Conversational commerce and catalogue AI for retail
The short answer
Retail buyers ask in their own words and expect an answer in seconds. Branemind builds WhatsApp commerce agents that answer from your live catalogue and act in your order systems, and search that understands what a product does rather than only what it is called.
Who we usually work with here
- Head of Ecommerce
- Accountable for conversion and basket size.
- CX or Support Head
- Accountable for response time and cost to serve.
- Category or Catalogue lead
- Accountable for findability and data quality.
Where it actually breaks
Search fails the way customers actually ask
A customer describes a problem or a fitment, and keyword search returns nothing because the catalogue is written in part numbers.
The enquiry arrives on WhatsApp and dies there
Answered hours later from a shared handset, with no record in any system and no way to follow up.
The same twenty questions, forever
Order status, returns, availability and fitment consume the people who should be handling the complicated cases.
Constraints that shape every build here
- A price or delivery date must come from the system, never from the model
- Opt-in is required before outbound messaging, and opt-out is permanent
- Catalogue quality bounds answer quality, and retrieval will expose every gap
Where we would start
In order of how often it pays off, not in order of what is most interesting to build.
WhatsApp enquiry to order
Qualification, catalogue answers and order actions in the channel the customer already opened.
WhatsApp Business AgentsCatalogue search and retrieval
Hybrid retrieval with reranking, so a described need finds the right part. Evaluated on your own query logs before launch.
Production AI EngineeringControls we hold ourselves to
- Commitments confirmed from system responses only
- Consent enforced in the send path
- Retrieval quality evaluated against real query logs, not a demo set
- Escalation on repeated retrieval failure rather than a plausible guess
Published work in this industry
Case study
Search that understands a part, not just its name
Forty thousand SKUs, half of them near-identical, and buyers who search by symptom rather than part number. A reranker and a graph fixed what keyword search never could.
- Search-to-cart
- +41%
- Zero-result queries
- −78%
- Support tickets
- −1/3
- 'which part do I need'
Case study
Collections on WhatsApp, run end-to-end by an agent
Recovery calls nobody answers, replaced by a conversation people actually finish. BM Relay negotiates, takes the payment, and knows exactly when to stop.
- Contact rate
- 94%
- vs 31% on voice dialer
- Promise-to-pay kept
- +38%
- Cost per recovery
- −62%
Questions we get asked
Will the agent quote prices?
Only prices your system returns. A model-generated price is a commitment nobody checked, so the agent confirms what the backend says and nothing else.
Our catalogue data is messy. Is that a blocker?
It is the work, not a blocker. Retrieval exposes gaps and contradictions quickly, and the first weeks of most engagements are catalogue repair. That is uncomfortable and it is also the part that produces the improvement.