An AI agent for lead generation: when to let it research, and when not to
An AI agent for lead generation plans a search, tries several sources and reports what it found. When that beats a fixed daily flow, and how to keep it honest.
Part of the guide AI employee for sales prospecting
An AI agent for lead generation is worth using when the question is open: "which kinds of business in a city we have never sold in are most likely to need us, and who are twenty of them?" Given a task like that, the agent writes itself a to-do list, searches with the tools you allow (places, job postings, news, a website), notes what it learns, and comes back with a report and a list. When the question is closed, such as "send me new dental practices in my area every morning", a fixed scheduled flow is the better tool: it is cheaper, faster and gives the same kind of answer every day.
This article is about finding leads, not working them. It explains how a research agent goes about the job, compares it with a fixed flow, walks through a made-up example in Seoul, lists the mistakes that send agents in circles, and shows how to set one up in xMap Flows.
How a research agent works through the task
A planning agent does not answer in one go. It breaks the task into steps and works through them, keeping notes between steps:
- It restates the goal and the limits you gave it: the city, the kind of customer, how many leads, what to leave out.
- It plans a few steps: look at which business types are common, check which of them are hiring, look for news of openings or expansion, then pick and rank candidates.
- It works each step with its tools and writes down what it found, adding steps if something unexpected turns up.
- It stops when the plan is done or it reaches the most steps you allowed, and writes a report with the list and how it got there.
That is why it suits exploration: it can follow a lead it did not expect. It is also why it needs limits, or it will keep exploring.
Agent or fixed flow?
| Question | Better tool | Why |
|---|---|---|
| New businesses of a known type, every morning | Fixed flow | Same steps every day; only the results change |
| Which segments to target in a new city | Agent | The steps depend on what it finds |
| Companies hiring a given role, weekly | Fixed flow | A clear search with a clear filter |
| Twenty strong prospects with a reason each, for a market entry | Agent | Needs judgment across several sources |
| Keeping a list current after it is built | Fixed flow | Repetition, with nothing sent twice |
A rule of thumb, not a measurement.
The two work well together. Let the agent do the first exploration once a month, then turn what it learned into a fixed daily flow that keeps the list topped up.
A made-up example: an office supplies firm entering Seoul
This example is made up. A company that supplies offices with stationery, coffee and cleaning products wants to start selling in Seoul. It does not know which kinds of company to start with. It gives an agent the task: "Find 20 companies in Seoul with offices of 50 people or more that are likely to switch office supplier in the next three months. Explain why each one. Leave out government offices."
The agent plans five steps. It searches the business districts for office buildings and co-working spaces, then searches job postings for office managers and facilities staff in Seoul, then checks news feeds the company named for office moves and expansions, then cross-checks the companies that appear in more than one source, and finally ranks them. Its report says that companies hiring an office manager while moving premises came up most often, lists twenty, and gives a sentence of evidence for each. The sales lead then sets up a fixed weekly flow on exactly that signal.
Mistakes that send agents in circles
- A task with no finish line. "Find leads in Seoul" never ends. Say how many, of what kind, and when to stop.
- No limit on steps. Set the most steps it may take. It always reports what it has when it reaches the limit.
- Tools it does not need. Every extra tool is another branch to explore. Give it the three or four that fit the task.
- Trusting the list without the evidence. Ask for a reason per lead and read a sample. Any claim it cannot point to is a claim to check.
- Using an agent for a daily routine. Once you know the signal, a fixed flow does it better.
Setting it up in xMap Flows
In xMap Flows, the agent block has a mode called Work through a big task, made for this kind of research:
- Schedule: once a month, or run it by hand when you start a new market.
- AI agent, How it works set to Work through a big task: the task in plain words, Most steps set to a sensible number such as 12.
- Tools linked to its Tools port: Nearby places for businesses in an area, Job postings for who is hiring, News for openings and moves from feeds you choose, and Browser for a public website that has no export.
- Make a report or Email: the agent's report and list.
What lands in your inbox is a written report with the plan it followed, what it found at each step, and the list of leads with a reason for each. In the run you can open the timeline to see every step and tool call. When you know the signal, build the fixed version: the template New places every morning, only new for openings, or Full outbound engine: find, research, approve, send for a list that goes on to approved outreach. For an agent that helps reps work the leads once they exist, see An AI agent for sales; for the whole picture, the guide to AI employees for sales prospecting.
Questions
What does an AI agent for lead generation do that a search does not?
It plans several searches, uses more than one source, connects what it finds and explains why each lead is on the list.
Is an agent better than a scheduled lead list?
Only for open questions such as entering a new market. For a daily list of a known kind of business, a fixed flow is cheaper and more predictable.
How do I stop it running too long?
Set Most steps on the agent. It always writes a report with what it has when it reaches the limit or runs out of the credits held for it.
Can I see how it reached its list?
Yes. The run keeps a timeline of the plan, each step, each tool call and the report, so you can check its evidence.