AI Sales Bot for Lead Qualification in SaaS
A SaaS company was wasting manager time on unqualified leads. We deployed an AI bot that qualifies leads by 7 criteria, books demos, and passes hot leads to CRM with full context.
- +55%demo rate increase
- ×2.4more qualified leads
Industry context
The inbound problem in SaaS is almost never quantity of leads but their composition. Forms are filled by students, competitors, freelancers shopping for a tool, and companies the product does not fit by size or market. A manager cannot tell them apart before the first call, so spends equal time on everyone, which is why a company with a seemingly healthy pipeline suffers from a long cycle. Another trait: in SaaS lead value decays fast. A person who submitted a form and got a reply in ten minutes, and the same person a day later, are two different probabilities, because within a day they have looked at two competitors.
The challenge
Managers talked to everyone, including people who would never buy. Hot leads waited their turn, and some were simply lost between chat and CRM.
Our approach
- 01
Qualification model
We formalized 7 qualification criteria together with the sales team.
- 02
Dialogue & demo booking
The bot runs the conversation, qualifies, and offers a demo slot on the spot.
- 03
CRM handoff
Hot leads flow into CRM with full conversation context.
How the solution works
- 01IntercomA visitor opens the chat
- 02GPT-4oQualifies against 7 criteria
- 03CalendlyBooks a demo
- 04HubSpotReceives the hot lead with a transcript
Intercom is the chat entry point, GPT-4o runs the 7-criteria qualification dialogue, Calendly books the demo, and HubSpot receives the hot lead with transcript and score. Managers only talk to on-target prospects.
Why this stack
Intercom stayed the entry point, because swapping a familiar site widget is unnecessary risk in a project whose value lies elsewhere. GPT-4o runs the qualification dialogue instead of a form with fields deliberately: someone who will not fill in seven fields will happily answer seven questions in conversation, and conversion into a qualified lead comes out higher. Calendly books the demo inside that same dialogue: every extra step between “I am interested” and a confirmed slot costs a share of leads, so booking must happen before the person closes the tab. HubSpot receives the lead with transcript and score: what a manager needs is not the fact of a submission but the reason this lead counts as hot, otherwise they will start the conversation from zero anyway.
Where projects like this usually break
A bot that became a turnstile
The temptation to filter harder is strong, but qualification must not feel like an interrogation. If the bot demands budget and job title before the person understands the product’s value, a share of on-target leads simply leaves. Value first, questions second.
Criteria invented by marketing
A qualification model built without the sales team is almost always wrong: it filters out the people managers actually close and lets through the ones who waste time. Criteria have to come from real won and lost deals, not from an image of the ideal customer.
A hot lead in the general queue
Qualification without changing response speed achieves almost nothing. If the bot marks a lead hot and a manager sees it in the morning, the advantage is gone. The handoff has to come with a notification and a clear rule about who responds right now.
Results
| Before | After |
|---|---|
| Reps talked to everyone | Reps only talk to qualified leads, ×2.4 more of them |
| Hot leads waited in line | The demo is booked right away in Calendly, demo rate +55% |
| Leads got lost between chat and CRM | The lead reaches HubSpot with a transcript and a score |
Qualification + demo booking within 2 weeks.
Who this fits
This kind of bot makes sense when inbound volume is high enough to sort and a visible share is off-target (typically from a few dozen a month). One prerequisite is non-negotiable: you can name the traits of a good customer based on closed deals rather than assumptions. If volume is low and every lead is valuable, qualification is pointless: a human will reply faster. If the product is complex and a committee decides, limit the bot to the first step: gather context and book a conversation without trying to score. And remember that qualification changes not the number of leads but what managers spend time on; if the problem is quantity, a different stage needs fixing.
Frequently asked
Does qualification scare some customers away?
Bad qualification does: the kind that opens with budget questions. The dialogue is built the other way around: the bot first answers the person’s questions and shows whether the product fits, and only along the way establishes what it needs. Then the questions read as part of a consultation.
What happens to off-target leads?
They are not discarded. The bot gives a useful answer, offers materials where relevant, and the lead is stored with the reason recorded. Companies grow and come back a year later on-target, so there is no sense in losing those contacts.
Who defines the qualification criteria?
The sales team together with us, based on closed deals. We analyze what won deals had in common and where lost ones broke down, and that is what shapes the model. The criteria are revisited later, because the market and the product change.
Can we use a CRM other than HubSpot?
Yes. Pipedrive, KeyCRM or an in-house system connect the same way through an API. What matters is not the specific CRM but that the transcript and the reason for the score arrive with the lead: without that a manager starts from zero and the benefit of qualification disappears.