Salesforce reports that sellers expect AI agents to cut email drafting time by 36% once the technology is fully implemented. The result describes expected efficiency among respondents, not a promise that an AI-written email will convert better.
Benchmark table
| Benchmark | Result | Practical interpretation |
|---|---|---|
| Expected email drafting time reduction | 36% | Measure minutes per approved message before and after a controlled test |
| Sales teams using agents now | 54% | Adoption is already common in the source sample |
| Sales teams expecting agents within two years | 34% | Drafting workflows will likely change soon |
| Survey population | 4,050 professionals | Results are reported survey evidence, not a universal norm |
Source: Salesforce State of Sales, published 2026 and accessed August 5, 2026.
The metric that matters is approved-message time
Do not measure the time to generate a draft. Measure time from brief to approved send, including fact checking, personalization, compliance review, CRM logging, and edits. If draft generation saves two minutes but review adds three, the workflow is slower.
Create a message brief with the account context, approved proof points, offer, desired next step, and prohibited claims. The AI may fill a draft from that brief. It should not invent customer facts, infer sensitive attributes, or choose a recipient without a defined rule.
Use a four-part review: factual accuracy, relevance to the account, clarity of the next step, and compliance with consent and suppression rules. Store rejected reasons so the prompt and source brief improve over time.
The outbound follow-up system shows how to document ownership and escalation. An outbound sequencing VA can prepare drafts and maintain the queue while a sales owner approves send decisions.
Methodology and limits
The report covers a double-anonymous survey conducted in August and September 2025 across 22 countries. The 36% value is an expectation after full implementation. It should be tested against the team's own message quality and approval process.
Frequently asked questions
Should AI send sales emails without review?
Not by default. Use human approval until the audience, claims, consent, and failure paths are documented and tested.
What is a good baseline?
Track approved messages per rep-hour, edit rate, factual-error rate, reply quality, and meetings created. Keep quality metrics beside speed.
Can generic drafts improve productivity?
They can reduce typing but may increase review time and weaken response quality. Personalization should be based on verified account context.
Related Research
Sources
- Salesforce, State of Sales, Seventh Edition (published 2026): https://www.salesforce.com/en-us/wp-content/uploads/sites/4/documents/reports/sales/salesforce-state-of-sales-report-2026.pdf