Trove Boutique Curated Intelligence
Issue №07 · Curated Intelligence · Editorial Commerce

A curator's intelligence,
at the speed of a recommendation engine.

Trove Boutique is a recommendation and curation platform for independent retailers, design boutiques and editorial commerce. We model taste the way a great buyer does — through references, context and conviction — and refuse to optimise for vibes-free clicks.

Apply for an atelier slot See the collection
Four propositions

What a buyer would build, if a buyer could write models.

Four things separate Trove from every recommendation system that optimises for the metric a buyer would never use.

01 · Taste embeddings
Models trained on real buyer reasoning, not engagement logs.
Recommendations should behave like a colleague: they remember what mattered last week, they explain their reasoning when asked, and they are willing to be wrong on purpose to be useful. We do not train on raw click-streams.
02 · Editorial-grade catalogues
Auto-curated category pages, edits, and lookbooks.
A filter grid surfaces products. An edit surfaces an argument. Argumentless commerce is high-friction and low-trust; Trove Curate generates editorial-grade context for every category and SKU without flattening the texture that makes them readable.
03 · Provenance aware
Materials, makers, origin — surfaced in the buying flow.
Every recommendation comes with curator-level evidence behind it. Where the oak comes from. Who did the joinery. Which dye-house touched the textile. Surfaced naturally inside the description, not exiled to a virtue-signal page nobody visits.
04 · Small-catalogue strong
Designed for 300 – 30,000 SKU catalogues.
Not a scaled-down enterprise system — a purpose-built one. The independent end of the market has been ignored by the major recommendation platforms for a decade; we built Trove to be exact for the shape of catalogue this audience actually keeps.
+11.4%
Units per transaction · pilot avg
−7.8%
Return-rate reduction
300–30,000
SKU catalogue range
19
Boutiques in atelier programme
The collection

Three pieces, one editorial discipline.

Piece I — Trove Curate
Editorial-grade category pages, edits and lookbooks.
Generated from your catalogue and your point of view. Imitates the voice of your buyers, not the voice of an AI.
EditorialMerchandisingStorefront

Read the piece →

Piece II — Trove Recommend
Recommendations modelled on reasoning, not raw engagement.
Trained on curator-labelled reference sets, post-purchase satisfaction and explicit affinity statements. Click logs are deliberately under-weighted.
RecommendationsTasteConversion

Read the piece →

Piece III — Trove Provenance
Materials, makers and origin as a working graph.
Surfaced naturally inside the buying flow. Auditable. Refused where we cannot evidence the claim.
ProvenanceSustainabilityTrust

Read the piece →

From the journal

There is a difference between a recommendation system that models taste and one that models engagement. The boutiques that have understood this difference are pulling away from the ones that haven't, and the gap is widening every quarter.

The temptation, when a recommendation engine fails to lift a conversion metric, is to chase the metric. Lower the price-anchor threshold. Surface more novelty. Inflate the click-through with brighter thumbnails. None of this works for a boutique whose customer cares how things were found — and increasingly few customers do not.

Three recent pieces from the journal explore the choice in detail. Each is short. Each is direct. None of them apologises for any of the trade-offs.