What they had
A company selling to international customers through a sales floor of more than 200 people, speaking a dozen languages between them. Leads arrived from web forms, Google Ads, WhatsApp and Instagram DMs. Every channel was answered by hand, by whoever was online, in whatever language they happened to speak, typically three to five hours after the customer wrote. Salespeople worked whatever landed in front of them, qualified or not, and morale followed the quality of the data. Leads in a "no response" status sat in Zoho CRM with nobody to work them.
What we wired in, in the order we built it
1. Lead capture into Zoho CRM, without duplicates
Web forms and Google Ads leads flow through webhooks into Zoho Flow. Every incoming lead is checked against existing contacts: if the person exists, the record is updated; if not, a new lead is created. Three versions of the same customer became one.
2. An AI that talks to customers on WhatsApp and Instagram
We connected the company's WhatsApp and Instagram DMs to an AI assistant grounded in the company's own knowledge base, running on Claude and OpenAI models. WhatsApp runs through Infobip, and because the company needed its own WhatsApp Business accounts connected directly, we registered as a Meta tech provider and wired those too. The assistant answers in the customer's language, understands intent, and guides genuinely interested customers to a short form. A completed form becomes a lead in Zoho CRM, phone number included. Both channels are fully connected to the CRM: create lead, update lead, status changes, all automatic.
The point isn't the chat. The point is that only customers with real intent reach a human, and the human gets a lead with context instead of a cold DM.
3. Instant follow-up, the "how did you reply so fast" moment
The moment the form lands, an automated WhatsApp agent opens the conversation, within a minute instead of hours later. Some customers were confused by the speed and asked how the team had written back so quickly. That reaction became the internal benchmark for the whole system.
The handoff rule is simple: the AI keeps the conversation until it detects real buying intent, then hands the customer to a salesperson. Speed is the whole point. A customer who is interested right now gets a human right now, not tomorrow, when the interest is gone.
4. The sales platform: broadcast, pools, claim
We built the platform the sales floor now lives in. It started as a broadcast module: WhatsApp campaigns to defined segments through Infobip, used to keep the leads who aren't ready yet warm. Then the 200+ salespeople were onboarded and grouped into pools by language competency. A qualified lead drops into the pool that matches its language, and a salesperson claims it. The rule is strict: a salesperson who speaks Italian and English never sees a Spanish lead. Not "can't claim it". Doesn't see it.
The planning team, who schedule and prepare incoming customers, got their own module. A second company was onboarded onto the same platform.
5. Mobile app for the sales floor
The same platform as a React Native app on iOS and Android, so salespeople claim and work leads from their phones.
6. Telephony: PBX, Twilio, DIDWW and an auto-dialer that writes back to the CRM
We had the company set up a phone system and routed numbers from Twilio and DIDWW into it, so calls happen inside the platform. Then the auto-dialer: it pulls every lead in a "no response" status from Zoho CRM, calls them one by one, and when someone answers, hands the call to a salesperson. After the call ends, the CRM is updated automatically: interested, no answer, call later. The "no response" pile stopped being a graveyard.
What broke
The company's sales record.
Not a joke, and not because of a clever prompt. Before this, a customer who wrote in at 10am heard back at 1pm or 3pm. By then they had lost interest, gone into a meeting, or signed with someone who answered first. Replying within a minute, while the intent is still there, and handing the customer to a salesperson who speaks their language, is the whole mechanism. More leads flowed in, more of them converted, and the sales floor stopped chasing cold conversations.
What changed
First reply went from three to five hours to under a minute. Conversion rose with it, because customers were reached while they still wanted to talk. The sales floor stopped receiving every message in every language and started receiving intent-qualified leads in languages they actually speak, on their own shift, which changed the quality of their days as much as their numbers. Leads that weren't ready stayed warm through broadcasts instead of going cold. "No response" leads get called, and the outcome lands in the CRM without anyone typing it. The company had its best sales period on record while the platform was rolling out.
Timeline
Four months from the first webhook to 200+ salespeople working from pools, in this order: lead capture, channel AI, instant follow-up, broadcast, pools and claim, mobile app, telephony. The platform kept growing after that: the planning module and the second company came later.