WhatsApp and Web AI Agents
A WhatsApp AI agent is an autonomous agent that answers your clients inside WhatsApp, using your own data, without a person having to look anything up. The same agent can answer on your website, so a client gets one consistent answer in whichever channel they open. E-comienza builds these agents for booking, customer support and financial data querying. Prosperia is the one running in production.
What a WhatsApp AI agent is
A WhatsApp AI agent is a language model agent connected to your systems and reachable through a WhatsApp number. A client sends a message the way they would message a person. The agent understands the request, retrieves the relevant data, and answers.
What makes it an agent rather than a bot is what happens next. It can generate a report, look up a record, compare two periods, or hand the conversation to a human when that is the right move. It is not choosing from a menu of canned replies.
The channel is the point. Your clients already have WhatsApp open. There is no adoption curve, no app to convince anyone to download, and no portal password to reset.
WhatsApp and Web, one agent
We build the agent once and expose it in both places. The client who prefers WhatsApp and the client who prefers a chat window on your site get the same reasoning, the same data and the same answer.
That consistency is worth more than it sounds. Two channels backed by two different scripts is how a business ends up giving two different answers to the same question.
Web chat also covers the visitor who is not yet a client. The same agent can answer product questions on the site and, once someone is a client, answer questions about their account in WhatsApp.
What these agents do
Three kinds of work are in scope today. Booking, where the agent handles scheduling and rescheduling conversations. Customer support, where it answers the repeated questions your team answers every day. And financial data querying, where it reads real records and returns the figure the client asked for.
Financial data querying is the hardest of the three and the one we have in production. It requires the agent to interpret a vague human question, map it to the right period and the right measure, pull the actual number, and present it in a sentence a non-accountant understands.
The common thread is that the answer comes from your system of record, not from the model memory. The model handles language and reasoning. Your data supplies the facts.
Prosperia in production
Prosperia, at prosperiaapp.com, is an autonomous agent on WhatsApp and Web for accounting professionals and their clients. Clients ask natural-language questions about their finances and receive instant reports, with no manual lookup by the firm.
A real exchange: the client asks what last quarter net income was. The agent answers that Q3 net income was $847,320, an increase of 12.4% versus Q2, and offers to generate the full report. The client then asks for a comparison with last year, and the conversation continues.
Notice what the firm did not do. Nobody opened the accounting system, exported a figure, wrote an email or scheduled a call to deliver a number the client could have had immediately.
Why accounting and finance firms
Accounting firms have a structural problem the agent fits exactly. The data exists and is accurate, but it is locked behind a person. A client question that takes thirty seconds to answer still costs a meeting, because the client cannot reach the data and the accountant is the interface.
An agent removes the queue without removing the accountant. Routine lookups are answered instantly. The professional keeps the work that requires judgement, which is the work clients are actually paying for.
The same shape appears in other businesses. Any business whose clients keep asking for information the business already holds is a candidate.
How a deployment works
First we scope the questions. We take the questions your clients actually ask, in their words, and decide which ones the agent will own and which ones always go to a person.
Then we connect the data. The agent needs a reliable path to the records that answer those questions, and it needs to know which client may see which data.
Then the agent goes live in the channel, and we tune it against real conversations. Real users ask things nobody writes into a spec, and the first weeks of real messages are the most useful design input there is.
Grounding and safety
The agent answers from your data. It does not answer financial questions from general knowledge, and when it cannot support an answer it says so rather than producing a plausible number.
That rule matters most precisely where the numbers matter. A wrong figure delivered confidently inside a WhatsApp thread will be acted on. A clear statement that the agent cannot answer sends the client to a person, which is the correct outcome.
Access rules sit alongside grounding. Each client sees their own data and nobody else data, and the agent is built with that boundary in it rather than around it.
Frequently asked questions
What is a WhatsApp AI agent?
A WhatsApp AI agent is an autonomous agent reachable through a WhatsApp number that answers messages using your own data. A client writes a question in plain language and the agent retrieves the relevant record and replies, without a member of your team having to look it up first.
Can a WhatsApp agent answer questions about my clients financial data?
Yes. Prosperia does this in production for accounting professionals and their clients. A client asks about last quarter net income and the agent answers with the figure from the real records, for example Q3 net income of $847,320, up 12.4% versus Q2, and offers the full report.
Do my clients need to install anything?
No. Clients use WhatsApp, which they already have, and message the agent like any other contact. There is no app to download and no portal login to create. If you also want a web chat, the same agent can answer on your website with the same data behind it.
What can a WhatsApp or web agent do besides answering questions?
E-comienza builds agents for booking, customer support and financial data querying. Beyond replying, an agent can generate a report, look up and compare records across periods, handle a scheduling conversation, and hand off to a person when a request needs human judgement.
What stops the agent from inventing an answer?
The agent answers from your system of record rather than from model memory, and it is built to say when it cannot support an answer. That matters most with numbers. An unsupported figure delivered confidently would be acted on, so the agent sends the client to a person instead.
Can one agent work on both WhatsApp and my website?
Yes. E-comienza builds the agent once and exposes it in both channels, so a client gets the same reasoning and the same data whichever one they open. Prosperia runs on WhatsApp and on the web this way. Two channels with two separate scripts is how businesses end up giving two different answers.
