Assetize

All learning series · Suite application · Part 1

Maximo Predict

Interpret predictive outputs with their data and model context.

Start here

Confirm this product is ready, then use the lesson. The readiness checks are the gate for configuration, for a first journey and for production.

  1. Are you ready to use Maximo Predict? Licensing and AppPoints, the records that must exist, and three checklists.
  2. Basics
  3. Configuration
  4. Practical example
  5. Practice and verify
  6. Advanced concepts

Are you ready to use Maximo Predict?

Predict trains a model on history that has an asset, a time and an outcome. The 9.2 guidance prices it at Premium. Health scores are a different product.

Before configuration

The user is Premium, Predict is deployed, and you can name the event you are trying to predict.

Before the first successful journey

A training set with asset identity, timestamps and a label you trust, plus a stretch of time you will not train on.

Before production use

The model, the label definition and the action you take from a prediction have owners. A high score is not a work order until someone makes it one.

Licensing, entitlement and AppPoints

The MAS 9.2 licensing guidance prices Maximo Predict at the Premium tier. The application table shows Predict only in the Premium column. There is no Predict install quantity on the install table. Health is Base. Visual Inspection is Premium and also has an install line. Do not price Predict as either of those. A data-science platform outside the suite, if you use one, has its own licence.

Keep four facts separate for Maximo Predict. Entitlement is the proof of entitlement and the License Information for IBM Maximo Application Suite. The MAS 9.2 licensing guidance supplements that agreement. Available AppPoints are the pool shown on the AppPoints Usage dashboard: Suite administration, then Suite, then Administration, then AppPoints Usage. Assigned access is the security group together with the user type on the person: Limited, Base or Premium, and Authorized or Concurrent. Consumption is what that assignment draws. Authorized users and administrators reserve AppPoints continuously. Concurrent users draw AppPoints while they are signed in. Install AppPoints stay reserved for the life of that environment. A non-production install of a component draws no install AppPoints when the same component is already installed in production. Administrators and User Management users open the dashboard. Other people need the LICENSEUSAGE security group. Compliance enforcement and idle timeout sit on the dashboard Configuration tab and are suite-administrator settings.

The same guidance publishes these user conversion ratios: Limited is 5 AppPoints for one concurrent user and 2 for one authorized user; Base is 10 concurrent and 3 authorized; Premium is 15 concurrent and 5 authorized; an administrative Base user is 10 AppPoints and an administrative Premium user is 15, both as authorized users. A Base administrator who already reserves 10 AppPoints can be given Premium application access with no further AppPoints. Those ratios were read from the MAS 9.2 licensing guidance on 18 October 2026. Your agreement sets how many AppPoints you purchased. An application that opens shows that the component is deployed. Read the proof of entitlement and the dashboard as well.

Installed and enabled applications

Confirm Predict is deployed and a dataset can be registered. Installed means the application opens for a Premium user. Configured means a dataset, a target and a horizon exist. Usable means a model produces a result on data it did not train on, and you can see which asset it refers to.

Foundation configuration

The assets, the site and the history live in Manage or in the feed you connect. Failure codes or work-order outcomes are a common label. Meters and monitor points are inputs when they exist for the same asset and time. You do not need a health score to start, and a health score is not a Predict label unless you define it as one.

Product configuration

Define the target and the horizon before you pick an algorithm. Split a holdout period. Grant model builders Premium access and grant reviewers read access. Review the default that treats a missing label as a non-event.

Infrastructure and integration

Predict runs as a suite application. Training capacity is part of the suite’s sizing. An external historian needs a path, an identity and a licence. No map or mobile client is required for the first model.

Data and operational readiness

The first journey needs enough history to train, a label that means one thing, and a holdout window. The result names an asset and a time. A missing label is data. A Premium refusal is entitlement. A result you cannot tie to an asset is configuration.

Setup order for a prediction

  1. 1. Premium entitlement No Predict install line. Health’s Base tier is not this.
  2. 2. Predict deployed A Premium user can open it.
  3. 3. Target, horizon and labelled history The label is defined before the model runs.
  4. 4. A holdout result on a named asset Then decide what action the result will trigger.
Readiness for Maximo Predict
RequirementWhy neededMandatory/Conditional/OptionalWhere to configure or verifyMinimum working exampleVerification procedureConsequence if missingRemediationSource
Premium entitlementPredict is a Premium application. The install table has no Predict quantity.MandatorySuite administration → Suite → Administration → AppPoints Usage. Manage → Security → Security Groups, and the suite user type.A model builder on Premium.Grant Predict and confirm the dashboard shows Premium.A Base user is expected to open Predict because they can open Health.Use a Premium user. Keep Health on its own tier.Licensing in Maximo Application Suite 9.2
Predict deployedIt is its own application.MandatorySuite administration → CatalogPredict opens and a dataset can be defined.Record the version.Only Health is deployed.Deploy Predict before you load a training set.Modules and applications in Maximo Application Suite 9.2
A target and a horizonThe model needs a question.MandatoryThe dataset definition.Predict a failure within 14 days, label defined as a breakdown work order.The definition states the event and the window.A model is trained with no written target.Write the target first.Modules and applications in Maximo Application Suite 9.2
Labelled history with asset and timeUnlabelled rows do not teach the event.MandatoryThe training extract.Asset identity, timestamp and the failure label on each training row.Count labelled and unlabelled rows. Inspect one failure and the asset it names.The file has sensor values and no outcome.Add the label or stop. Do not rename a health score into a failure.Modules and applications in Maximo Application Suite 9.2
A holdout windowThe result has to be on data the training did not include.MandatoryThe split you configure.The last period held out of training.The scored rows are inside that period and name an asset.The only score is on the training rows.Score the holdout before you talk about accuracy.Modules and applications in Maximo Application Suite 9.2
An action that is still a human decisionPredict does not by itself create the work order.OptionalThe work process you will use if you act on a result.One training asset where a person creates a work order from a result.The work order references the asset and the person, not only the model.Results open work orders with no review.Keep creation as an explicit step.Modules and applications in Maximo Application Suite 9.2

A. Ready to configure

  • The user is Premium and the dashboard agrees.
  • Predict is deployed.
  • The target and the horizon are written down.
  • You have a labelled extract with asset and time, or you know you do not.

B. Ready for the first end-to-end journey

  • The dataset registers with the label visible.
  • Training uses the earlier period.
  • The holdout score names an asset and a time.
  • You can explain one result from the label definition.

C. Ready for production operation

  • The label, the horizon and the model have an owner.
  • External data sources have licences and a path.
  • Acting on a score is a reviewed work process.
  • Health and Predict results are labelled as different numbers.

Sources for this readiness check

Licence statements come from the IBM topic named on the row. Installation of a component is a separate check from entitlement.

Before you start

MAS learning path. Match the component documentation to your installed release and maintenance build before configuring it.

  • Predict and the documented analytics dependencies for your deployment.
  • A training dataset with asset identity, measurements and trustworthy outcome labels.

IBM sources + Assetize exercise

01 · Basics — Understand the building blocks

Predict builds on asset-performance information to estimate degradation or failure. Its outputs need interpretation in the context of the model and the data used.

A prediction is not an observed failure. Validate performance on data that was not used to train the model.

Topics in this series

  • Prediction target
  • Data quality
  • Training and evaluation
  • Interpreting outputs
  • Operational feedback

02 · Configuration — Work through the setup

Use a training environment and match the actions to the IBM guide for your installed build.

  1. Define the prediction target and time horizon before selecting a model.
  2. Inspect data completeness, timestamps and the meaning of failure labels.
  3. Separate training and evaluation periods so future outcomes cannot leak into training inputs.
  4. Follow the documented model workflow and review predictions for the held-out population.
  5. Compare false alarms, missed events and lead time before defining a maintenance response.

03 · Practical example

Fictional teaching example

A fictional pump prediction warns of elevated failure probability within 30 days. That horizon is different from predicting failure tomorrow and requires a different response policy.

04 · Practice and verify

Create an evaluation table with prediction time, horizon, prediction, observed outcome and chosen action.

Expected result

Each result has a model/data context and can be evaluated against later observations without using those observations as inputs.

Common mistakes

  • Using future maintenance outcomes as training features.
  • Presenting a probability as certainty.

05 · Advanced concepts

Explore drift, retraining and action thresholds. Track whether interventions change the observed outcome and therefore the interpretation of model accuracy.

IBM references

Product explanations link to IBM sources. Scenarios and exercises are original Assetize teaching material. Reviewed 2026-10-01.

Maximo ecosystem overview · Manage learning paths · Practice data in Workbench