Sulzer Pumps U.S.
- Manufacturing
- Industrial Pump Repair
- Service
Custom Software Development
Voice-driven operational intelligence platform
- Voice AI
- Derived State Machine
- Predictive Scheduling
- Operational Dashboards
- Multi-Site Access
- SAP
Table of Contents
Shop Floor Visibility Software: How Sulzer IQ Gave Sulzer Pumps U.S. Real-Time Shop Status
A voice-driven platform that replaced paper travelers and manual status boards with job state derived automatically across two U.S. service centres.
Executive Summary
Sulzer’s U.S. service centres repair industrial pumps through a two-phase workflow tracked in SAP, with dozens of technicians per shift. They had no shop floor visibility software in place. Supervisors could not see which trade would be overloaded next week, which jobs were quietly stalling, or what technicians were observing at the bench. An earlier Kanban-style prototype had already failed, because it asked gloved technicians to stop working and type.
We built Sulzer IQ as a monitoring system rather than a task board. Job state is derived automatically from two inputs already happening naturally: the daily SAP export and four short technician voice commands. Nobody updates a status by hand. Every dashboard recalculates on load.
The Challenge: No Real-Time Production Visibility
Operational reality on the floor stayed invisible until it became a problem.
No look-ahead. Supervisors could not see which trade, whether welding, machining, or assembly, would be overloaded in the next two to four days. Bottlenecks surfaced only after they had caused a delay.
State lived in people’s heads. Whether a job was stalled, waiting on a vendor, or running behind was something a supervisor had to walk the floor to discover.
The floor does not type. Technicians wear gloves and move between tools. Any manufacturing work order tracking system requiring manual clicks was going to be ignored, which is exactly what happened to the prior prototype.
Tribal knowledge was walking out the door. Technicians narrate what they see constantly: wear patterns, anomalies, judgment calls. None of it was captured.
SAP had to stay the system of record. Any solution needed to sit alongside SAP without duplicate entry, or supervisors would abandon it.
This five-item list is the strongest featured snippet candidate on the page. Five parallel items, each opening with a bolded noun phrase, is the structure list snippets favour. Build it as a native bulleted list, not as icon cards, or the structure is lost to the parser.
The Solution: Job State Derived, Never Entered
We started by shadowing the workflow rather than writing a feature list, tracing how a job moves from intake through inspection to customer approval and repair, and mapping where supervisors lose sight of it.
That discovery produced the reframe the project turned on. Sulzer IQ had to be read-only. If a supervisor can update a status by hand, the status eventually becomes wrong.
Four-step diagram sits here, replacing the five-step Empathize/Define/Ideate/Design/Test graphic in the Vitalog template:
Finding the Signal Already in SAP
Real SAP work order visibility depends on using signals the floor already produces. Requiring every inspection step to be formally closed in SAP would have failed, because in practice that rarely happened cleanly. Instead, we identified the real-world marker that a customer had approved the repair: repair work orders appearing in the daily export. The state machine triggers on that, with SAP status as a secondary check.
Step-level state, covering In Progress, Pending External, Complete, Stalled, and Skipped-Flagged, rolls up to work order and job level automatically.
Defining a Voice Command Set the Floor Would Use
Voice AI for manufacturing operations only works when the command set is small enough to be unambiguous. We defined four: start, done, shipped, received back. Each is speakable hands-free without breaking stride.
Beyond those four, the system stores every other piece of technician narration. That record belongs to Sulzer and becomes the base for AI-assisted scoping and predictive maintenance work later.
Predictive Completion Dates
Planned hours, trade assignments, and available capacity project a completion date for each job phase, compared against the customer due date to flag On Time, At Risk, or Late. The same engine looks forward. A skills matrix maps every technician against the trades and tasks they are qualified for, and at what proficiency. The prediction engine reads that matrix alongside planned hours and current load to forecast which technicians should take which jobs next, and how long each job is likely to take them. Capacity stops being a headcount number and becomes a question of who can actually do the work. Supervisors get a staffing view before the shift rather an a reallocation after a job has already slipped. No manual scheduling input, and no one responsible for maintaining a schedule.
Dashboards & Adoption
Three views, each answering one supervisor question.
Trade Look-Ahead.
Which trade is heading for an overload two to four days out.
Assembly Risk.
Which jobs are at risk of missing their date, and why
Tech Utilization.
Who is active, who is idle, adjusted for shift.
All three recalculate on every page load from the current SAP export and voice log. There is no reporting step and nobody is responsible for keeping them current.
Adoption was the risk that killed the previous attempt, so the system was built to tolerate how the floor actually runs. Multi-technician contributions, shift changes, stalled steps, and skipped steps are handled without lockouts. Supervisors get a flag to investigate, not a system that blocks work. Multi-location Super Admin access and full EN/ES support shipped inside the original Phase 1 timeline, alongside three scope additions carried through a formal change register without moving the date.
Results
Reduction in supervisor time spent physically checking job status
Days of advance warning on trade bottlenecks, against zero before
Days of advance warning on trade bottlenecks, against zero before
Share of shop floor jobs carrying live derived status
Share of shop floor jobs carrying live derived status
Manual data-entry steps added to any role: zero
Conclusion
If your team tracks work across paper travelers, spreadsheets, or a status board someone has to remember to update, this pattern probably applies. The data that would give you predictive visibility usually already exists inside your systems and inside the conversations happening on your floor. It is not connected. We find that connection, then build the smallest system that surfaces it automatically.
Beyond those four, the system stores every other piece of technician narration. That record belongs to Sulzer and becomes the base for AI-assisted scoping and predictive maintenance work later.
Predictive Completion Dates
Planned hours, trade assignments, and available capacity project a completion date for each job phase, compared against the customer due date to flag On Time, At Risk, or Late. The same engine looks forward. A skills matrix maps every technician against the trades and tasks they are qualified for, and at what proficiency. The prediction engine reads that matrix alongside planned hours and current load to forecast which technicians should take which jobs next, and how long each job is likely to take them. Capacity stops being a headcount number and becomes a question of who can actually do the work. Supervisors get a staffing view before the shift rather an a reallocation after a job has already slipped. No manual scheduling input, and no one responsible for maintaining a schedule.
Frequently Asked Questions
Technicians speak four short commands at natural points in the job: start, done, shipped, received back. The system maps each to a step-level state change and rolls it up to the work order automatically. No forms, no clicks, no separate logging step at shift end.
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.
Have a project idea in mind? Let's get started growth advisors ready to talk
Tell us about your project and what you’re looking to build. A Liquid Technologies specialist will help you identify the right AI, data management, mobile app, or website development solution based on your goals, budget, and timeline.
- AI-powered solutions — Build intelligent AI models and solutions tailored to your business needs.
- Data-driven solutions — Manage, organize, and leverage your data to support smarter business decisions.
- End-to-end development — From mobile apps to websites, we build scalable digital experiences designed around your goals.
