How AI Turns Spoken Field Observations into Complete Maintenance Notifications

Key takeaways

• The person standing at the equipment has the best information, but busy schedules, unclear fields, and delayed entry cause details to be lost before they reach the CMMS/ERP. 
• Across the customers we’ve profiled in case studies, adding a mobile app lifts mobile execution by 15-20% on average.
• An AI agent trained on 25+ years of maintenance planning and scheduling processes guides technicians through user-configured questions by voice.
• One spoken description can produce a notification with type, location, short description, long text, cause code, and priority.
• The AI runs on top of the mobile app and the CMMS/ERP connection already in place, so no new integration is required. 

The best information about an equipment issue is at the equipment. A technician walking the plant can see, smell, hear, and feel what is wrong, whether it is a vibration, an abnormal odor, or a leak. Too often, that detail does not survive the trip back to the CMMS/ERP. An AI agent inside a mobile app changes the capture step: the technician describes the issue by talking, and the agent turns it into a complete notification.

Why does field data get lost before it reaches the CMMS/ERP?

Everyone in the field is busy. A single daily walk path can surface 100 different issues, which forces constant decisions about what to prioritize, how much detail to write down, and how much to remember. 

Then comes the question of how to enter it. Technicians often wonder which field to use. Is a priority required? Is a long description needed? Does the damage code matter? The requirements are often ambiguous. 

Timing adds another problem. An issue gets jotted on a sticky note and entered the next day or that evening. That lag affects accuracy, and information can be entered incorrectly in the ERP or CMMS. A detail can also be skipped because the issue is familiar and typing it out does not seem worth the time. 

The person who finds the issue is often not the person who plans the job. A single job can pass through contractors, electricians who throw the switches, mechanics who bring the lock boxes, and operations teams who stress test the work. Each party needs the right information, which means it has to leave the head of whoever identified the issue. 

What does incomplete notification data cost?

Poor notification data shows up downstream as jobs starting late, planners chasing detail, costs running over, and a backlog that does not reflect the real story. 

Planners chasing detail is a common theme. Some prospects ask for a mobile app so they can plan in the field. Good notifications offer another route: planners can work more efficiently and effectively across a larger swath of equipment and area when the notifications they receive are complete. 

Across customers with case studies, mobile execution typically increases 15-20% with mobile. Teams also get more data and better data, and they can standardize how that information is collected. 

Without a mobile device, the workflow is manual. Take a picture on a phone, send it to yourself, find the attachment, download it, locate where to attach it, and hope the planner or someone downstream looks for it. 

What's the difference between entering data in SAP, on mobile, and by voice? 

How does the AI agent work?

The Prometheus Group AI agent is trained on 25+ years of maintenance planning and scheduling processes, and it references the customer's own data. It is also skilled at translating SAP or other CMMS/ERP information into plain English. Because the technician is logged into the application, the agent knows who they are and where they are. 

The agent runs as a guided prompt that walks the technician through each step. The company sets up those steps, so they are configurable rather than one size fits all. The level of detail that matters can differ from one plant to another. Without AI, the closest option is requiring fields to be filled out. With AI, technicians can say what they would naturally say to someone over the radio, and the agent converts it into good notification data.

What does creating a notification by voice look like? 

Here is the flow demonstrated for a leaking chiller:

1. Open the assistant. One button opens the AI assistant, and no typing is required.
2. Describe the issue. The technician says: "The chiller is leaking on the Northwest Side." Chiller has a specific meaning at that plant and is probably labeled similarly in SAP. The agent uses that context to identify the equipment.
3. Answer guided follow-ups. The agent asks what the equipment was doing and whether the technician did anything before submitting. The answer: it was operating at reduced capacity, the technician tried to throttle it back, and it was still leaking.
4. Match codes. The agent searches the codes in SAP and relates them to the asset and the damage described, including the cause code.
5. Assess priority. The agent asks whether production is affected, whether safety is at risk, and how likely that risk is. The answer: low safety risk, the leak is just water, and production is reduced. These answers feed a priority matrix, an out-of-the-box S/4HANA feature.
6. Review the notification. The result includes the notification type, location, area, a short description, and long text concatenated from the technician's answers, plus the priority.

The technician never typed a word. The agent also simplifies the language in the long text, which addresses a common customer complaint: feedback arrives, but people describe issues inconsistently. Summarizing is something AI already does well.

The same observation can be handled differently depending on setup. A cracked seal might be a damage code under a particular assembly at one organization and a fault tracked in long text at another. Either scenario can be supported.

The review screen is configurable too. The goal is to minimize the fields presented so the user is not overwhelmed, while still placing the information in the notification.

FAQ

How much setup is required to get the AI notification features running for a field team? 
Less than most teams expect. The AI operates on top of the mobile app a team already runs, using the same connection to SAP that was already built. Effectively no integration is required. Any configuration or requirements identified during discovery are handled during implementation. 

Can the guided follow-up questions be customized? 
Yes, and configurability is a foundational part of the mobile app. Customization by site and by asset type is common. Different sites may have different regulations, including across countries, so a team may want to ask different questions or capture different information. By asset type, questions can differ depending on whether an issue could hurt someone or is inconsequential. 

What devices does Prometheus Mobile run on? 
Prometheus Mobile works on the devices your team already has. The app is native for iOS, Android, and Windows, runs on phones and tablets, and is available from public app stores. There are three like-for-like builds across those operating systems. Prometheus Group is not a hardware vendor, so the app is device agnostic, and it is designed to work offline first. No special hardware is required. 

How does this tie back into existing CMMS/ERP data quality? 
Everything in the mobile application comes directly from the CMMS/ERP, in an almost one-to-one read and write. There is no siloed data. The system pulls the data, formats it so it makes sense to the end user, and leaves the underlying data unchanged. When the information is complete, the app writes it back and pushes it to the CMMS/ERP. It is deeply integrated with the CMMS/ERP and cannot function without it. 

Capturing complete data at the source

CMMS/ERP data quality should not depend on how well every technician knows the system. Maintenance teams do not need to be SAP experts. Technicians can focus on their jobs and speak directly to a device, capturing complete information at the source the first time. Instead of memorizing fields and screens, they use one window and their voice. 

To see how voice-created notifications could work in your environment, talk with the Prometheus Group team about a personalized conversation.