How to Use AI with Salesforce for Sales: Pipeline Hygiene, Call Updates, Lead Triage and Forecast Prep

AI is most useful in Salesforce when it works on your real opportunity, account, lead and activity data: finding pipeline gaps, turning call transcripts into opportunity updates, sorting new leads and preparing you for the forecast call. Most reps should start with a report export into their company approved AI assistant plus a short Salesforce field guide, and add a connector only when their admin approves one.

This guide covers five workflows for prospecting reps, closing reps and managers, and what to keep out of AI's hands. Salesforce offers its own AI features; check current Salesforce documentation and your admin for what your org has. The workflows here work with any capable AI assistant. On HubSpot instead? See the HubSpot guide. For finding the people who aren't in your CRM yet, see How to Use AI with LinkedIn Sales Navigator.

Three ways to bring AI to your Salesforce data

Where to start: with exports. Every prompt below works on an export and works the same way later through a connector. When a connector is approved, start read only.

What AI can see in your org

Salesforce orgs are often tightly controlled. Three platform basics shape what AI gets:

1. Field level security. Your profile and permission sets decide which fields you can see and edit. A hidden field won't appear in your export, or usually in a connector acting as you.
2. Sharing rules. You may see only your records, your team's or a region's.
3. Report types. A report can only combine the objects its report type allows. If you need opportunities with contact roles, or opportunity history, the right report type has to exist. Ask your admin.

These settings differ by company, so confirm yours with your admin.

Before you start: build a Salesforce field guide

AI doesn't know that your org has two opportunity record types, what "Commit" means on your team or that a stage change fails unless a custom field is filled. Write it down once. It works like a seller context file for your CRM (how to build one).

# Our Salesforce field guide

## Opportunity record types
{Record type}: {which deals use it}

## Stages, exit criteria and forecast category
{Stage}: {what must be true to leave it} · Forecast category: {value}

## Key fields (label, API name, rule)
Close Date (CloseDate): {rule, for example "never in the past"}
Next Step (NextStep): {rule, for example "a dated action"}
{Custom field label} ({API name ending in __c}): {meaning, allowed values}

## Validation rules we hit
{Stage}: requires {fields}

## Leads
Statuses: {list}. Convert when: {rule}

## Rules
Never invent a value. Write MISSING if unknown.
Use our exact stage names, picklist values and field labels.

Labels are what you see on screen. API names are what a connector or an admin import uses. Listing both lets your admin act on AI suggestions without guessing which field you meant.

Attached are the column headers and 20 sample rows from two Salesforce report exports: my open opportunities and my open leads. Draft our Salesforce field guide using the template in this Project. Use only what appears in the files. List what you can't determine (exit criteria, validation rules, API names, forecast category mapping) as questions for my Salesforce admin.

Then ask your admin or RevOps lead to review it. Thirty minutes of their time improves every workflow below.

Workflow 1: Opportunity hygiene in five minutes

Pipeline hygiene makes the forecast trustworthy. Salesforce style forecasting adds one more check: whether each deal's stage and forecast category agree.

The report: open opportunities with stage, forecast category, amount, close date, next step, last activity date and the number of contact roles (if your report type allows it).

Using our Salesforce field guide and the attached export of my open opportunities, check each opportunity against our org's rules:
1. Stage and forecast category that disagree with the field guide (for example Commit at an early stage)
2. Fields the field guide says this stage and record type require, left blank
3. Close dates in the past, or outside the quarter the forecast category implies
4. Next steps that aren't a dated action
5. Fewer contact roles than the field guide expects at this stage
6. No activity in 14 days

Return a table: opportunity, record type, issue, suggested fix, the question I must answer to fix it. Sort by amount, largest first. Use the exact picklist values from the export. Don't invent values. This export reflects my permissions and may be incomplete.

Fix what you can yourself. The open questions become your agenda for the next buyer conversation. Once the check is reliable by hand, make it a weekly routine; see AI Sales Automation.

Workflow 2: From call transcript to opportunity update

AI can draft the logged call, the opportunity updates and the recap email in one pass. The Salesforce part is checking each update against what you can edit and what validation rules will accept.

Here is my call transcript with {contact} at {account}, the current values on opportunity {name} and our field guide. Produce:
1. A logged call: subject line and 5 bullets, pains in the buyer's words, next steps with owners and dates
2. Proposed updates as a table: field label, API name, current value, proposed value, supporting transcript quote
3. Any stage change, the exit criteria it meets or misses and the fields a validation rule will require
4. People mentioned who aren't contact roles yet, with their likely role
5. A recap email under 150 words, no new pitch
Use only the transcript. Mark anything unclear UNCLEAR.

Read the quote column, then the validation flags. The two ways this goes wrong are an update the buyer never said and an update Salesforce will reject at save. The full workflow, including recording consent, is in AI for Sales Call Notes.

Workflow 3: Lead triage before you convert

In Salesforce, leads sit apart from contacts and accounts until someone converts them. That split causes a familiar mess: a "new" lead at an account your team already owns, or three leads from one company worked by three people.

The reports: open leads (name, title, company, email domain, source, status) and an account list with owner and type.

Attached: my open leads and our account list, both exported from Salesforce today.
1. Match each lead to an existing account by company name or email domain, with your confidence. Flag near matches (spelling variants, subsidiaries) for me to check.
2. Flag leads at customers or accounts with an open opportunity, with the account owner, so I coordinate first.
3. Group leads from the same company.
4. Rank the top 10 unmatched leads by fit with the best fit customer profile in my seller context file, one line of reasoning each.
Recommend only. I'll convert, merge or reassign in Salesforce myself.

Never let AI merge or convert records. Merges are hard to undo, and conversion often triggers routing your ops team designed.

Workflow 4: Forecast call prep

Reports show where deals sit now. AI can explain what moved. Use two exports of the same opportunity report a week apart, or an opportunity history report if your org tracks field history (ask your admin).

Compare these two exports of my open opportunities (last Monday and today) using our field guide. Tell me:
1. Changes in stage or forecast category, either direction
2. New, closed won and closed lost opportunities, with totals
3. Close dates pushed, by how many days, and any pushed twice
4. Amount changes over 10 percent
5. My Commit and Best Case totals for this quarter, calculation shown
6. The 3 deals most likely to slip, with evidence
7. Three questions my manager will likely ask
Use only the exports.

Check the math. AI can miscount or filter the wrong rows in large files. Spot check two totals against the live report before the call, or ask an assistant that can run code to calculate them with code. Managers can run the same prompt on the team's report before one on ones; see AI for Sales Managers.

Workflow 5: Account coverage map for enterprise deals

Enterprise deals often span a parent account and several child accounts. AI can turn their contacts into one map of who you know and who is missing.

The report: contacts across the parent and child accounts, with title, account, contact role on open opportunities and last activity date.

Using the attached contacts report for {parent account} and its child accounts, and opportunity {name}:
1. Group contacts by child account and buying role: economic buyer, champion, technical evaluator, user, procurement, unknown
2. Flag anyone with no activity in 60 days
3. List missing buying roles by child account
4. Name the 3 people to reach next, and why
Don't invent people, titles or roles. Write UNKNOWN when unsure.

AI doesn't know people who aren't in Salesforce, and if pushed it may invent them, a classic hallucination. Treat the gaps as a research list and confirm names and titles yourself; How to Use AI with LinkedIn Sales Navigator shows how to do that without copying profiles into AI.

What not to automate in Salesforce

Rolling this out to a Salesforce team

Enterprise orgs move through admins, security reviews and change windows. Plan for that from day one.

1. Confirm the policy

   Which assistant is approved, and may Salesforce exports go into it?

2. RevOps owns the field guide

   and updates it when stages, record types or validation rules change.

3. Standardize the reports

   Your admin builds the four or five reports these workflows need, with only the required fields, and decides who may export them.

4. Pilot with three reps for two weeks

   on Workflows 1 and 2, using the standard reports. Track how many hygiene flags each rep clears and how many proposed updates fail validation.

5. Add a connector later, if at all

   Admin and security choose the user it runs as, read only first, and the objects it may touch.

6. Share the prompts

   as prompt templates in a team workspace.

The AI for Sales Certification teaches these workflows on reps' own deals, with tool agnostic labs that work in Salesforce or HubSpot. Sales leaders can see team options.

Frequently asked questions