Skip to content
Koidra
Book a demo

Agentic insights

Predix

Reports that say what changed and what to do next, and failures seen before they arrive.

Monday starts with the decision instead of the spreadsheet, and a failing bearing becomes a repair you booked rather than a line that stopped.

How it works

What one record produces, and what you put in The whole operating record, holding sensors, shift notes, work orders and production batches, produces four things: a daily note, a weekly summary, a recommendation on what to change against the KPIs you set, and a work order opened early when a trend shows a failure coming. Your KPIs and your questions go in, and both are answered from the same record. The whole operating record sensors, shift notes, work orders, production batches Daily note what moved since yesterday Weekly summary and what it cost What to change against the KPIs you set Work order, early a failure seen coming Your KPIs and questions as numbers, or in words arrives without being asked for both answered from the same record
Every day and every week, you get a short report on your plant. You also get advice on what to change to hit your targets, and a work order before a machine fails. Ask a question and the answer comes from the same data: sensor readings, shift notes and past work orders.

What it takes over from

  • The weekly report someone rebuilds by hand every Monday
  • Asking three people what happened last week
  • Waiting for a consultant to say what to change
  • A separate condition-monitoring tool whose alerts never become a work order

Pays for itself by

Putting the numbers and the recommended change in front of the Monday meeting, so the hour goes to deciding. Catching a failing bearing early enough to replace it on a planned stop, before it stops the line.

The number a buyer checks: Hours a week spent building reports, and hours of unplanned downtime.

Reports that arrive automatically

Predix turns your operational data into regular reports without requiring someone to build the same analysis every day or week. Daily reports highlight what changed. Weekly reports summarize performance, cost and recurring issues.

Because the reports use the full operational record, they bring together maintenance costs and completion trends, inspection results, asset history, energy use per unit of production, production and quality metrics, and the saved charts and plant-specific KPIs your team already watches.

Different plants and different roles need different information. A plant manager may want production, energy and downtime trends. A maintenance lead may care more about recurring failures, overdue work and maintenance costs. A data center operations team may need an entirely different set of indicators. We work with your team to identify the questions that matter most and shape reports around them, starting from the decisions your team needs to make rather than from a fixed template.

One record, a report per roleFrom one operational record, a daily note of what changed and weekly summaries shaped for each role: production, energy and downtime for a plant manager; recurring failures, overdue work and costs for a maintenance lead.One operational recordDailywhat changed since yesterdayWeekly, for the plant managerproduction, energy, downtimeWeekly, for the maintenance leadrecurring failures, overdue work, cost
Your team does not have to remember to go looking for the important changes. The report each role gets is shaped with you, not picked from a template.

Ask the follow-up question

A report may show that energy use per ton increased during the night shift. The next question depends on the situation. Which line changed? Was a different product running? Was moisture higher? Did someone change a setpoint? Was a piece of equipment unavailable?

You can ask these questions in plain language. Predix looks across sensor trends, operator notes, production records and maintenance activity to provide the context behind what changed.

The follow-up question, answered from one recordA report shows energy per ton rose on the night shift. The follow-up, which line, which product, was moisture higher, did a setpoint move, is answered by reading sensor trends, operator notes, production records and maintenance activity together.Energy per ton rose on nights. Why?Predixreads them togetherFour sources, one timelinesensor trendsoperator notesproduction recordsmaintenance activityThe cause, with its evidence
One question, four sources, one answer. A tool that sees only sensors can explain only the sensors.

What to change, against the KPIs you set

Explaining the past is half the job. The other half is saying what to do about it.

You tell Predix what you are trying to move. That can be a metric, such as energy per tonne or unplanned downtime hours, or a sentence, such as fewer changeovers lost to a wet feedstock. Given that target, it reads the same record the reports come from and recommends what to change: a setpoint, a schedule, a threshold, a maintenance interval, a batch to stop running on the night shift.

A recommendation carries its reasoning, drawn from the trends, the operator notes and the maintenance history it was built from, so the person deciding can see why before they act. It does not act on its own. Acting is what KoPilot is for, and a recommendation is how a site finds out whether it wants that.

From a target to a recommendationYou name the target, as a number or a sentence. Predix reads the record and recommends a change, a setpoint, a schedule, a threshold or a maintenance interval, with its reasoning attached. A person decides. Acting is KoPilot's job.Your targetenergy per tonne down, orfewer wet-feedstock changeoversPredixA recommendation, with its reasoninga setpoint, a schedule,a threshold, an intervalYou decideKoPilot actswhen a site wants that
A recommendation carries its reasoning, and a person decides. Acting on its own is what KoPilot is for.

Why it needs both doors open

AI can only answer questions using the information it can access. A system that only sees sensor data can explain what happened in the sensor data. It cannot know that production dropped in the same week an operator recorded repeated manual overrides, or that a critical asset was under maintenance. Koidra brings machine data, operator observations, production context and maintenance history into the same operational record.

That is what lets Predix answer questions across the plant rather than within a single system, and why it needs both doors open, DataPilot and Maintenance Management, and is sold with both rather than on top of them.

Why Predix needs both doors openMachine data, production context and operator observations come through DataPilot; maintenance history comes through Maintenance Management. Predix reads all of it, which is why it is sold with both and not on top of one.DataPilotmachine data, production context,operator observationsPredixanswers across the plant,not within one systemMaintenance Managementmaintenance history
Both doors, one record. Neither half answers a question about the plant on its own.

Predictive maintenance, from the data you already have

The third thing Predix does with the record is watch it. Because it reads your sensor trends and your maintenance record together, it can see a failure coming and open the job before the failure arrives, with the asset, the checklist and the assignee already attached. It needs both Maintenance Management and DataPilot, and it comes with both.

A basic alarm works well when a single value crosses a fixed limit. Many equipment problems develop without any single signal crossing one. The three conditions below are the kind a threshold cannot express, and in each case the result is not another alert: the system opens a work order with the asset, procedure and assignee already attached, so the issue is followed through to completion.

Predictive maintenance, from the data you already havePredix reads the sensor trends and the maintenance record together. When a trend says a failure is coming, it opens the work order early, with the asset, the checklist and the assignee already attached.Sensor trendsfrom DataPilotPredixsees the failure comingMaintenance recordfrom Maintenance ManagementWork order, earlyasset, checklist, assignee attached
The trend and the record together see what neither sees alone, and the job is opened before the failure rather than after it.

Conditions a threshold cannot express

A sensor that has stopped changing

AND(STDDEV(bearing_temp, 12) < 0.01, mill_running)

The machine is running but the bearing-temperature reading is almost perfectly constant, so the problem is probably the sensor rather than the process.

Then: A maintenance task for the instrument that needs inspection, instead of a generic process alarm.

A risky relationship between two signals

dryer_exhaust_temp - dew_point < 10

Neither signal is abnormal on its own. Exhaust temperature is closing on the dew point, which brings condensation and material buildup.

Then: Maintenance scheduled before the buildup becomes a larger equipment problem.

Gradual degradation against the machine's own history

A vibration signal climbs slowly over days while staying under its absolute alarm threshold, so no single reading ever trips anything.

Then: Deterioration caught against the machine's own recent baseline, and a maintenance task raised before the condition becomes critical.

How a failure announces itself for weeks A vibration trend over several weeks, drawn as a sawtooth. Each cycle climbs as dust builds on the rotor, drops when the machine is cleaned, and starts the next cycle higher than the last. An alarm threshold runs across the top of the chart. Every individual peak sits below it until the final one crosses, at which point the machine fails, even though the trend had been rising for weeks. alarm threshold cleancleancleanclean it breaks week one week six Vibration, cleaned four times, never investigated once
Each clean resets the symptom and none of them touches the cause, so the trend returns higher every time. Four separate work orders, each closed on its own, because the thing connecting them is a line no maintenance system was looking at.

Where this sits

Two ways into one platform, a layer that needs both, and control on top. Each is a complete product with its own payback. How the pieces fit.

  1. Start here

    Maintenance Management

    Work orders, preventive maintenance, asset records and spare parts, on the phone your technicians already carry.

  2. Start here

    DataPilot

    Every signal the plant produces, in one place, live and historical.

  3. Agentic insights

    Predix

    Reports that say what changed and what to do next, and failures seen before they arrive.

  4. Autonomous control

    KoPilot

    Closed-loop control of the process, minute by minute, with no human at the panel.

On top of this sits KoPilot, scoped against the process it runs.

Includes Maintenance Management + DataPilot.

See Koidra on your own plant data

Bring a week of readings from one line. We will show you what the platform sees, what it would have flagged, and what it would have scheduled.

Book a demo