Augmented Reality in Field Service: 7 Use Cases That Work on a Phone

Augmented reality in field service has a reputation problem. For several years it meant a headset, a six-figure pilot, and a demo that looked great on stage and never reached the technicians who actually fix things.

That version is over. Microsoft stopped making HoloLens 2 in 2024 and is retiring Dynamics 365 Remote Assist and Guides at the end of 2026. Counterpoint Research counted a 28% drop in global VR headset shipments in 2025. What survived is the useful part: showing a technician exactly which component a step refers to, on the device they already carry.

This post covers the seven AR use cases I see working in field service today, what each one needs, and where the headset-era versions went wrong. Two of the seven are AR in the strict sense, a mark placed on the equipment or on a model of it. The other five are the camera, AI and workflow pieces that make those two worth deploying; without them, AR stays a demo.

What augmented reality in field service actually means

Strip away the hardware and AR in field service is one idea: showing the technician the part, not just naming it. Instead of reading "check the connector on the control board" and then looking for it, the technician sees it visually marked in the video stream.

That happens in two different ways, and it is worth keeping them apart:

• Live, in the camera view. A remote expert, or anyone else in the session, draws on the technician's live camera feed, and the annotations stay locked to the equipment as the camera moves. Annotations in real space are placed by people in the call. Nothing is overlaid automatically.
• In the procedure, on a 3D model. A guided procedure shows a model of the equipment with the relevant part highlighted at each step. The model sits in the guidance on screen; it is not projected onto the real unit.

Everything below is one of those two, applied to a different moment in the job.

1. Remote expert guidance on the live camera view

The oldest use case, and still the one with the fastest payback. A technician, or the person onsite, opens a link, points the phone at the equipment, and a remote expert sees what they see. The expert, and anyone else in the session, draws arrows and circles that stay locked to the object as the camera moves, so "that one" is never ambiguous. This is the only use case where AR marks sit on the real equipment, and they come from the people in the call.

What it needs: an app on a phone. No headset, no content authored in advance.

Where it pays: second visits. A large share of repeat dispatches are a technician going back with the right part or the right person. A ten-minute AR remote assistance session confirms the fault and the part before anyone drives.

2. Equipment identification from the camera

Before any guidance can be right, the system has to know which unit it's looking at. Computer vision identifies the equipment model from a photo and loads the knowledge published for that model: manuals, service bulletins, repair videos, work notes.

What it needs: a photo, and documentation attached to the equipment model.

Where it pays: mixed installed bases. A technician who services 300 models cannot carry the right manual for each one in their head. Identification removes the "which procedure applies" step that used to eat the first fifteen minutes of a visit.

3. Guided repairs with the part highlighted on a 3D model

This is what most people picture when they hear "AR work instructions," and it is the use case that stalled hardest in the headset era, because every 3D model took a specialist days or weeks to build from CAD. That constraint is gone. In ResolveGrid's 3D Studio, five or six photos of a unit produce a high-quality model in 30 to 60 minutes.

Once the model exists, a guided procedure shows it on screen at the start so the technician can orient, then re-presents it with the relevant part highlighted as each step calls for it. The technician looks at the model, then at the unit; nothing is projected onto the equipment itself. "Remove the fuser cover" is a sentence. The same sentence with the cover lit up on a model of this exact unit is an instruction. I wrote more about this in Augmented Reality Work Instructions: A Practical Guide.

What it needs: a 3D model (an hour) and the procedure built as digital work instructions in a no-code builder, from documents you already have.

Where it pays: complex units, newer technicians, and anything where "which bolt" matters.

4. Troubleshooting from a photo of the fault

Not every AR use case is an overlay. Some of the value is the camera as input. A technician photographs the error display and the agent reads the code and answers from the approved manual, citing the passage it drew from. Photograph a connector and it checks whether the cable is seated. Ask in plain language, "what does E-27 mean on this unit," and get an answer grounded in your documentation rather than a guess.

What it needs: published documentation for the model, and an agent that declines to answer when it can't find a grounded source.

Where it pays: the long tail of faults no procedure covers, and the hours after the senior technician has gone home.

5. Verifying critical steps with a photo before moving on

Checkboxes get ticked. Photos get taken or they don't. Critical steps in a guided procedure can require a photo before the next step unlocks: torque marks, a seated connector, a cleared drain. The photo becomes part of the record.

What it needs: a workflow builder that supports photo-gated steps.

Where it pays: safety-critical work, warranty disputes, and any procedure where a skipped step shows up as a callback three weeks later.

6. Capturing proof of work into the work order

Readings, photos, completion status and the session itself flow into the work order in ServiceNow, Salesforce, ServiceMax or your own system. The service record stops being whatever the technician remembered to type in the truck afterwards.

What it needs: an integration into the system of record, through the Web API.

Where it pays: the back office. Dispatchers see what was found, parts planning sees what was used, and the next technician sees what the last one saw.

7. Training on the same model and procedures used in the field

The 3D model and guided procedures that help a technician on site are the same assets that prepare a new hire before they arrive. Walk the procedure on the model in the office, then run it for real with the agent alongside. Salesforce's 2026 field service research found that 66% of leaders report rising mobile worker turnover, with insufficient training on new technology the top reason cited. Training that reuses field content is the cheapest way to close that gap.

What it needs: nothing beyond use case 3. The content is already built.

Where it pays: onboarding, and the gap between "finished the course" and "can fix the unit alone." More on this in AI-powered training for field service teams.

Where the headset era went wrong

Four reasons, and they explain why the use cases above look the way they do.

The hardware. Headsets struggle with gloves, bright sun, tight spaces and roofs. Most technicians carry a phone, so that is where field guidance has to work. Every use case above runs on a phone or tablet.

The content cost. 3D content built per model from CAD, by a specialist, meant coverage never got past a pilot. An hour per model changes the math.

The overlay didn't know anything. The headset-era promise was pre-authored holograms pinned to the real machine, and they showed the standard procedure regardless of which error the unit was throwing or what the last visit changed. The fix is an agent that picks and adapts the procedure and shows the part on a model, with live annotation on the real equipment left to the people in the session.

Nothing came back. The instruction was a one-way broadcast. Use cases 5 and 6 exist because the record matters as much as the guidance.

How to start

1. Pick one asset model with high repeat-visit or escalation rates.
2. Turn on remote expert guidance first. It needs no content and pays back on the first avoided dispatch.
3. Build the model. Five or six photos, 3D Studio, an hour at most.
4. Upload what you already have for that model and structure the one or two procedures that cause the most callbacks.
5. Run it with three or four technicians on live jobs for two weeks and measure first-time fix, repeat dispatches and escalations against the prior period.

If the numbers move, add the next model.

Where ResolveGrid fits

ResolveGrid combines an AI agent, computer vision, fast 3D modeling, guided workflows and AR remote expert support in one technician workflow, on any phone or tablet. The agent identifies the equipment, delivers the right step from your approved knowledge, shows the technician where on the unit it applies, verifies the work with a photo and brings in a human expert with full context when needed. Early ResolveGrid customers cut dispatches by up to 50%.

Want to see these use cases on your own equipment? Request a demo.

FAQ: augmented reality in field service

Do you need a headset for augmented reality in field service?

No. Every use case in this post runs on the phone or tablet the technician already carries. Headsets add hands-free viewing but bring cost, battery and environmental limits that make them a poor fit for most service work.

What is the difference between AR remote assistance and AR work instructions?

AR remote assistance is people in a live session drawing on the technician's camera view, with the marks anchored to the real equipment. AR work instructions are pre-built procedures or real-time guidance that show each step on a 3D model of the equipment, on screen, with the relevant part highlighted. Most service teams end up using both: instructions for the known procedures, a live expert for everything else.

How long does it take to deploy AR in field service?

Remote expert guidance can be live the same day, since it needs no content. Guided procedures take about an hour per equipment model for the 3D model plus the time to structure the procedure from existing documentation.

Does augmented reality in field service improve first-time fix rate?

That is the metric it is meant to move, by getting the right part and the right step to the technician before and during the visit. Measure it per asset model against the prior period rather than fleet-wide, so the effect is visible early.