Quick answer
An AI lead generation workflow has seven stages: ICP, account sourcing, enrichment, fit scoring, outreach drafts, send and follow-up, and handoff with measurement. Let AI do the research and drafting, and keep a named person on three approvals: the ICP and sources, every message's facts, and what counts as a qualified lead.
This guide is for: sales, marketing and revenue operations teams at B2B companies that want AI to handle research and drafting without losing control of who is contacted and what is said.
Lead generation with AI is a pipeline with owners, not a tool
Most pages that rank for this topic list tools. The harder question for a team is the order of work and who is accountable at each step.
Where AI helps and where a person decides
AI is useful where the work is repetitive and can be checked in a sample: finding accounts that match a profile, attaching company facts with their sources, scoring against written criteria and drafting a first message. A person should decide where a mistake is expensive or hard to undo: which accounts are in scope, which claims may appear in a message, and whether a reply is a real buying signal.
Write the facts down before the message
A fluent draft can still say something false about a company. Require every sentence about the prospect to trace to a source field, and treat a field marked unknown as unusable. The reviewer then checks facts, which is faster and more reliable than judging tone.
The seven stages
| Stage | What AI does | What a person approves | Output |
|---|---|---|---|
| 1. ICP and exclusions | Drafts the ideal-customer profile from won deals and the exclusion list (competitors, existing customers, regions you do not sell to). | Approves the ICP and every exclusion before any account is sourced. | A written ICP with a version and an owner |
| 2. Account sourcing | Finds accounts that match the ICP from approved sources: your CRM, your own site traffic, public company pages, licensed data. | Approves the sources and the daily volume cap. | A candidate list with a source for every row |
| 3. Enrichment and verification | Adds firmographic and role data, checks that each fact has a source, flags what it could not verify. | Reviews a sample and every flagged row; nothing unverified is used in a message. | Accounts with a source and a verified-or-unknown flag per field |
| 4. Fit scoring | Scores each account against the ICP and writes the reasons next to the score. | Reviews the top and the borderline accounts and can overrule a score. | A ranked list with reasons a person can read |
| 5. Outreach drafts | Drafts a message from verified facts only: who the account is, why it fits, one relevant point. | Reads and approves every message, or a template and its allowed fields. | Approved drafts, each tied to its facts |
| 6. Send and follow-up | Queues approved messages, applies the suppression list, schedules follow-ups, and routes replies. | Owns the send, the opt-out handling and the reply to anything but a clear yes or no. | A send log with opt-outs and replies |
| 7. Handoff and measurement | Hands a qualified reply to the right seller with the context and compares sources by meetings, not by volume. | Decides what counts as qualified and which sources stay. | A meeting record linked to its source and stage |
Three approval gates
You do not need a person at every step. You need one at the steps where an error leaves your building or enters your pipeline.
Worked trace: four days from ICP to a handed-off reply
This hypothetical trace shows the stages and gates working together. The numbers are illustrative, not a customer result or a benchmark.
Compliance checks that belong in the workflow
Outreach rules differ by country and by who the recipient is, so treat the points below as checks to build in, not as legal advice. Ask your counsel about your own markets.
Email to people in the United States
The FTC's CAN-SPAM guide sets eight requirements for commercial email, including a truthful header, a subject line that matches the content, a clear identification that the message is an advertisement, a valid postal address, and an opt-out that is honored within 10 business days. It also says that hiring another company to send for you does not move the responsibility away from you, and it states a penalty of up to $53,088 for each violating email. In a workflow this becomes a send gate: no message leaves without the opt-out and the address, and every opt-out lands on the suppression list.
Sending to Gmail
Google's sender guidelines separate two groups. Every sender who emails personal Gmail accounts needs SPF or DKIM authentication, valid reverse DNS records for its sending domain or IP, a TLS connection, correctly formatted messages, and a spam rate kept below 0.30% in Postmaster Tools. Senders of more than 5,000 messages a day to Gmail accounts have additional requirements: SPF and DKIM and a DMARC record, alignment of the From domain, and one-click unsubscribe for marketing messages with a clearly visible unsubscribe link in the body. Google suggests aiming below 0.10% for the spam rate. The workflow should treat the baseline as mandatory for any volume and the bulk list as the target to meet before you scale.
Business contacts in the United Kingdom
The UK Information Commissioner's Office separates corporate subscribers, which are companies and similar bodies, from sole traders and some partnerships, which get stronger protection. A named person at a company can still object to direct marketing under UK data protection law, and the ICO advises keeping a suppression record rather than deleting objectors, so you can screen later lists against it. The ICO says this guidance is under review, so check the current version before you rely on it.
What to measure
| Measure | Why it matters |
|---|---|
| Meetings per approved account, by source | A source that fills the list but produces no meetings is cost, not coverage. |
| Reviewer overrule rate | How often a person changes a score or rejects a draft. A rate near zero can mean the reviewer is not looking. |
| Share of fields marked unknown | A rising share means a source is degrading before it shows up in replies. |
| Opt-outs and complaints per 1,000 messages sent | Watch the trend per source and per template, not only the total. |
| Time from reply to a named owner | A qualified reply that waits for assignment is a lead lost to process. |
This guide gives no benchmark for any of these. Your ICP, offer and data sources set your own baseline, so record it for four weeks before you compare.
How OpenMax supports this workflow
An AI BD employee with a human in the loop
OpenMax's BD use case shows an AI employee that researches prospects from company data, CRM and email, scores fit and drafts outreach, with the output reviewed by a person before anything is sent. See the BD use case for the sample run, and the guides on lead scoring, account research and lead routing for the stages around it.
Limits and human-review boundaries
A workflow organizes the work; it does not make a contact lawful, a fact true or a prospect interested.
- Do not message anyone from a field you could not verify, and do not let a score replace a person's judgment on a borderline account.
- Do not infer sensitive traits or personal details from a name, a photo or a location.
- Do not scrape or combine data in a way the source's terms or the law do not allow; ask for the source's terms before you add it.
- Do not count a sent message as progress. Count replies, opt-outs and meetings.
- Test each change on a small batch, keep the suppression list in one place and record who approved what.
Frequently asked questions
What is an AI lead generation workflow?
A sequence of defined stages in which AI does research, enrichment, scoring and drafting, and named people approve the decisions that carry risk: who is contacted, what is said and what counts as a lead.
Can AI send outreach without a person reading it?
It can be configured to, but the law and the sender rules apply to the sender, not to the tool. For a new workflow keep a person on every message or on an approved template and its allowed fields, and widen only after the reviewer overrule rate and the complaint rate are known.
How is this different from buying a lead list?
A list is an input. The workflow adds the ICP, source tracking, verification, scoring with reasons, approval and measurement around it, so that you can see which source produces meetings.
Which stage should a small team automate first?
Enrichment and scoring with reasons. They are the most repetitive, the easiest to review in a sample, and they do not contact anyone.
Does this guarantee more meetings?
No. It is a structure for doing the work with control. Results depend on the ICP, the data sources, the offer and the people reviewing.
Sources, editorial method, and limitations
OpenMax editors read the three public pages below on October 15, 2026 and summarized only what they state. The seven-stage workflow and the three approval gates are an original synthesis, not a vendor's process. No customer result, response rate or benchmark is claimed.
- FTC — CAN-SPAM Act: A Compliance Guide for Business — eight requirements for commercial email, including accurate headers, a physical address, an easy opt-out honored within 10 business days, and responsibility for third parties who send for you; the page states a penalty of up to $53,088 per violating email.
- Google Workspace Admin Help — Email sender guidelines — a baseline for every sender to personal Gmail accounts (SPF or DKIM, reverse DNS, TLS, spam rate below 0.30%) and additional requirements for senders of more than 5,000 messages a day (SPF, DKIM and DMARC, one-click unsubscribe).
- UK ICO — Business-to-business marketing — corporate subscribers versus sole traders and some partnerships, consent and the soft opt-in, the right to object to direct marketing, and keeping a suppression record.