AI on Salesforce is rapidly evolving. Consulting on AI projects has its own patterns.
Where AI fits in Salesforce:
- Einstein for Sales — opportunity scoring, lead scoring, forecasting predictions, activity capture.
- Einstein for Service — case classification, article recommendations, agent assist, chatbots.
- Einstein for Marketing — send-time optimisation, content selection.
- Agentforce — autonomous AI agents with tools, reasoning, workflows.
- Prompt Builder — reusable prompt templates with merge fields.
- Atlas Reasoning Engine — Salesforce's own LLM platform.
- Data Cloud + Einstein — unified data fueling AI.
Discovery for AI projects:
1. Use case identification. Where does AI add value?
- Repetitive cognitive tasks (classify, summarise, route).
- Pattern recognition (likely to churn? likely to win?).
- Generative work (draft email, summarise meeting).
- Conversational interfaces.
2. Data readiness. AI needs data:
- Volume — enough data to train / RAG against.
- Quality — clean, labeled, accessible.
- Permissions — AI accessing data needs to respect sharing.
- PII — Trust Layer / data masking strategy.
3. Outcome definition. What does success look like?
- Quantitative — case classification accuracy, forecast accuracy.
- Qualitative — agent feedback, user adoption.
4. Trust & governance.
- Einstein Trust Layer — masks PII, audits prompts.
- Bias auditing — does the model treat groups fairly?
- Explainability — can you defend AI's decisions?
- Human-in-the-loop — when does AI suggest vs decide?
Implementation considerations:
- Prompt engineering — Custom Metadata for prompts; iterate.
- RAG over Knowledge — Knowledge articles + Data Cloud for grounded answers.
- Tool definition — Agentforce agents call Apex methods; design carefully.
- Cost monitoring — LLM calls are billed; track per-feature usage.
- Fallback paths — when AI service down, app must still function.
Change management:
- Sales reps trust AI scoring? Build trust through transparency.
- Service agents accept AI suggestions? Make AI helpful, not annoying.
- Customers comfortable with chatbots? Clear handoff to humans matters.
Common pitfalls:
- AI as solution looking for a problem. "We should add AI somewhere" — wrong direction.
- Underestimating data work. AI needs clean data; you do that work first.
- Over-promising. AI doesn't replace humans; it augments.
- Ignoring cost. LLM calls add up; monitor.
- No human-in-the-loop. AI mistakes happen; need oversight.
Senior consultant insight: AI projects look glamorous but are mostly data work. Consultants who claim "we can ship Agentforce in 6 weeks" are usually leaving out the data prep.
Realistic timeline for production AI: 6-12 months for serious work, including data, build, test, change management.
The most senior framing: AI is a feature, not a strategy. The strategy is the customer experience or business outcome. AI helps achieve it.