06 · After launch
Model care: machine learning model maintenance after launch
Machine learning model maintenance after launch: data collected as people use your app becomes training data, so the model gets better with use. Hard cases are labelled, the model is retrained and tested, and a one-page accuracy report arrives every month. A monthly block of engineer hours.
Who it is for
Is this the right step for you?
- The model is in production and you have no ML engineer, or one.
- Users report wrong results, and the complaints arrive faster than fixes.
- The app collects user data, and none of it goes back into the model.
- You are hiring your own ML engineer and need cover and a handover until then.
What you get
Model care: the deliverables
- 01
A data loop: recordings where the model failed or was unsure are collected from the app and labelled.
- 02
Regular retraining on that data, tested before each new model ships.
- 03
A test set that grows with real use, so accuracy is measured on the cases your users bring.
- 04
A one-page accuracy report every month.
- 05
Fixes when a new phone or OS version breaks something.
- 06
Priority response when the model misbehaves.
- 07
A handover when you hire your own engineer.
How it works
From first message to done
- 1
At release: the baseline test set and accuracy are recorded, and the app starts collecting hard cases.
- 2
Every month: new cases are labelled, the model is retrained and tested, the report goes out.
- 3
When something breaks: you report it, an engineer picks it up with priority.
- 4
When you hire: documentation, walkthrough and a handover period.
One AI engineer part-time, with a tech lead on call.
Accuracy at or above the release baseline, measured every month on a test set that grows with real use.
Compared with the alternative
One in-house ML engineer carries the whole model and can leave. A monthly block of hours gives you a team that already knows the model, costs less than a hire, and hands over when you are ready to bring the work inside.
Questions
What teams ask about model care
Yes, after a two-week technical research to set the baseline.
Accuracy below the baseline on the agreed test set, or a failure after an OS or device update.
Yes. Collection follows your privacy policy and consent flow, and we can help design the opt-in.
The engagement runs month to month.
How we work
What you can count on
Who maintains it after delivery?
Documentation and a handoff plan come with every engagement. The goal is that your team can take the model over.
What if the AI works but users do not care?
The monthly report tracks the business metric next to the accuracy figure.
What if our engineer leaves?
The engineer on your model is backed by a tech lead and a team, so cover continues through holidays and hires.
Next step
Book a scoping call
Book a scoping call. For a model already in production the call agrees the baseline test set and the monthly hours. A two-week research sets the baseline when the model was built by someone else.