AI Isn't a Tool. It's Your Teammate: Salesforce's Playbook for Customer Support
Most support teams have budget and access to AI. The blocker is trust and adoption. Salesforce's new AI Fluency Playbook shows how they made 85% of employees confident with AI-and why those who use it daily report 64% higher productivity.
Here's the real insight: this isn't an IT rollout. It's a people strategy.
Technology Alone Is Not a Strategy
Natalie Scardino, Chief People Officer at Salesforce, says it plainly: "technology alone is not a strategy." The difference-maker is how you govern AI and weave it into workflows your agents already live in.
For CX, stop asking "what can the bot do?" Start asking "how fluent is my team at working with the bot?" That shift changes everything-from adoption, to outcomes, to morale.
The Three Pillars of AI Fluency
- Engagement: Tackle fear and skepticism head-on. Ruth Hicken, VP of Workforce Innovation, urges teams to experiment and be okay failing. The goal is comfort and curiosity, not perfection.
- Activation: Move from "I tried it once" to "I use it every day." Make AI part of standard operating procedures, not a side project.
- Expertise: Build human, agentic, and business skills. This is where AI becomes a partner for brainstorming, orchestration, and complex problem-solving-not just task execution.
Real Support Outcomes: Redeployment, Not Replacement
Salesforce put its own approach to work. "Service Agent" handled over 2 million support requests. Costs went down, sure-but the big win was people.
"Because of the benefits and efficiencies of Agentforce, we were able to reshape the function, redeploying our support engineers into forward deployed engineering roles," Scardino shared. And 50% of new hires over 12 months were internal moves. Offload the repetitive work to agents; move humans into higher-value roles.
The Dirty Data Warning
There's a catch. Pierre Mattouche, SVB of Adecco IT and Digital Transformation, warns that agents "will just amplify dysfunction" if your processes are broken. Vague escalations or outdated knowledge? AI will spread those mistakes faster.
Fix and standardize first. Decide where the human stays in the loop. Once people see they remain in control, adoption jumps.
A 90-Day CX Playbook You Can Run Now
- Weeks 0-2: Foundations
- Baseline: containment, FCR, AHT, CSAT, backlog, agent effort.
- Map top 10 intents and top 5 escalation paths. Remove ambiguity.
- Update your knowledge base. One source of truth or nothing works.
- Define "human-in-loop" moments: compliance, refunds, sentiment risk, VIPs.
- Weeks 3-4: Pilot
- Pick 2-3 narrow use cases: password resets, order status, entitlement checks.
- Set policies: tone, data access, escalation rules, refusal criteria.
- Shadow mode first, then limited production with clear guardrails.
- Weeks 5-8: Adoption
- Daily practice: 10-15 minutes of prompt reps per agent with feedback.
- Office hours with a lead to review failures and fix prompts/KB gaps.
- Publish win stories weekly to normalize usage and build confidence.
- Weeks 9-12: Scale and Redeploy
- Expand to new intents and channels. Keep scorecards visible.
- Retrain select agents for roles like bot trainer, knowledge engineer, or workflow designer.
- Update job ladders and incentives to reward AI-enabled outcomes.
Guardrails That Keep You Safe
- Clear escalation triggers: policy, sentiment, dollar thresholds, identity verification fails.
- Knowledge freshness SLA: what gets updated, by whom, and how it's approved.
- Data scope: what the agent can see or do (entitlements, refunds, order edits).
- Audit trail: store prompts, outputs, actions, and handoffs for QA.
- Red-team weekly: test bad inputs and edge cases to harden the system.
What to Measure (and Share)
- Containment rate by intent and channel
- First contact resolution and AHT deltas
- Deflection vs. backlog and SLA adherence
- CSAT on AI-only, AI-assist, and human-only paths
- Agent productivity and time-to-proficiency
- Employee confidence using AI (simple pulse survey)
- Error rate, safety incidents, and escalations avoided
New Roles for Modern Support Teams
- AI Support Specialist (agent assist + orchestration)
- Bot Trainer / Prompt Engineer (improves prompts, tests edge cases)
- Knowledge Engineer (structures articles for AI consumption)
- Workflow Designer (maps automations, defines handoffs)
- Forward Deployed Support Engineer (handles complex, technical issues)
- QA and Policy Reviewer (audits outputs, enforces standards)
Tools Mentioned
Salesforce used Agentforce and its "Service Agent" approach to automate high-volume support while moving people into higher-impact work. Learn more from Salesforce here: Agentforce.
Sharpen Your Team's AI Fluency
If you're building skills by role, here's a practical starting point: AI training paths by job function. Focus on prompt skills, KB design, and human-in-loop practices for support teams.
The Takeaway
AI works best as a teammate. Build fluency, not just features. Fix your processes, set guardrails, ship small pilots, and keep humans in control.
Do that, and you'll get the same upside Salesforce saw: higher productivity, stronger retention, and talent moving up to work that actually matters. As Mattouche put it, "The workplace will feel faster but calmer… more automated, but more human."
For a quick primer on why data quality matters before automation, this overview can help: IBM: What is data quality?
Your membership also unlocks:
