Why WhatsApp is where the measurement breaks
In Malaysia the enquiry usually arrives on WhatsApp, and that is where the trail goes cold. The conversation sits on one person's phone, the follow-up depends on them remembering, and nobody can say later which campaign produced which sale. The channel works; the measurement does not exist.
This is not a WhatsApp problem. It is a boundary problem. The business has a system for orders and a system for accounts, and a gap where the first conversation happens.
What does it mean to automate WhatsApp enquiries?
It means the enquiry is captured, qualified and routed without anyone re-typing it. A message arrives, the useful details are extracted, a record is created in the CRM, the right person is notified, and follow-up happens on a schedule rather than from memory. The conversation still feels human; the administration around it stops being manual.
Note what is not automated: the selling. Judgment, negotiation and anything requiring accountability stay with a person. Automating those is where these projects fail most expensively.
What it changes, with real numbers
In one documented automotive engagement, restructuring campaigns around genuine WhatsApp enquiries rather than raw reach moved cost per conversation from RM14.49 to RM1.88, producing 77 enquiries in the rebuilt campaign.
The interesting part is not the percentage. It is that the figure existed at all. Before the work, cost per conversation was not a number the business could have produced. Being able to see it is what made the campaign fixable.
A comparable engagement in Saudi Arabia produced 125 messaging conversations at SAR11.98 each after an audit-led rebuild of a fragmented ad account.
Where to start if enquiries are unmanaged
The first version should be small enough to finish and specific enough to measure. Most businesses do not need an AI agent on day one; they need the handoff to stop leaking.
- Get every enquiry into one place, even a simple shared record
- Capture the source so campaigns can be judged against outcomes
- Define what qualified means before automating any qualification
- Automate the follow-up that currently depends on memory
- Measure cost per conversation, then per qualified enquiry
When AI actually helps
Add AI once the structure exists, not to create it. Where it works well is reading an unstructured message and extracting the useful fields, drafting a first reply for a person to approve, and flagging which conversations have gone quiet and are worth reopening.
Where it goes wrong is letting it answer questions it cannot be accountable for — pricing edge cases, delivery commitments, anything a customer will hold you to. The test is simple: if a wrong answer would cost you the customer, a person stays in the loop.
What good looks like after a few months
Enquiries arrive in one place regardless of channel. Every one has a source attached, so campaign spend can be judged against enquiries rather than impressions. Follow-up happens whether or not anyone remembers. The cost of a conversation is a number the business checks, not a guess.
None of that requires replacing how the team sells. It requires the administration around selling to stop being invisible.