Machine Learning in the Workplace¶
The most difficult part of deploying machine learning is rarely the code; it is the culture change.
The Integration Challenge¶
Building an accurate model is only 20% of the battle. The remaining 80% is getting people to actually use it.
- Trust: If end-users (e.g., call centre reps, warehouse managers) don't trust the model, they will ignore its recommendations. Trust is built through transparency and involving them in the design process.
- Workflow Disruption: A model that requires a user to open a new tab, copy-paste an ID, and wait 30 seconds for a prediction will fail. Models must be integrated silently and seamlessly into existing workflows (e.g., Salesforce, the ERP system).
- Feedback Loops: Every prediction the model makes needs a real-world outcome attached to it eventually to know if it was right. If a user overrides a model recommendation, you need to capture why they overrode it to retrain the model later.
KSB Mapping¶
| KSB | Description | How This Addresses It |
|---|---|---|
| S5 | Deployment, value assessment, and ROI | Translating model performance into business impact |
| S6 | Communicate through storytelling and visualisation | Presenting ML results to non-technical stakeholders |
| B4 | Consideration of organisational goals | Framing technical results in terms of business objectives |
| B1 | Inquisitive approach | Exploring creative ways to explain model behaviour |