

None of this is wrong. In fact, scheduling optimization alone is real money. Salesforce’s new State of Field Service research came out the same week. It found that 57% of organizations using AI-powered scheduling and dispatch report higher revenue per job. If your dispatchers are still juggling spreadsheets, an AI-first redesign will pay for itself.

And the vendors are moving at the same speed as the consultants. That same week, Salesforce pushed Agentforce’s help agent to general availability with outcome-based, pay-per-resolution pricing. It also signed a $1.6 billion Agentic Enterprise License Agreement with the Department of Veterans Affairs. The blueprint isn’t theoretical. The money is flowing.

Now look at what the same research actually says

Here’s the uncomfortable part, and it’s in Salesforce’s own numbers. Ninety-five percent of field service organizations already use AI. Eighty-five percent plan to invest more. Adoption already happened. This is not an industry that needs convincing.

Yet 66% of leaders say mobile worker turnover has increased over the past two years. And 61% say their mobile workers have limited access to the customer data they need once they’re onsite.

Sit with that second number. After a decade of FSM platforms, CRM integrations, and now agentic AI, six in ten technicians still arrive without the context to do the job. The industry has spent billions making the office smarter and the truck is still driving blind.

That’s not a change-management problem. That’s not an org-design problem. It’s an information problem with a very specific shape. The most important data in any field service case is what the equipment physically looks like right now. And almost none of the AI-first architecture captures it.

The blueprint’s missing layer

Consulting frameworks run on things that fit in boxes and arrows: roles, systems, data flows, decision rights. But field service’s defining constraint doesn’t fit in a box. It’s rust on a fitting. A breaker panel wired by someone’s cousin in 1987. An error code that means three different things depending on the board revision. The physical world, in all its unlogged, unstructured mess.

Every prescription in the AI-first blueprint operates downstream of a case description. Predictive models run on sensor data. That’s great for the connected installed base, but useless for the 80% of field equipment that isn’t instrumented. Agents triage tickets — based on what a frustrated customer typed into a form. Dispatch optimization routes the technician efficiently — to a job somebody diagnosed wrong, with the wrong part on the truck.

Garbage in, elegantly orchestrated garbage out. I made a version of this argument about agentic support platforms — an agent can’t resolve a case it can’t see. It applies with double force to a McKinsey-grade org redesign. You can restructure reporting lines forever. If nobody looks at the actual machine until a truck arrives, your AI-first organization is optimizing paperwork around a blindfold.

Seeing the problem is an org-design decision too

The fix is not exotic. Before you dispatch, someone on your team looks at the problem through the customer’s phone camera. This happens during triage, while the case is still a phone call or a ticket. No app install, just a link. Now the diagnosis rests on reality instead of a customer’s guess. The predictive model gains visual ground truth. And the agent triaging the case finally has the one input that actually determines what happens next.

The downstream effects are exactly the ones BCG promises from the rest of the stack. Fewer wasted truck rolls, right part on the first visit, faster resolution. We’ve written about what AI deflection actually does to truck-roll economics, so I won’t rehash the math here. The point for this piece is structural: remote visual context isn’t a tool you bolt on after the transformation. It’s a layer of the blueprint — the perception layer — and it’s the one BCG left out.

It also happens to be the cheapest layer. Instrumenting your installed base with IoT sensors takes years and capex. Rebuilding your data platform takes a program office and a prayer. Putting “look before you roll” into your triage workflow takes a decision.

The org chart can’t fix what it can’t see

I expect a wave of AI-first field service transformations over the next two years. Most of them will follow something close to BCG’s blueprint. Some will work. The ones that don’t will fail the same way AI initiatives already fail in this industry — 95% adoption, rising turnover, technicians onsite without context. They’ll pour intelligence into the layers above the problem while the problem itself stays invisible.

The AI-first field service organization isn’t the one with the best agent architecture. It’s the one that sees the broken machine before the truck leaves the depot. Everything else in the blueprint works better once that’s true — and none of it works well until it is.