ServiceNow Virtual Agent for Manufacturing

Virtual Agent for manufacturing is the chat assistant that lets plant operators and technicians report equipment downtime, check work order status and request materials without filing a ticket by hand. It is built for shop-floor use on tablets and phones during shifts.

Manufacturing’s Unique Virtual Agent Requirements

Manufacturing organisations use ServiceNow to manage production schedules, maintenance requests, quality audits and asset lifecycle. Virtual Agent in this context handles routine operational requests: work order status checks, equipment downtime notifications, shift handoff documentation, and preventive maintenance scheduling. Unlike office-based service requests, manufacturing requests often come from non-desk workers using tablets or mobile devices on the shop floor.

Manufacturing plants run 24/7. Virtual Agent must support shift work, recognise requests from operators across multiple time zones and languages, and handle urgent escalations without delay. A worker who discovers a machine malfunction needs quick acknowledgment and a plan, not a ticket number.

Virtual Agent Conversations in Manufacturing Context

Common manufacturing intents for Virtual Agent include: “Report equipment downtime” (create a maintenance request and assign to the next available technician), “Check work order status” (retrieve the WO and show current step and ETA), “Request a tool or material” (check inventory and create a purchase order if out of stock), and “Log shift production numbers” (update production records for the shift).

Each conversation should be brief and specific to the operator’s context. If a CNC machine is down, the operator needs to know if it is in the maintenance queue and who is assigned. They do not need a 10-question diagnostic; they need confirmation someone is coming, plus an escalation path if the downtime extends beyond a threshold.

Configure Virtual Agent to recognise equipment identifiers (machine serial numbers, asset tags) and operator shift information. When an operator reports downtime, pre-populate the maintenance request with the equipment ID and shift, reducing the questions Virtual Agent needs to ask.

Integration with Maintenance and Production Systems

Virtual Agent connects to your CMDB to look up equipment status. When an operator asks about a lathe, the system retrieves its maintenance history, current work orders and known issues. This context helps Virtual Agent decide whether the request is routine or urgent.

Connect Virtual Agent to your MES (Manufacturing Execution System) or ERP production data so conversations about work order status pull real-time data. If an operator asks “Is my order ready for shipment?”, Virtual Agent queries production status, inventory, and QC flags before answering.

Use Virtual Agent to escalate urgent equipment failures to the maintenance team without delay. If downtime is likely to exceed a critical threshold, trigger an automated call to the maintenance supervisor and create a priority work order.

Mobile and Accessibility

Manufacturing workers use tablets and phones on the floor. Ensure Virtual Agent chat interface is optimised for small screens. Keep conversations to one or two questions per turn. Support voice input or dictation if workers are wearing gloves. Test the interface on Android tablets, which are common on plant floors.

Virtual Agent must work offline or with poor connectivity. If your plant has WiFi dead zones, configure the app to cache conversation history so users can see previous responses even if the connection is intermittent.

Escalation and Follow-Up

Manufacturing escalations are time-critical. If a work order is assigned but not started within a defined window, Virtual Agent automatically escalates to the supervisor. If a quality check fails, Virtual Agent escalates to the quality manager immediately.

After resolving a maintenance request, Virtual Agent should follow up with the requesting operator to confirm the issue is resolved and ask for feedback. This feedback helps you track repeat issues and plan equipment replacement.

Compliance and Safety

Manufacturing plants operate under safety regulations. Virtual Agent should not provide instructions for equipment operation; that remains the technician’s and supervisor’s responsibility. But Virtual Agent can confirm that mandatory lock-out/tag-out procedures have been completed before a technician is assigned.

Track all conversations for audit purposes. Manufacturing audits often require evidence that procedures were followed. Archive Virtual Agent conversation logs with the associated maintenance records.

Getting Started

Start with one high-volume intent: equipment downtime reporting. Build a conversation that an operator can complete in 30 seconds. Train operators on how to use it during shift handoff. After two weeks of real usage, collect feedback and refine. Add a second intent only when the first is running smoothly.

Oblytech works with manufacturing plants to identify which conversations will have the highest impact, design them for mobile and shift work, and integrate with your production and maintenance systems.

Deciding which plant questions the agent may answer and which it must hand over

On a plant, the most important design decision for a conversational agent is the boundary between answering and handing over. A question about where a work order stands can be answered from the record. A question about whether a press is safe to restart cannot, because the honest answer depends on a physical inspection that only a qualified person can perform. The agent should therefore be given three kinds of response: an answer drawn from a governed record or approved knowledge article, an action that opens or updates a record, and a handover to a named person with the context already captured. Mixing these is where plant deployments often go wrong, when the knowledge base starts to contain procedural steps that were written for a training course and are then read as instructions at the machine.

Handling shift change without losing an open conversation

Conversations on a plant frequently span a shift change. An operator reports a fault at the end of the night shift, a day technician takes the job, and the record shows nothing about what the operator saw. Configure the agent so that every open conversation carries a shift identifier and a short structured summary, and so that the summary is visible to the incoming supervisor at the start of their shift. This keeps the handoff inside the platform, rather than relying on a written log at the machine that may not be read.

1. Require the operator to confirm the asset and area before any record is opened, so the summary is attached to the right equipment.
2. Store the operator’s own words alongside the agent’s classification, so later reviewers can see what was reported.
3. Show open conversations by shift to the incoming supervisor, sorted by whether a person has been assigned.
4. Close a conversation only when the assigned person records an outcome, not when the operator stops typing.

When the agent should not be offered at all

Some plant situations should bypass the agent entirely. An emergency stop, a gas or chemical release and a person trapped in equipment all require an immediate call to the designated response route, and a chat window is the wrong tool. Similarly, a machine that has been locked out for maintenance should not be offered a status conversation that might suggest it is ready. Place the response route at the top of the screen on shared terminals, and keep the agent out of any flow that involves an energy isolation.

Testing the agent with the people who will use it

Test the conversations on the floor, with operators and technicians who did not write them. Ask them to report a fault in their own words, and watch where they stop, misspell an asset tag or give up. Failed conversations are among the most useful data, because they show which asset names, abbreviations and local terms the agent does not recognize. Keep a running list of these terms and map each one to the asset or intent it refers to, rather than adding synonyms in bulk.

Measures that show whether the agent is useful

Resolution rate alone can mislead on a plant, because a conversation that ends with a handover is not a failure. Track the share of conversations that reach a recorded outcome, the time between a report and the first human response, the number of handovers that were reopened because the summary was incomplete, and how often operators abandon the agent to phone the supervisor. Abandonment is the signal that the agent has lost the operator’s trust, and it is often a result of too many questions before anything happens.

Questions raised by plant managers during rollout

Can the agent tell an operator whether a machine has been serviced recently?

Yes, if the maintenance history is held in the platform and the record is current. The agent should give the date and the outcome of the last recorded service, and it should not infer that a machine is safe from the absence of an open task.

Who maintains the answers the agent gives?

A named owner in each plant area should approve every knowledge article the agent uses, and articles should carry the date of their last review. Articles written for training should be kept out of the agent’s source set, because they often describe conditions the operator will not see on the floor.

Should contractors use the agent?

They can, under their own identity and scoped to the areas they service. Contractors should be routed to the same handover paths as employees, and their conversations should be retained with the same audit trail.

For the knowledge side of this, see the manufacturing knowledge management guide, and the virtual agent and self-service service page describes how a conversation is kept to a defined task or handed to a person.