InsightForce — Solution Overview

InsightForce is a Salesforce-native Slack intelligence layer built on Agentforce. It captures organizational signal from Slack conversations, structures it into a purpose-built Salesforce data model, and surfaces it through an AI agent that leadership can query directly from within the CRM. No middleware. No manual extraction. No third-party sync.

The integration pipeline is built entirely on Salesforce platform primitives: a hand-crafted OAS 3.0 spec registered as an External Service, authenticated via External Credential and Named Credential, and exposed as Invocable Actions that feed four Agent Actions in Agentforce Agent Builder. The agent uses Flex prompt templates for auto-categorization and executive digest generation, and writes all captured intelligence into a six-object custom data model.

How It Works

When a user queries the InsightForce agent, it routes the request to one of four topics — Search & Retrieve, Capture & Categorize, Trend Detection, or Leadership Digest. Each topic maps to one or more Agent Actions backed by live Slack API calls: channel search, conversation history retrieval, thread traversal, and message write-back. Retrieved content is processed through prompt templates and persisted into Salesforce as structured records, making every captured insight searchable, reportable, and actionable from inside the org.

The agent operates on a shared bot token (xoxb-) with scoped OAuth permissions and routes all API traffic through Salesforce's Named Credential framework, keeping authentication centralized and auditable.

Integration Architecture

The end-to-end pipeline follows a single pattern:

OAS 3.0 Spec → External Service → Named Credential → Invocable Action → Agent Action → Agentforce Agent

Four Slack operations are defined in the spec and registered as External Service operations in Salesforce:

Authentication uses a Custom-type External Credential with a Bearer token on a shared principal. The Named Credential binds to https://slack.com/api as the base URL. Each External Service operation auto-generates an Invocable Action, which is then manually registered as an Agent Action and scoped to the InsightForce agent in Agent Builder.

Data Model

Six custom objects, deployed in three waves to resolve cross-object dependencies.

Insight__c — Core Intelligence Record

Every captured Slack signal lands here. The central object in the data model.

Contributor__c — Slack User Profiles

Slack users enriched with Salesforce context. Created or updated at capture time.

Trend__c — Detected Themes

Agent-identified recurring themes across Insight__c records. Created and incremented by the Trend Detection topic.

Product_Feature_Request__c — Structured Product Signal

Created when the categorization prompt template identifies product feedback. Links back to the source Insight__c record.

Correction_of_Error__c — Risk and Process Failure Capture

Created when categorization identifies a process failure, misalignment, or risk signal in Slack.

Insight_Reference__c — Trend Junction Object

Many-to-many junction between Insight__c and Trend__c. Carries an agent-assigned relevance score.

Object Relationships

Deployment Sequence

Three-wave deployment to satisfy Salesforce metadata dependency resolution.

Agent Topics and Actions

Prompt Templates

Both templates use the Agentforce Flex Template type and are invoked inline by Agent Actions.

Auto-Categorization

Input: raw Message_Text__c. Output: JSON payload mapping to Category__c, Urgency__c, and Summary__c. Markdown stripped before upsert.

Executive Summary

Input: merged Insight__c records from a configurable lookback window, filtered by urgency threshold. Output: structured digest covering recurring themes, key decisions, open risks, and recommended actions. Can be piped directly to postMessage for Slack delivery to a designated leadership channel.

Requirements