# From Prediction to Action: How to Turn AI Outputs Into Decisions

[Salesforce](https://yomu.fyi/company/salesforce) · Scott Nyberg · Aug 24, 2026

**Type:** Problem & solution

## Summary

Salesforce addressed an operational challenge where sellers faced roughly 12,000 dashboards and over 20 applications outputting machine learning predictions without clear next steps. The engineering team reframed machine learning outputs as raw signals rather than standalone answers. To bridge the gap between assessment and action, they built a Next Best Action layer that combines model signals, business logic, and contextual institutional knowledge into actionable recommendations. They integrated this layer with an AI agent using Model Context Protocol (MCP) tool contracts, enabling dynamic discovery and explicit handling of missing data. Finally, rather than introducing a separate dashboard destination, the agent serves on-demand recommendations directly inside Slack where sellers already collaborate.

## Context

Sellers at Salesforce faced roughly 12,000 dashboards and more than 20 applications producing conventional machine learning predictions, scores, and alerts. Although models reliably assessed conditions such as churn risk or stalled opportunities, users had to manually interpret the scores and decide what actions to take.

## Approach / What changed

The team created a Next Best Action layer that merges three inputs: model signals, business logic rules, and institutional knowledge. They connected this layer to an AI agent using the open standard Model Context Protocol (MCP), defining explicit tool input schemas, return types, and absence handling. To avoid creating another dashboard destination, recommendations are delivered on demand directly inside Slack.

## Takeaways

- Treating model outputs as signals rather than answers requires combining them with business logic and contextual institutional knowledge to generate actionable recommendations.
- Model Context Protocol provides explicit schemas and contracts that allow AI agents to dynamically determine needed tools at runtime and recognize the absence of data rather than guessing.
- Delivering AI recommendations via a pull-based interface in existing workspaces like Slack prevents the cognitive overhead of adding new dashboards or unsolicited notifications.

**Tags:** [Architecture](https://yomu.fyi/topic/architecture), [LLMs](https://yomu.fyi/topic/llm), [Machine Learning](https://yomu.fyi/topic/machine-learning), [Recommendation Systems](https://yomu.fyi/topic/recommendation-systems)

[Read original post](https://engineering.salesforce.com/from-prediction-to-action-how-to-turn-ai-outputs-into-decisions)