Case Study · Salesforce · March to June 2026
Salesforce Service Cloud Consultant
Improved the Service Cloud operation that took utilisation from 10% to 50%.
Fig. 01 · Demand analysis: 342 support emails, ten case types
One repeated question drove a third of all volume.
Recreated from the engagement's demand analysis. Client-internal data not reproduced.
The Brief
The client trains mothers and women returners into Salesforce careers. Its support operation ran largely outside its own platform: the same questions answered by hand, knowledge held in people rather than systems, and Service Cloud utilised at roughly ten percent of its licensed capability. The engagement was to change that, and to answer, honestly, whether an AI agent belonged in the solution.
The Approach
Nothing was configured until the demand was understood. A corpus of 342 support emails was pulled and categorised into a ten-type case taxonomy, revealing that a single repeated question drove a third of all volume, and a fifth of the inbox was junk that should never reach an agent. Every downstream decision, from case-type fields to Knowledge structure to macro design, traced back to that evidence.
The build ran in two-week DSDM sprints behind formal gates. With no access to the client's production org, a complete demonstration environment was built from scratch in OrgFarm (cases, queues, assignment rules, Knowledge) and passed Gate 1 sign-off before a single production setting was touched. Production followed a no-sandbox release pattern: pre-build audit, cloned known-good components, build-test-iterate, every action logged.
The Demand
Nothing configured until the demand was understood. Every downstream decision traced to this.
- 33% One repeated question: answered by a single Knowledge article and macro
- 20% Junk and misdirected mail: filtered at intake, never reaching an agent
- 47% Eight further case types: each mapped to queues, macros and articles
What Was Built
Gated DSDM sprints. A complete demo org built in OrgFarm and signed off before a single production setting was touched, then a no-sandbox production release with every action logged.
- 10 types · Case taxonomy Queues, assignment rules and Service Console, derived from the demand analysis
- 17 macros · Macros & Quick Texts Lightning Macro Builder, each macro linked to an email template; 16 merge-field Quick Texts alongside
- 10 articles · Knowledge base Structured for case deflection and agent grounding
- Agentforce · Service Agent, built System prompt, Data Library grounding, topics, Embedded Messaging (Deploying it was the client's cost call)
- 26 docs · Governance suite Gate sign-offs, Jira structure, decision log, documented handover
The team now runs its own system, on its own documentation, with an audit trail behind every decision.
The delivery
What Was Delivered
- Ten-type case taxonomy with queues, assignment rules and Service Console configuration derived from the demand analysis
- 17 production macros in Lightning Macro Builder, each linked to a Lightning email template; 16 merge-field Quick Texts
- Ten-article Knowledge base structured for case deflection and agent grounding
- Working Agentforce Service Agent, built and costed at list pricing; the client chose not to deploy on cost, and now runs the reusable macros and Quick Texts through their own admin
- Training package: 16-page SOP, troubleshooting reference, 8:25 training video
- 26-document governance suite (Jira structure, gate sign-offs, decision log) closed with a documented handover
From discovery to commission
2025 · The three-week discovery · March 2026 · This build, commissioned
The Outcome
Platform utilisation rose from 10% to 50%. The support team runs its own system, on its own documentation, with an audit trail behind every decision, including the client's decision, on cost, to run the reusable build rather than deploy the agent. The engagement closed formally in June 2026, and the strongest evidence of the 2025 discovery that preceded it is that this build was commissioned at all.