ATPL

AI · Build with AI

Conversations handled at any hour, handed over when needed.

For businesses answering the same questions on WhatsApp, the website and Instagram all day: a chatbot takes the first reply, qualifies the enquiry, and passes the hard ones to a person with context.

Describe the problemAll AI capabilities

Same questions

The same ten questions every day

Support and sales conversations handled automatically, with a clean hand-over.

Describe the problem

  • Hours

    Enquiries arrive at ten at night on WhatsApp and the sale is lost by morning.

  • Repeat

    Your support team spends the day on order status, timings and price lists instead of real problems.

  • Handoff

    The last chatbot you tried trapped customers in menus and they typed "agent" to escape.

Across channels

One bot that hands over cleanly

  • 01

    One bot across channels

    The same conversation logic on WhatsApp Business, the website widget and Instagram, with each channel’s own rules respected.

  • 02

    Answers from your systems

    Order status from the ERP, table availability from the booking system, stock from the store: live lookups, not canned text.

  • 03

    Hand-over with context

    When the bot cannot help, a person gets the whole transcript, the customer’s details and the bot’s own summary of what is needed.

  • 04

    Lead capture into your CRM

    Qualified enquiries land in the CRM or a sheet with the fields filled, and the bot books the call if a calendar is connected.

  • 05

    Tone, limits and refusals

    What the bot may promise, what it must never say, and the languages it answers in are written down and tested before launch.

Bot or human

Where it fits, where it doesn’t

Saying when this is the wrong tool is the fastest way to be believed about when it is the right one.

Fits when

  • High volumes of similar questions on WhatsApp, the website or social channels.
  • Businesses with systems the bot can look things up in: orders, bookings, stock, prices.
  • Teams willing to keep a person available for the hand-over.

Doesn’t fit when

  • Complaints, refunds and anything emotional. The bot should recognise these and hand over at once.
  • Businesses with no written answers; a bot cannot invent your policy and should not try.
  • Low volumes where one person on WhatsApp already answers inside ten minutes.

Your data

Conversations run on your WhatsApp Business account and your website; transcripts are stored in your database, not the model provider’s. Each reply sends the model only the current conversation and the lookups it needs, under terms that exclude training. Customers are told they are talking to a bot and how to reach a person.

First reply draft

A visitor question becomes a reply

The first useful answer, with the hand-off fields ready if a person has to take over.

What a client typed

We run three restaurants and get about a hundred WhatsApp messages a day asking for the menu, whether we deliver to their area, table bookings for the weekend and where their order is, and one person answers all of it from her phone.

Brief ready

What it becomes

Problem
Repetitive WhatsApp enquiries about menu, delivery area, bookings and order status handled by one person.
Scale
Three outlets; around a hundred conversations a day, peaking at weekends.
Systems
WhatsApp Business, the POS, the table booking sheet, the delivery aggregator panels.
First step
A WhatsApp bot answering menu, delivery area and order status from the POS; bookings and complaints hand over to staff.

How it ships

Channel, then CRM

  1. bot/01

    Understand

    We read a month of your real conversations and group them: what the bot should answer, what it should hand over, what it should never touch.

  2. bot/02

    Design

    We script the hand-over and the refusals before the happy path, and choose the model last, after the channel limits are known.

  3. bot/03

    Build

    The bot runs on a test number with your team posing as customers, and every wrong answer becomes a test case.

  4. bot/04

    Stay

    Hand-over rate and unanswered questions are reviewed monthly, and the bot’s knowledge is updated when your prices or policies change.

How we work

ATPL builds chatbots on the same method as its restaurant and commerce systems: live lookups, logged conversations, a person one tap away. ATPL will confirm which deployments may be named.

    Chosen for the job

    Node.js · Next.js · Python

    Decide the shape

    Before you start

    Plain answers on scope, data and what a first proof usually takes.

    An AI chatbot for one channel such as WhatsApp is typically a three to five week build, plus monthly costs for the WhatsApp Business API, model calls and hosting. ATPL scopes the first version to the questions that make up most of the volume, which keeps both the build and the running cost small.

    Yes. ATPL builds chatbots on the WhatsApp Business API, so the bot answers from your verified business number, uses approved message templates where WhatsApp requires them, and hands over to a human agent inside the same conversation.

    It says so and hands over to a person, with the transcript and a short summary attached, rather than guessing. ATPL designs the hand-over before the rest of the bot, because how a chatbot fails decides whether customers trust it.

    Rule-based bots are better for short fixed flows like OTP or order tracking; AI chatbots are better when customers type freely in mixed Hindi and English. ATPL usually combines both: fixed flows for lookups and payments, a language model for understanding and replying.

    Continue to