We score where AI actually pays off in your CRM, from lead scoring to knowledge answers, then sequence twelve months of work with a metric and an estimate for every step.
Salesforce ships AI features every release, and your board has asked what you are doing with them. The honest answer depends on your data, your licences and where your people lose time. A roadmap gives you that answer in writing, with numbers.
Each card states what the feature does, what it needs from your data, and the number we use to judge it. If the number cannot be measured today, the first task is to measure it.
Ranks inbound leads by likelihood to convert from history and firmographics. Needs twelve months of won and lost leads with consistent fields. Metric: conversion rate of top-decile leads versus the rest.
Condenses long case threads and call transcripts so agents and managers read in seconds. Needs case history and email-to-case. Metric: handle time and wrap-up time per case.
Recommends the next step on an account or case from signals and rules. Needs clean stage data, logged activity and a feedback loop. Metric: acceptance rate, then win rate on accepted actions.
Drafts replies and follow-ups in your tone, grounded in the record. Needs templates, tone guidance and a human review step. Metric: first-response time and edits per draft.
Answers agent and customer questions from your knowledge base with citations. Needs a curated knowledge base, retrieval (RAG) and access control. Metric: deflection rate and accuracy on an evaluation set.
The sales-side cases are described as a product on our sales copilot page; the service-side ones under customer support AI.
Value is time saved, revenue influenced and risk reduced. Effort is data readiness, licences, integration work and governance. We score both in a workshop with the people who own the numbers, then plot the result.
Neither answer is right for every use case. The questions below decide it per feature; the usual result is both, with Agentforce in the console and an MCP or RAG service behind it. The reasoning is spelled out in Salesforce and LLMs.
| Question | Lean Agentforce / Einstein | Lean external LLM via our services |
|---|---|---|
| Where does the data live? | Mostly inside Salesforce or Data Cloud | Spread across CRM, ERP, documents and ticketing |
| Who uses the feature? | Reps and agents inside Salesforce | Customers, partners and other applications too |
| How much model control do you need? | Platform-managed models and Trust Layer are enough | You need model choice, evaluation sets or fine-tuning |
| What is the cost shape? | Licence and per-conversation add-ons are acceptable | Token usage on your own infrastructure is cheaper at your volume |
| Who audits it? | Platform guardrails and logs satisfy your policy | You need your own guardrails and audit trail (see AI security) |
Foundations first, then the low-effort wins, then the use cases that need a year of clean data. Every quarter ends with a gate where a pilot either scales or stops.
Data readiness fixes, consent flags, a governance line and evaluation sets. Case summarisation piloted in one queue with baseline metrics captured two weeks before switch-on.
Summarisation to every queue. Email drafting with a mandatory review step. Knowledge base curation starts, and the retrieval service is built and evaluated offline.
Knowledge answers live for agents first, then for customers behind deflection metrics. Lead scoring trained on twelve months of outcomes and compared with the reps' own ranking.
Recommendations in the console with a feedback loop. Monthly evaluation runs continue. The roadmap is re-scored with a year of real numbers.
Two to four weeks of the metric before anything changes, captured from Salesforce reports the team already trusts. No baseline, no build.
One team or queue for four to six weeks, with a comparison group where the volume allows. The gate is a number agreed in advance, not a demo.
A fixed evaluation set re-run monthly for accuracy and cost per outcome. The method is described in measuring AI ROI.
The free AI readiness assessment takes ten minutes and gives a first view of data, skills and governance gaps. Bring the result to the roadmap workshop, or use it to decide whether a workshop is worth it yet. For AI beyond the CRM, start with AI strategy.
What the platform's own agents can do, and when to reach outside them.
ReadA guide to connecting CRM data to external models safely.
ReadMeeting prep, notes and next steps for reps, inside the CRM.
ReadDeflect, assist and escalate with knowledge-grounded answers.
ReadThe same scoring method applied across the whole business.
ReadImplementation, clouds, AI, development, integrations and support.
BrowseNo. The roadmap tells you which licences each use case needs, and where an external model behind an MCP or RAG service is the better fit. Several use cases run on your existing edition plus our services; others genuinely need the platform add-ons.
Two to four weeks: one week of data and org discovery, a scoring workshop with sales, service and IT, then a written roadmap with a metric and an estimate band per use case. Larger orgs with several business units take the longer end.
Usually case summarisation or email drafting. The data already exists, the risk is low because a human still sends the message, and the metric (handle time, first-response time) is already measured. Lead scoring needs twelve months of clean outcome data first.
Yes. Agentforce and Einstein configuration, Apex and LWC, integrations, and the retrieval and MCP services that connect Salesforce to external models are all delivered by the same team that wrote the roadmap.
Each use case gets a risk classification, a data-access boundary and an evaluation set before it ships. The approach follows our AI governance practice, and the roadmap flags any use case that would fall under EU AI Act obligations.
Consistent stage and status fields, activity logging that reps actually do, consent flags on contacts, and a knowledge base somebody curates. The discovery week measures each of these; see our guide to data readiness for AI.
Tell us about your systems, your data and your constraints. We reply within one business day with a straight answer and a suggested next step.