AI lead qualification from web forms: check the company before anyone calls
A form answer goes to an AI agent that looks the company up, checks whether it fits, answers the visitor on the page and tells sales only about the leads worth a call.
Part of the guide AI employees
AI lead qualification from web forms means that every enquiry is checked before a salesperson touches it: an AI agent reads what the visitor typed, looks the company up, compares it with what a good customer looks like for you, and decides what happens next. The visitor gets a useful answer on the page at once, and your sales team gets only the enquiries worth a call, each with a short note on why.
This is for sales and revenue operations teams whose contact form brings in a mix of real buyers, students, job seekers, suppliers and people who typed a company name that does not exist. It walks through the three blocks that do the work, a made-up example, the questions the agent should ask itself, and where people usually go wrong.
Why a form needs a second look
A form tells you what someone claims. It says the company is called Harbour Bakery Group, that it has about 40 sites and that it is in Toronto. It does not tell you whether that business can be found, whether it is the size it says, whether it is already one of your customers, or whether it sits in a city you serve. A person used to check this by searching, opening a few pages and writing a line in the CRM. That check is what an agent can do in seconds, every time, at two in the morning as easily as at noon.
The point is not to reject people. It is to make sure a strong enquiry is answered first and a weak one still gets a polite, useful reply instead of silence.
The three blocks: a form, an agent and a reply
- Form page starts the flow. You choose up to 12 fields: name, work email, company, website, city (a place field, so it is looked up), what they need (a choice), and a long text box for the message. Each answer runs the flow once, with the answers as one item.
- AI agent reads the answers and decides. Its instructions describe your good customer in plain words. Its tools are the blocks you link to its purple Tools dot, and it calls them only when it needs them.
- Reply answers the visitor on the form's own page with what the agent wrote, for example the next step and when to expect a call.
Alongside the reply, an Email to the sales team carries the qualified leads only, or a row goes into the team's sheet. The form page shows the answer when the run ends within a minute, so keep the agent's look-ups short: a few tool calls, not a research project.
Which tools the agent uses to check a company
- Map business scan, set to look for each place itself, searches online maps for the company by its name near the city typed. It brings back the address, phone, website, category, rating and whether it is open, which tells the agent the business is real and roughly what it is.
- Nearby places counts the other branches or outlets of the same kind around the address, useful when your offer suits multi-site businesses.
- Match with my table looks the company up in your customer list in My data, so an existing customer is passed to account management instead of new business.
- Apollo, on your own Apollo account, adds the company's industry and size from its website when you need firmographics.
In the tool's settings, write {{agent}} where the agent should fill in the value on each call, such as the company name or the place to search around. Everything else stays exactly as you set it, so the agent cannot widen a search or change what a tool does.
A worked example (made up)
Imagine a company in Madrid that sells kitchen management software to restaurant groups with five or more sites in Spain and Portugal. Its website form asks for the name, work email, restaurant group, website, city, number of sites and a message.
On a Tuesday evening three forms arrive. The first names a group in Valencia with eight sites. The agent searches the name on online maps near Valencia, finds a well-rated restaurant with a website matching the one typed, uses Nearby places to see several sites of the same brand in the city, checks the customer table and finds nothing. It writes: Fits: multi-site group, eight sites claimed and several found, not a customer. The visitor reads that a specialist will call within one working day; sales receive the lead with the agent's note and the details it found.
The second is a single cafe in Lisbon. The agent finds it, sees one site, and replies with a link to the self-service plan instead of promising a call. The third names a company that cannot be found anywhere near the city given and a message that asks for a job. The visitor is pointed to the careers page and nothing reaches sales.
Nobody read three forms. The one enquiry that mattered had its call booked before the morning stand-up.
What to put in the agent's instructions
The agent follows its instructions and nothing written in the form, so the instructions are where your judgement lives. Keep them concrete:
- Who you sell to: the kinds of business, the size, the places you serve.
- What counts as proof: found on the map in the city given, a website that matches, a minimum number of sites.
- What to do in each case: qualified, too small, outside your area, existing customer, not a buyer.
- What to write back, in how many words, and what never to promise (prices, dates, discounts).
- Which field to fill: a short
verdictandreasonthat the email to sales can show.
If the reply to visitors must meet rules every time, such as staying under 80 words and never quoting a number the agent did not find, put Check the work between the agent and the Reply. An answer that breaks a rule goes back to be written again before anyone sees it.
Mistakes to avoid
- Too many tool calls. The page waits for the answer. Set Most tool calls low and let the agent stop once it is sure; a run longer than a minute sends the result by email instead.
- Treating not found as fake. A new or very small business may not be on any map yet. Tell the agent to call it unconfirmed and send it on with a lower priority, not to discard it.
- Letting visitors steer what is sent. On a public form, people can type anything. Keep the agent's sending tools behind an Approval if it writes anything beyond the reply on the page.
- Testing only with perfect enquiries. Try a misspelt company, a personal email, a job request and an existing customer before the form goes live, and read how the agent answered each one.
Every run keeps how the agent decided: each tool it called, what it was given and what came back. When sales disagree with a verdict, open the run, press See how it answered and change the instruction that led it there.
Qualifying form enquiries is one part of a wider sales setup with AI employees, from building lead lists to writing first messages; the guide to AI employees for sales prospecting covers the rest.
Questions
Does the visitor see the agent's verdict?
Only what you tell Reply to show. Most teams show the next step in friendly words and keep the verdict and reason for the email to sales.
What if the run takes longer than a minute?
The form page says thank you and the result is emailed instead, so nothing is lost. Fewer tool calls keep most answers on the page.
Can it check whether the company is already a customer?
Yes. Link Match with my table as a tool and point it at your customer list in My data; the agent can look the company up by name, website or phone.
Can the form be put on our own website?
Yes. The form has its own page that you can link to or place on your site, and it takes answers while the flow is live and its Form page is On.
Will it send emails to the visitor on its own?
Only if you give it an email tool. Sends from an agent go out with the run, never while it thinks, and an Approval after the agent holds them until someone on your team agrees.
Templates that do this
Read next
More on ai employees and automation · Every article · Templates