An AI sales agent that closes,
not just answers, across the channels you sell on
A generic chatbot answers questions. A sales agent is supposed to close. We build the second kind. An agent that follows your actual playbook. One that tracks objections it has not seen before, and knows exactly when to stop talking and hand off to a person.
A closer, not a FAQ bot
An AI sales agent is software that carries a sales conversation from the first message to a closed order or a qualified handoff. It works on the channel a buyer actually picked. It needs a real playbook behind it: the scenarios, objections and tone a good human seller would use, not a generic “how can I help you” bot. It pays off for teams selling through Telegram, WhatsApp, LINE, Instagram or a website chat widget, at enough volume that a person cannot answer every message within minutes. If your sales cycle depends on judgment calls a script cannot make, this will not replace your best closer. It makes sure nobody with a real question waits an hour for a reply.
A playbook, a closed toolset, and per-channel limits
The playbook comes first. We sit with your sales process and write down the scenarios, the objections buyers actually raise, and the tone that fits your brand. We also mark the exact points where the agent must stop and hand off. The agent itself runs on Claude or GPT with a closed set of tools. It can look up a product, check stock, create an order, or escalate. Nothing outside that list. It cannot wander into promises you did not approve. Each channel gets its own rate limits, tuned to that platform’s published limits and policies. A bot that replies too fast on WhatsApp risks the account, not just the conversation. Every sale, every escalation and every unanswered question lands in a log your team can read.
One channel first, proven, then the rest
We start with your current sales material: call recordings, chat transcripts, FAQ documents, whatever exists. We turn it into a structured playbook with scenarios and objection responses. That playbook becomes the agent’s instructions, tested against real conversation transcripts before it goes live on a single channel first. Once the first channel is stable, we add the others, reusing the same brain and playbook, and tune rate limits and message pacing per platform. CRM or Telegram order alerts go in from day one, so a closed sale is never only visible inside the chat log. We run a tuning period after launch, reading transcripts weekly and refining objection responses that are not landing.
What breaks a messaging account
The biggest real risk is a playbook gap discovered live: a buyer asks something the script never covered, and the agent either stalls or improvises badly. Weekly transcript review in the first month catches this before it becomes a pattern. The other risk is channel account health. Messaging platforms suspend accounts that message too fast or too many strangers. Rate limits are not optional polish. They are what keeps the account alive past week one. Budget for light ongoing tuning even after the formal 30 days end, since buyer language and objections shift as your product or market does.
Timeline and price
| Option | Price | What it covers | Timeline |
|---|---|---|---|
| MVP | from $2,500 | One channel, playbook from your existing material, CRM alert on sale | 3 to 4 weeks |
| Production | from $6,000 | Multi-channel, per-platform rate limits, upsells and follow-ups, analytics dashboard | 6 to 8 weeks |
| Full control (handover-ready) | from $7,000 | Everything in Production, plus a full handover package: architecture docs, test suite, admin access audit, and a walkthrough so your own team or another vendor can run it without us | 8 to 10 weeks |
Running cost on top of the build is usually $20 to $80 a month in model calls and messaging fees, depending on conversation volume.
What stays yours
You own the playbook document, the full source code, and the channel accounts the agent runs on. Nothing is locked inside a third-party SaaS subscription. The agent is your code, running on infrastructure in your name. Documentation lets any developer, or a person with an AI assistant, pick it up and keep extending it.
Related
See the development service page for how we scope and build custom software. See the AI support agent product for the handling side once a lead stops being a sale and starts needing help. For the orchestration layer when one agent is not enough, see the AI agent orchestration platform. For the underlying runtime architecture, see AI agent runtime if your team wants the component-level build. Real builds: the seven-channel AI sales agent case study and the B2B certification sales agent case study. Ready to see what a playbook for your product would look like? Get in touch and bring whatever sales material you already have.
FAQ
How much does an AI sales agent cost?
From $2,500 for one channel with a playbook built from your existing sales material. A multi-channel agent with CRM integration and per-platform rate limits runs $6,000 to $7,000. The exact price depends on how many channels and how complex the catalogue is.
How long does it take to launch?
Three to four weeks for a single-channel agent once the playbook is written. Adding channels after the first one is faster, usually one to two weeks each, since the brain and the playbook are already built.
What is the stack?
Python and FastAPI for the backend. Claude or GPT as the model. PostgreSQL for conversation and order state. Each channel's native API: Telegram Bot API, WhatsApp Business API, LINE Messaging API, or a website widget.
Who owns the agent and the playbook?
You. The playbook document, the code, and the account are yours. If you want us to keep extending it, changes are billed per task, from $150, with no monthly fee. Otherwise the handover package includes everything a developer needs to run it alone.
What happens when the agent does not know an answer?
It hands off to a human instead of guessing. Every unanswered question gets logged, and we review that log with you to decide whether it becomes a new scripted answer or stays a human-only case.