What was done
A payments provider that runs debit card programs for credit unions across Canada wanted an AI layer in front of its partner support case queue. Consultancy Lane Four built a customer-facing Agentforce agent on the partner portal that answers from the company's knowledge base and opens a case when a person is needed. The main work was non-technical: the client came in with a 25% deflection target, but nearly 80% of in-scope inquiries needed actions in external systems the agent couldn't reach, so scope and KPIs were reset to what automation could realistically handle.
Outcome
The consultancy reports the agent launched and remains in production with more efficient workflows, but publishes no deflection or resolution numbers. The useful lesson is that system access, not the AI, set the ceiling on deflection.
Evidence: Vendor case study (client unnamed). Figures are as reported by the source and have not been independently audited by AgentsNorth.
Prompt Reconstructed by AgentsNorth
The source didn't publish its prompt. This is our reconstruction of a starting prompt for a similar workflow, not the company's actual prompt.
You are the support assistant on a partner portal for credit unions that use our debit card program. Answer questions only from the approved knowledge articles and link the article you used. If the request needs an action in a system you can't access (card status changes, settlement or dispute lookups), don't guess: collect the credit union name, contact, card program and a clear description, then create a support case and give the case number.