Frontline Workers Are the Last People to Get AI. They’re the First People Who Actually Need It.
Last week, Evri — the UK’s largest parcel carrier — announced it’s rolling out Microsoft 365 E7 to 6,000 employees. Copilot. Agent 365. The full suite. They called it “the next phase of their AI journey.” The announcement was well-produced. The language was ambitious. It talked about cutting manual tasks, getting clearer insights, helping teams “work smarter not harder.”
I read the whole thing looking for mention of their drivers.
Couldn’t find it.
That’s not a knock on Evri specifically. It’s the pattern. In fact, this is how every major AI deployment in logistics, field service, utilities, and telecoms works right now. The back office gets the tools. The person in the van — the one actually touching the parcels, fixing the cables, replacing the meter — gets nothing. Or maybe a QR code that links to a PDF.
Around the same time, Salesforce announced that its Agentforce IT Service product was designed to “end the portal-to-ticket era” — replacing static support forms with conversational AI that routes and resolves issues autonomously. Smart move. Clean UX. Genuinely useful for the employees sitting at desks who used to fill out those forms.
Still mostly a desk worker problem.
Here’s what I keep coming back to: the hardest support problems — the ones that cause the most cost, the most repeat visits, the most customer frustration — aren’t happening at desks. They’re happening in the field. On the roof. Under the sink. In the server room. At the loading dock. They’re happening in places where the person trying to solve the problem can’t easily type a support ticket, can’t pull up a knowledge base article, and can’t get a specialist on a screenshare.
The people who need AI the most are the people who can’t use it yet.
The AI Rollout Playbook Is Wrong
There’s a logic to why enterprises start with knowledge workers. They’re easy to instrument. After all, they already live in Microsoft 365 or Google Workspace. They have laptops and enough downtime between tasks to interact with an AI assistant. The ROI case is simple to model — count the hours saved on email, summarization, report generation.
But “easy to instrument” is not the same as “most in need.”
A knowledge worker who uses Copilot to summarize a meeting saves twenty minutes. That’s nice. Meanwhile, a field technician who shows up to a job without visual confirmation of what’s actually wrong and drives two hours for the wrong part — that’s a $400 mistake. Times however many times it happens per day across your fleet.
The cost of AI deprivation compounds differently in the field. For desk workers, not having AI is a productivity gap. For frontline workers, not having AI is a failure mode. As a result, every unnecessary truck roll, every wrong diagnosis, every repeat dispatch becomes a quantifiable loss hitting your P&L right now.
And yet the enterprise AI investment is flowing upward, not outward.
The Unique Problem Frontline Workers Have
When someone in your back office hits a problem they can’t solve, they can describe it in text. They can screenshare. They can paste an error log. In short, they can do the things that modern support systems are built to receive.
When your field tech hits a problem they can’t solve, the problem often exists in three dimensions. A weird noise. A component that looks right but isn’t quite seated properly. Maybe a visible crack the original work order didn’t mention. Or a device configuration that doesn’t match what’s in the system because someone swapped it out last quarter without updating the record.
In other words, none of that translates easily into a text box.
This is why “just give them the same tools” doesn’t work for frontline deployment. Copilot, for example, is brilliant for summarizing documents. It cannot tell a technician whether the wiring they’re looking at matches the spec. Likewise, Agent 365 is impressive for routing IT tickets. It cannot visually confirm whether a field installation actually passed inspection.
Frontline workers don’t just need conversational AI. They need AI that can see what they’re seeing.
That’s a different capability, and it requires a different deployment philosophy.
Visual Context Changes Everything
I’ve spent years watching support calls in the field, and there’s a consistent pattern. The best outcomes happen when a remote expert can actually see the problem. Not hear a description of it. Not read a ticket about it. See it.
When you can see what the person in the field is looking at, the diagnosis time compresses dramatically. You stop playing the telephone game where the technician’s words have to somehow convey the physical reality they’re standing in front of. Instead, you see the actual state of the equipment. You give an actual instruction. The problem either gets solved or gets correctly escalated.
The question was always: how do you scale that? You can’t have a senior engineer available on a call for every technician in the field.
AI changes the answer. Not by replacing the remote expert entirely — not yet — but by making that connection faster, smarter, and more accessible. By doing the work of getting visual context into the loop before anyone has to make a decision. By helping the technician quickly capture what they’re seeing and get it to the right person, with the right context already attached.
What 830 IT Leaders Are Getting Wrong About Agentic AI in Support covers how production AI deployments fail at the exact moment they hit something in the physical world. That failure mode isn’t going away. If anything, it gets more expensive as the AI layer scales up without the visual ground truth to back it.
The ROI Case Is Hiding in the Field
Here’s the argument that I think hasn’t been made loudly enough yet: the ROI on AI for frontline workers is larger than the ROI on AI for back-office workers. It’s just harder to see because you have to measure it differently.
Back-office AI ROI: hours saved × cost per hour.
Frontline AI ROI: repeat dispatches eliminated + wrong parts avoided + escalations deflected + first-contact resolution rate improvement + technician hours recovered from dead-end jobs.
That’s a bigger number. Indeed, it’s usually a much bigger number. The reason enterprises aren’t chasing it is because it requires solving a harder problem — getting AI into an environment that isn’t a keyboard and a screen.
But “harder” isn’t “impossible.” It’s just requiring someone to actually think about the deployment differently.
What Has to Change
The Evri announcement isn’t wrong. Back-office AI creates real value. Salesforce’s agentic service products are genuinely smart. None of this is bad.
What’s bad is that 6,000 desk workers getting Copilot is treated as a landmark AI rollout for a company whose entire business is delivered by people in the field. What’s bad is that “ending the portal-to-ticket era” is the innovation story while the field tech without a way to get visual support in real time isn’t anyone’s headline.
Enterprise AI strategy needs to answer a harder question than “where can we deploy the tools we already have?” The harder question is: where in our operation is the most value being destroyed by information gaps, and how do we close them?
For most companies in logistics, utilities, field services, and hardware support — the answer is in the field. It’s always been in the field.
The tools to do something about it exist now. However, the deployment philosophy hasn’t caught up yet.
That’s the actual gap.
Viewabo helps organizations bring visual context into the field — so the people physically touching the problem have the same information advantage as everyone else.