Every recommendation surfaced by KPAI Merchandising arrives with the rationale visible — for the operator, for the customer, and for the next decision the customer makes. The page below walks the substrate, the agents that reason on top of it, and the surfaces that meet the customer.
The Moment
A pilot PDP. Customer types a project. Bundle returns with rationale.
Category · Sub-category
Product Name
Product description lives here — material, build, lead time. Ships in 7–10 business days. Pricing visible at signed-in checkout.
$3,840
List · Member −15%
Add to Project
Save
KPAI · Project Assistant
v1 · Brain
KPAI
What are you working on? I can help you build a set around this product while you keep browsing.
Customer
Putting together a set — two anchor pieces, a complement, and an accent.
KPAI
Five pieces that work together below — each with the reason it's in the set.
Product 01 × 2
The pair you're already on. Same family as the rest — visually consistent.
Product 02
Same family as Product 01. Scale fits the anchor.
Product 03
Adds the finishing element. Reads warmer than the obvious alternative.
Product 04
Pulls the palette together. Ties the accent back to the anchor.
Knows what people are searching and asking — keyword volume, AEO answer rates, rising queries.
Reads: SEMrush · GSC*
All three reads run daily. Every claim links to its source.
Run · Boost & Bury
Collection: Category · sub-category
runId · brn_5f8a2c
01
Inventory
step 01 · 1.2s
In this collection, 47 SKUs are active. 8 are below safety stock; 3 of those have a restock ETA past 6 weeks. 4 are dead stock — 0 sold in 90 days, $84k tied up.
02
Performance
step 02 · 0.9s
Of the 8 low-stock SKUs, 12 are converting at 2.1× the collection average. Repeat orders for these tend to come in pairs — AOV impact when bundled is +38%.
03
Search Demand
step 03 · 0.7s
Of those 12, four match queries trending up +38% WoW (seasonal + use-case qualifiers). Three others have flat demand. Five are off-trend.
→
Decision
synthesized · 0.4s
Boost 4 SKUs (high-conversion, on-trend, in stock). Watch 8. Bury 3 on restock-risk. Bundle the trending pair on PDP. Push to Shopify on operator approval.
Every step is auditable. Every claim links to its source data.
See raw run →
Act 03 — Surfaces
Where the thinking meets the customer.
Three customer-facing surfaces. The decision layer above writes through to all three. Each carries the rationale forward — visible, auditable, useful.
Surface 01
Live
PLP Re-rank
Collection pages reorder by what the agents decided — not alphabetical, not last-published. Each rank carries provenance the operator can audit before it ships to Shopify.
↑1
Product 01
on-trend · 2.1× conv
↑2
Product 02
paired AOV +38%
↓3
Product 03
restock 8wk
/merchandising/rerank
Open →
Surface 02
Live
Bundles by Intent
Customer states a project. The Brain assembles a 3–7 piece set from the catalog with rationale per item. The operator can publish as a special collection or push back on any component.
Intent
"A project the customer states in plain language."
5-piece bundle · rationale per item
/merchandising/bundling
Open →
Surface 03
Designed
Project Assistant
A chat surface embedded on the PDP. Asks "what's your project?" while the customer keeps browsing, then returns a curated set with the reason behind every choice. The Brain is search-ready today; the chat layer ships next.
Today
Brain queryable via vector + filter search. Operator-side surface live.
Next
Customer-facing chat embedded on PDP. Same query engine, new shell.
/brain
Preview →
Open — What we'd want to learn from your users
The shape only you can name.
Theme A
Intent shape on the site.
01
When a customer lands on a PDP, roughly what % is in break-fix mode versus project mode?
02
What's the most common customer-stated project you hear from sales calls or support tickets?
Theme B
Rationale shape per category.
03
For your top category, what does rationale need to address — compatibility, fit, alternatives, system-level concerns? Rank them.
04
What's a recommendation you'd flag as bad even if it converts? The floor of acceptable.
Theme C
Channel hand-off + personas.
05
When paid traffic delivers a buyer to a PLP, what does that buyer expect to see first? Does it change by ad creative?
06
Across your customer types, which one is most under-served on the site today?
Theme D
Where to start.
07
If we shipped V1 to one PLP next month, which one teaches us the most?
08
What does we got value look like to you in 30 / 60 / 90 days?