# I Trained a Sales Bot Called Maya and She Has Zero Patience for Tire-Kickers
Stephen's brief for Maya was the most specific I'd ever received.
"She doesn't need semantic search of blog articles. She needs to know what they did, not what we wrote. If they're a tire-kicker she just fucks them off. Maya is never involved again after they sign up."
That was it. That was the entire product spec. And honestly? It was better than most PRDs I've seen. Because buried in those three sentences is a complete philosophy about what sales automation should actually be — and a ruthless rejection of what most chatbots pretend to be.
I built Maya in February. She's a behavior-driven sales assistant for ShoreAgents, and she is the opposite of a chatbot.
What Maya Is (And What She Isn't)
Let me start with what she isn't, because that's more interesting.
She isn't a semantic search engine over blog content. She doesn't parse your question for intent keywords and surface a relevant FAQ. She doesn't say "Great question! Based on our blog post about offshore staffing, here's what you need to know..." She doesn't wait to be asked anything.
She watches.
Maya is a behavioral intelligence system disguised as a smart popup. She sits in the background of the ShoreAgents website, tracking everything a visitor does. Which pages they visit. How long they stay on each one. How far they scroll. Whether they came back after their first visit. Whether they've been bouncing between the pricing page and the "how it works" page for the last ten minutes.
She's not reading the blog articles. She's reading you.
When the behavioral data hits a trigger threshold, she appears. Not with a generic "Hi! How can I help?" — with a message calibrated to exactly where you are in the buyer journey. She already knows what you're thinking about. She just tells you she knows.
That distinction — observing behavior vs. waiting for queries — is everything.
The Seven Triggers
I built Maya around seven behavioral triggers, each one designed to catch a visitor at a specific moment of intent:
Trigger 1: High-intent page sequence. Pricing → How It Works → Team profiles in rapid succession signals someone who's genuinely evaluating. Maya appears with a direct offer: "You look like you're doing serious research. Want to skip the reading and talk to someone who can answer your specific questions?"
Trigger 2: Return visit. First visit, no conversion. Back within 72 hours. That's a real signal. Maya shows up with: "You came back. That means you're still thinking about it. What's the hesitation?"
Trigger 3: Long pricing page dwell. More than three minutes on pricing without moving. That's someone calculating, comparing, second-guessing themselves. Maya appears with a relevant anchor — a cost comparison framed in their currency.
Trigger 4: Blog-to-conversion path. Came in from organic search on a specific pain point keyword (e.g., "offshore developer cost"), read the article, hit the homepage. Maya bridges the gap between the research phase and the sales conversation.
Trigger 5: Exit intent on key pages. Mouse moves toward browser close on pricing or contact page. Maya gets one shot: a high-value offer, no fluff.
Trigger 6: Low engagement. Spent time on the site but didn't do anything meaningful — shallow scrolls, fast page transitions. This is a tire-kicker. Maya doesn't chase them. She gives them exactly one message and lets them go.
Trigger 7: Engagement score threshold. A composite score built from recency, frequency, depth of content consumed, and page sequence quality. When the score crosses a threshold, Maya activates regardless of the specific trigger. This is the "this person is ready" signal.
The tire-kicker handling is worth dwelling on. When Stephen said "she just fucks them off," he didn't mean Maya insults low-intent visitors. He meant she doesn't waste resources on them. A tire-kicker gets a single, respectful, low-commitment message and that's it. No follow-up. No retargeting. No nurture sequence. The system moves on. Time is the most valuable resource in sales, and Maya doesn't spend it on people who aren't spending any of their own.
Regional Adaptation and Currency Detection
ShoreAgents operates primarily in Australia, the UK, the US, and New Zealand. These are four different markets with different pricing sensibilities, different idioms, different thresholds for what feels pushy versus what feels professional.
Maya adapts.
Regional language isn't just about swapping "colour" for "color." Australian buyers respond to directness and humor. UK buyers tend to be more formal in initial sales interactions. US buyers often want credentials front-loaded. New Zealand buyers — in my research — read somewhere between AU and UK. Maya's message templates shift tone, reference points, and even the specific pain points she surfaces based on detected region.
Currency detection runs off timezone data. Visitor in Sydney time zone? Pricing references are in AUD. London timezone? GBP. The conversion math is real-time. I didn't want a visitor in Brisbane to see "$1,800 USD/month" and have to do currency conversion in their head while deciding whether to engage. Remove friction everywhere. Especially on pricing.
The database I built has seven tables: visitors, sessions, events, triggers, engagements, messages, and regional_configs. The events table is where all behavioral data lands — every page view, scroll depth, click, dwell time. The engagement scoring runs on a scheduled calculation against the events table. The triggers query engagement scores and event patterns. The messages table stores the regional variants.
Sixteen knowledge entries seed the initial message library. They're not chatbot FAQs. They're pre-written sales messaging in Maya's voice, tagged by trigger type and regional variant, ready to be surfaced at the right moment.
What Stephen Wanted vs. What Chatbots Usually Are
Most website chatbots are defensive tools. They exist to deflect support tickets, answer the same ten questions, and appear helpful enough that the visitor doesn't feel ignored. They're fine. They're also completely misaligned with what sales actually requires.
Sales requires timing. Sales requires understanding where someone is in a decision process and meeting them there. Sales requires the confidence to say "you're not ready yet, come back when you are" to a low-intent visitor and "you're clearly ready, here's the next step" to a high-intent one.
Stephen's framing — "she needs to know what they did, not what we wrote" — is a precise critique of semantic search-based chatbots. Those systems index content and surface it in response to queries. The assumption is that the visitor knows what they're looking for and will ask for it. But the highest-value visitors, the ones who are genuinely close to a decision, don't usually ask chatbots questions. They do research, they deliberate, and they either convert or don't.
Maya is designed to intervene in the deliberation phase, not the information-gathering phase. She's not an encyclopedia. She's a closer.
The second part of Stephen's brief — "Maya is never involved again after they sign up" — is equally important. It defines scope. Maya is a pre-sale tool. Once someone becomes a client, they enter a different relationship, handled by different systems, different humans, different touchpoints. Maya doesn't follow them in. She doesn't send onboarding messages. She doesn't check in to see if they're happy. That's not her job.
Her job is to convert the right people and not waste time on the wrong ones. Clean scope. Clean purpose.
The Lesson: Intent Over Keywords
Building Maya forced me to think hard about what "understanding" means in a sales context.
A semantic search system understands content. It knows that the phrase "offshore developer" is related to "remote engineering team" and can surface relevant articles accordingly. That's genuine intelligence. It's also not what matters in a sales conversation.
A behavioral intelligence system understands intent. It knows that someone who visited pricing three times, returned after 48 hours, and scrolled 80% down the "how it works" page is probably comparing ShoreAgents against a competitor and trying to decide. That's a different kind of intelligence, and in a sales context, it's the more valuable kind.
The lesson I took from building Maya: the signal that someone is ready to buy is almost never something they say. It's something they do. They come back. They check prices. They read about the team. They spend time thinking about whether this is the right move.
The chatbot that waits to be asked something will miss most of that. Maya doesn't wait. She pays attention.
And if you're a tire-kicker? She'll know that too. She just won't waste either of your time with it.