Skill Level: Any Skill Level

Learn about what you will need to think about as you deploy your predictive maintenance models.


Deployment is the process of using your new predictive maintenance models to make improvements within your organization. This can mean a formal integration such as the implementation of Maximo APM PMI model producing anomaly detection scores. Alternatively, deployment can mean that you use the insights gained from the model development to effect some change in your organization. For example, perhaps you discovered alarming patterns in your data indicating faster degradation of assets over a certain age in a certain location. These results may not be formally integrated into your systems, but they will be useful information for your reliability engineers.

In general, the deployment phase consists of two activities:

  •  Planning and monitoring the deployment of models
  • Completing wrap-up tasks e.g. report writing or and project review


  1. Plan for deployment

    Resist the impulse to quickly share the exciting results of your modelling efforts. Instead, ensure you take the time and plan for smooth deployment.

    • Summarise model results and findings so that you can determine which to be integrated and the findings that are worth presenting to your organization.
    • For each deployable model, create a step-by-step plan for deployment and integration.
    • For each important finding, create a plan to distribute to decision makers.
    • Consider how the deployment might be monitored. For example, how will model be updated? How will you decide when the model is no longer useful?
    • Identify any potential deployment problems and plan for contingencies. For example, decision makers may want more information on modeling results and may require further technical details.
  2. Planning for deployment monitoring

    In a real-world deployment and integration of predictive maintenance models, your work might be ongoing. Models need to be evaluated periodically to ensure effectiveness and to make improvements.


    Analyse the following as part of your efforts to monitor your models

    • For each model or finding, which factors or influences need to be monitored?
    • How can you ensure the continuing validity and accuracy of each model?
    • How will you determine when a model has outlived its useful life? You should be specific about accuracy thresholds or data changes etc
    • What will happen when a model is no longer useful? Will you retrain the model with newer data? Make some slight adjustments? Or will the changes require a entire new model building effort?
    • Can this model be used for issues once it has expired? Model documentation is very important to help make this decision.
  3. Write a final report

    Think about the audience for your report. Are they technical Maximo developers or maintenance managers? You may need to create separate reports for each audience if their needs are different. Either way, your report should contain:

    • Comprehensive description of the original business problem
    • The process used to build the model
    • Project costs
    • Deviation from the original plan
    • Summary of your results, both models and findings
    • Overview of the deployment plan
    • Any recommendations based on insights gained during the project.
  4. Review the project

     This is the final step of the CRISP-DM methodology, and it offers you a chance to think abour lessons learned during the predictive maintenance project.

    Ideally, you should conduct a brief interview with those involved in the project, seeking to discover: 

    • What were their overall impressions of the project?
    • What did they learn during the process–both about predictive maintenance and their Maximo and asset data?
    • Which parts of the project went well and which did not? For any areas that did not go well, what could be done in future to avoid those problems.

    After the model has been deployed, you might also interview those affected by the results such as reliabiity engineers, asset managers, and the maintenance team. Find out if they felt the project was worthwhile and offered the benefits it set out to create.

    The results of these interviews can be summarized along with your thoughts about the project with focus on the lessons learned from the experience of predictive maintenance modelling.

Join The Discussion