Voice-First Agents for Field Teams

Gloves, ladders and a phone in a pocket

A heating engineer finishes a boiler service in a loft. The job sheet needs filling in, a part needs ordering and the next customer wants an arrival time. Their hands are dirty, their phone is in a pocket, and the form on the app has 22 fields. So it all gets written up in the van at half past five, from memory.

Voice-first agents are aimed squarely at that moment. Instead of an app that happens to have a microphone button, the main way to use the system is talking to it, and the agent turns speech into structured records, lookups and actions. Voice AI has improved enough by 2026 that this is practical for field teams, not just call centres.

What field teams actually use voice agents for

• Dictating job notes that the agent turns into a completed job sheet with fields filled
• Asking for asset history: "when was this unit last serviced and what was replaced"
• Checking van or depot stock and reserving a part for collection
• Working through safety or inspection checklists by answering questions aloud
• Logging time, mileage and materials used at the end of each job
• Telling the office about delays so customers are updated automatically

Of those, turning dictated notes into structured records is usually the most valuable. It improves data quality as much as it saves time, because notes captured at the job are more accurate than notes reconstructed hours later.

Transcripts are not the goal. Structured data is.

Many voice projects stop at transcription and produce a wall of text nobody reads. The useful step is extraction: the agent listens, then fills specific fields such as fault code, parts used, readings, follow-up required and customer signature status, and reads back anything it was unsure about.

A voice agent that saves a transcript has made a recording. One that fills the job sheet and asks about the missing reading has done the admin.

This is where agent design matters. The agent needs the job's form schema, reference lists such as part numbers and fault codes, and rules about required fields, so it can ask "what was the flue gas reading" before the engineer leaves the site.

Designing for real field conditions

Offline support is often the deciding factor. On-device speech recognition lets staff dictate with no signal, with the agent processing the rest when a connection returns, a pattern we describe in on-device AI for business apps.

Confirmation and safety

Voice introduces a specific failure: a confident mishearing. "Fifteen" and "fifty" sound alike on a building site. Any value that matters must be confirmed back in a way that is quick to correct.

1. Read back numbers, part codes and quantities before saving or ordering
2. Require an explicit yes for anything that orders, charges or notifies a customer
3. Show a written summary on screen for the worker to glance at when safe
4. Never make someone interact with the agent while driving beyond simple audio updates the law and your policy allow
5. Log the audio for disputed records for a short, defined retention period, and tell staff you do

The driving point is not a formality. Check the rules where your teams work and your own policy before designing any in-vehicle interaction.

Voice agent vs a better mobile form

Voice is not automatically better. For quick taps, such as marking a job arrived or complete, a large button is faster. For long free-text notes, checklists done with busy hands, and questions that would otherwise need a phone call to the office, voice wins. A good field app offers both and lets people choose.

It is also worth asking whether staff will talk to an app in front of customers. Many are comfortable doing so in the van and less so in a customer's kitchen. Pilot with a few volunteers before rolling out to a 40-person team, and listen to what they say about it.

How we would approach it

SpiderHunts would start with one job type and one outcome: a complete, accurate job sheet captured before the engineer leaves site. We build the extraction and confirmation first, integrate with the job management system, and measure how many sheets are complete on first submission compared with the old form.

Customer-facing voice is a different project with different trade-offs, covered in AI voice agents for sales and support. Internal field agents like this are built through our AI agent development service, often alongside the offline-capable apps described in offline-first apps for field teams.

Frequently asked questions