Example project, figures to be confirmed

WhatsApp support agent for a service business

Two staff answering the same twelve questions all day, and nothing after hours.

Client

A Nairobi service business, 20 staff

Sector

SMEs and startups

Services

AI agents & automation

Routine enquiries

Resolved without staff

Method
Share of inbound messages closed with no human reply, measured over the first month against a manual baseline week.

After hours

Now answered

Method
Enquiries arriving outside business hours that receive a substantive reply within minutes.

Daily log

Of what it could not answer

Method
Every unanswered question recorded and reviewed, which drives what gets documented next.

The story

Context

A service business taking most of its enquiries through WhatsApp, with two staff handling the inbox alongside their other work.

The problem

Two staff spent most of the day answering the same twelve questions on WhatsApp, and after-hours enquiries went unanswered until the next morning, by which time some had gone elsewhere.

What we did

An agent grounded in the company's own documents, with booking lookup, a clean handover to a human whenever it was unsure, and a daily log of everything it could not answer.

Read a month of real messages

Before writing anything, to find the twelve questions that were actually being asked.

Piloted against a baseline

One week measured by hand, then the same week measured with the agent running, so the go/no-go was a number.

Kept a person in the loop

Anything the agent was unsure of hands over cleanly to a named member of staff, with the conversation intact.

The outcome

Most routine enquiries are resolved without staff, after-hours enquiries are answered, and the daily log of unanswered questions has become the list of what to document next.

Quote generation for the three most common job types, with human approval before anything is sent.

Stack

ClaudeWhatsApp Business APIPythonRAG