Conversational Commerce Is Not a Chatbot Strategy
Deflection rate is the wrong metric. What matters is whether the conversation ends in a sale.
The metric everyone optimises is the wrong one
Ask a vendor how their chatbot performs and they will quote a deflection rate — the share of conversations resolved without a human. It is a support metric borrowed from a call-centre world, and in commerce it actively misleads.
A conversation that ends with a satisfied, informed customer and no purchase counts as a success by that measure. So does a conversation where a customer gives up.
What to measure instead
- Assisted conversion rate — sessions with a conversation that end in an order
- Average order value, assisted vs. unassisted
- Escalation quality — did the human receive full context, or start from zero?
- Time to first useful answer, not time to first response
Three things separate commerce agents from support bots
1. Catalogue awareness
A support bot reads documentation. A commerce agent needs live stock, current pricing, variant availability, and promotion eligibility. "Is this in stock in size M?" is not answerable from a knowledge base.
2. Memory across channels
A customer who asked about shipping on Facebook and returns on the website is one person. Without a unified profile, they are two strangers.
3. Graceful handoff
The escalation path matters more than the automation. A human picking up a conversation should see everything — the browsing history, the cart, the previous three messages.
Automation that cannot escalate cleanly is worse than no automation. It teaches customers that the channel is a dead end.
A realistic rollout
- Weeks 1–2 — instrument the highest-volume question types
- Weeks 3–6 — automate only the top three, with human fallback always visible
- Weeks 7+ — expand by measured conversion lift, not by coverage percentage
Coverage is vanity. Conversion is revenue.
Read about MS ChatCare AI.