← Back to AI & Future
AI-020 AI & Future 22 min read For: CTOs

Building an AI Roadmap on the Salesforce Platform

An AI roadmap that skips the prerequisite foundations will fail to deliver — the sequence matters as much as the destination.

VS

Vishal Sharma

Salesforce AI Specialist · Updated May 2026

What you will learn in this tutorial
  • The prerequisite foundations that must be in place before advanced AI features can deliver value
  • A sequencing model for Salesforce AI capability: augmentation, automation, and autonomy phases
  • How to prioritise AI investments against business value and organisational readiness simultaneously
  • The governance model required to manage AI risk as capability scope increases
  • How to communicate the AI roadmap to executives, business sponsors, and delivery teams differently
  • Common roadmap mistakes and how to avoid them

The Three Phases of Salesforce AI Adoption

Successful AI programmes on Salesforce follow a predictable capability progression. The phases are not always sequential in time — some organisations overlap them — but the prerequisites for each phase are real, and skipping them produces predictable failures.

Phase 1 — Augmentation: AI assists humans by surfacing insights, generating drafts, and making recommendations. Humans review everything and take all actions. Einstein recommendations, generative email drafting, case summarisation, and knowledge article suggestions are all Phase 1 capabilities. The technical prerequisite is basic data quality and CRM adoption. The risk is low because humans provide the quality filter.

Phase 2 — Automation: AI takes defined actions within bounded processes without human review of each action. Automated case routing, Einstein Lead Score-driven assignment rules, automated follow-up sequences triggered by AI signals. The technical prerequisite is higher data quality, reliable model accuracy, and exception handling for low-confidence cases. The risk is moderate — errors propagate before humans catch them, so the exception design matters.

Phase 3 — Autonomy: AI agents handle end-to-end processes across multiple steps and channels, escalating to humans only for exceptions. Agentforce customer service agents, autonomous prospecting agents, multi-step approval and fulfilment agents. The technical prerequisite is mature data architecture, robust integration with external systems, well-defined escalation paths, and established governance and monitoring. The risk is significant — autonomous errors affect real customers in real time, and the feedback loop between error and correction is longer.

💡
Insight

Most AI roadmaps that fail do so because they are Phase 3 ambitions built on Phase 1 foundations. The excitement around autonomous agents is legitimate — the capability is real and the potential value is significant. But an autonomous agent running on inconsistent CRM data, without reliable integrations, and without a tested escalation path will produce poor outcomes that damage trust in AI capability broadly, not just in that specific feature.

The Prerequisite Assessment

Before sequencing AI capabilities on a roadmap, assess the current state against five prerequisite dimensions. Weaknesses in any dimension constrain the entire roadmap, not just the features that depend directly on that dimension.

Data quality and completeness: Do the key objects and fields that AI features will reason against have sufficient quality and fill rates? A target of 85%+ completeness on key fields is a reasonable threshold for Phase 2 deployment. Phase 3 requires 90%+.

CRM adoption and usage consistency: Are the processes that AI will automate actually being performed through Salesforce reliably? If a significant proportion of sales activity happens in email and calendar without being logged in Salesforce, AI features that depend on activity signals will underperform.

Integration readiness: For AI agents that need to resolve customer enquiries, do the required integrations to external systems exist, are they reliable, and do they have sufficient SLA headroom to support AI-driven query volumes? Integration failures in agent actions degrade customer experience.

Knowledge base quality: For any AI feature that uses knowledge — agents, Einstein Search, knowledge recommendations — is the knowledge base sufficiently complete, current, and consistently structured? A knowledge base with 40% article staleness will produce AI outputs that confidently present outdated information.

Governance and monitoring capability: Does the organisation have the operational model to monitor AI behaviour, review samples, retrain models, and act on performance degradation? AI features are not deploy-and-forget — the governance capability must exist before Phase 2 deployment and be mature before Phase 3.

🔑
Key Concept

The prerequisite assessment produces a readiness profile, not a pass/fail score. Most organisations are ready for Phase 1 immediately, partially ready for Phase 2, and need specific investments before Phase 3. The roadmap should make those investments explicit — they are the dependencies that gate the advanced capabilities, not optional improvements.

Sequencing and Prioritisation

Once the prerequisite profile is established, AI capabilities are prioritised against two dimensions: business value potential and organisational readiness. High-value capabilities that the organisation is already ready for are the right starting point — they deliver measurable ROI quickly, build trust in AI capability, and create the evidence base for further investment.

A common prioritisation mistake is placing high-value, low-readiness capabilities at the front of the roadmap because they are the most exciting. Deploying an autonomous service agent when the knowledge base is poor and the integration with the order management system is unreliable will produce a poor customer experience regardless of the agent's inherent capability — and the failure is attributed to "AI doesn't work" rather than to the prerequisite gaps that actually caused it.

Sequence roadmap phases such that Phase 1 capabilities fund and validate the prerequisite investments needed for Phase 2. Case summarisation deployed in Phase 1 validates the knowledge base quality needed for Phase 2 automated routing. Automated routing in Phase 2 validates the escalation design and integration reliability needed for Phase 3 autonomous agents. Each phase is both a value delivery step and a validation step for the next.

Governance Model for AI at Scale

As AI capability scope expands from augmentation through autonomy, the governance model must scale with it. A governance model appropriate for Phase 1 (light-touch review, no mandatory approval gates) is insufficient for Phase 3 (autonomous customer interactions, automated financial transactions, sensitive data access).

The governance model for a mature Salesforce AI programme should include: an AI review board with representatives from technology, legal, compliance, and customer experience; a defined approval process for new AI use cases by phase (Phase 3 use cases require more rigorous approval than Phase 1); a model performance monitoring dashboard reviewed on a defined cadence; an incident response process for AI-caused customer or data issues; and an ethics review process for use cases that involve sensitive decision-making (credit decisions, access to sensitive personal data, automated customer communications).

Communicating the Roadmap

AI roadmaps require different communication for different audiences. Getting this right determines whether the programme gets appropriate funding, support, and realistic expectations.

Executive audience: Focus on business outcomes and phased investment. Lead with the value delivered by Phase 1 capabilities in near term, the Phase 2 value enabled by foundational investment, and the Phase 3 transformation enabled by sustained commitment. Do not lead with technology — executives fund business outcomes, not features.

Business sponsors: Focus on process changes and timelines. Be explicit about what the prerequisite investments mean for their teams — data quality work, knowledge base maintenance, operational process changes. Unrealistic expectations about "AI fixes everything" are most common at the business sponsor level and most damaging to programme trust.

Delivery teams: Focus on prerequisites, sequencing, and technical dependencies. Be explicit about what must be true before each roadmap phase can proceed. Delivery teams benefit from understanding why the sequence matters, not just what it is — this enables them to escalate prerequisite gaps rather than discovering them mid-sprint.

Leader Perspective

The most durable AI roadmaps are built on measured outcomes from completed phases, not projected outcomes from future ones. When Phase 1 produces documented ROI, Phase 2 investment is a fact-based decision rather than a hope-based one. Build the roadmap so that each phase earns the next through demonstrated value, not through an upfront multi-year commitment to a feature list that may not survive contact with real data.

Key Takeaways

  • AI adoption follows three phases: augmentation (AI assists), automation (AI acts within bounds), autonomy (AI agents handle end-to-end) — each requires the previous as a foundation
  • Prerequisite assessment across five dimensions (data quality, CRM adoption, integration readiness, knowledge base quality, governance capability) gates the roadmap sequence
  • Prioritise high-value, high-readiness capabilities first — deploying high-value, low-readiness capabilities first produces failures attributed to "AI doesn't work" rather than to prerequisite gaps
  • Each roadmap phase should validate the prerequisites needed for the next — Phase 1 validates data quality for Phase 2; Phase 2 validates integration and escalation design for Phase 3
  • The governance model must scale with AI capability scope — Phase 3 autonomous use cases require an AI review board, defined approval gates, and an incident response process
  • Communicate the roadmap differently to executives (business outcomes), business sponsors (process change and timelines), and delivery teams (prerequisites and sequencing)

Checkpoint: Test Your Understanding

1. An organisation wants to deploy Agentforce autonomous agents for customer service (Phase 3) but their CRM data completeness on key case fields is 72% and their knowledge base has 35% article staleness. What is the correct roadmap recommendation?

A. Deploy the autonomous agents immediately — the AI will improve data quality organically over time
B. Deploy in a limited pilot first and use the results to drive data quality improvement
C. Invest in data quality and knowledge base remediation as explicit roadmap phases before Phase 3 deployment — deploying without these foundations will produce poor agent performance that damages trust in AI capability
D. Escalate to Salesforce Professional Services to manage the data quality risk during deployment

2. Why should AI roadmap phases be designed so that each phase validates prerequisites for the next?

A. Salesforce's licensing model requires sequential phase activation before advanced features unlock
B. Each phase simultaneously delivers value and validates whether the foundational conditions for the next phase are in place — this converts prerequisite gaps from mid-project surprises to planned investment decisions
C. Sequential phases are required to maintain AI Trust Layer compliance across the programme
D. It ensures that all phases receive equal investment, preventing budget concentration on the most advanced capabilities

3. What is the most important element to emphasise when communicating an AI roadmap to executive stakeholders?

A. The specific Salesforce AI features included in each phase and their technical capabilities
B. The prerequisite investments required and the data quality work that must be completed
C. The governance model and AI review board structure required for Phase 3
D. Business outcomes and phased investment — executives fund outcomes, not features; lead with what each phase delivers and what investment it requires, not the technology beneath it

Discussion & Feedback