The seven workflows
Build a list against your ICP
Describe your ideal customer in plain words; an agent finds people and companies that match and explains why each fits. You review the reasons, not just the names.
Research every account
A one-page brief per lead: company, people, recent news and an angle. What used to take twenty tabs takes seconds.
Watch for buying signals
New roles, hiring, launches and posts about the problem, flagged as they happen. See 15 free buying signals.
Score inbound leads
Every form fill and inbound email checked against your ICP, with the hot ones briefed to you first.
Write personalised first lines
A specific opening line for each person based on the research, in your voice. Check a sample before you scale it up.
Draft follow-ups that add something
Follow-ups that bring a new angle or resource instead of "just bumping this".
Book the call
When someone says yes, offer real availability, book it and send reminders.
Describing your ICP to an agent
Agents find better leads when the description is specific. Compare:
| Vague | Specific |
|---|---|
| "Small businesses that need marketing help" | "Agencies of 10 to 50 people in the UK and US that run client onboarding, led by a founder or managing director, hiring account managers this year" |
| "Tech companies" | "B2B SaaS companies with 20 to 100 staff, a sales team of fewer than five, and a head of operations" |
Add who isn't a fit, too: "not recruiters, not enterprise, not agencies under five people".
What not to automate
- Deciding your ICP and offer. AI can suggest; the decision is yours.
- Sending without review. A wrong detail stated confidently costs more than the time saved.
- Conversations that need judgement. Objections, pricing questions and calls.
- Mass scraping. Bulk-copying profiles breaks platform rules and gets accounts restricted.
A simple weekly setup
| When | AI does | You do |
|---|---|---|
| Monday | Builds the week's list and briefs | Removes poor fits (10 minutes) |
| Daily | Scores inbound, drafts outreach and follow-ups | Approves or edits (15 minutes) |
| On a yes | Offers times and books the call | Takes the call |
| Friday | Summarises replies and booked calls | Adjusts the ICP and message (20 minutes) |
A worked example
An agency-software founder describes the ICP above on Monday. The agent returns 40 agencies with a reason for each, such as a recent hiring post or a founder post about onboarding. The founder removes six and sets a daily limit. The agent writes a LinkedIn note and an email for each, opening with that agency's reason, and follows up in the same thread. At the end of the week the founder looks at which reasons got replies and leans next week's list towards them.
Tools for each workflow
You can assemble these from separate tools (a database for lists, a research assistant, a sequencer, a scheduler) or use one agent that does them together. Separate tools give more control; one agent saves the copying between them. Either way, make sure replies and booked calls land in one CRM. See the best AI SDR tools for small business for options.
How to know it's working
- Fewer poor-fit leads on the weekly list.
- Approved drafts need fewer edits each week.
- Inbound leads get a reply within the hour.
- Booked calls per 100 people contacted rise over a month or two.
Frequently asked questions
How can I use AI for lead generation?
Use it to build lists against your ICP, research accounts, spot buying signals, score inbound leads, write personalised first lines, draft follow-ups and book calls, with a person setting the limits and checking the results.
Is AI lead generation worth it?
For small teams it usually saves hours a week on research and follow-up. It's worth it when the leads it finds turn into booked calls, so measure that.
Can AI find leads for me?
Yes. Give it a clear description of your ideal customer and it can find matching people and companies and explain why each fits.
What shouldn't I automate in lead generation?
Choosing your ICP and offer, sending without review, and conversations that need judgement.