Who they are

The Roost is a curated homeware marketplace, bringing together independent and design-led brands in a single destination - furniture, lighting, textiles, tableware, decorative pieces and paint. The proposition is a considered alternative to the big-box retailers: everything for the home, chosen properly, in one place.

A marketplace carries a measurement problem that a single-brand store does not. Performance is the aggregate of many brands, many categories and several very different buying behaviours, all running through one checkout and reported as one number. A headline conversion rate or average order value describes none of them accurately.

The question worth answering was not whether the store was selling. It was which parts of the model were actually working, and which were being carried by the others.

What we did

A fixed-scope Gold Audit - a strategic read of the business, built from the client's own raw data rather than a dashboard summary.

Marketplace reporting tends to average everything together. A furniture sale and a candle sale become one blended order value; a returning trade buyer and a first-time gift shopper become one customer. Those averages describe a business that does not exist, and decisions taken from them tend to be aimed at nobody in particular.

We brought several years of operational data into a single analytical framework:

• Order and transaction history - three trading years of orders, examined line by line rather than in aggregate.
• Customer records and cohorts - grouped by when they first bought, so that repeat behaviour could be measured properly over time.
• Full product and catalogue exports - every active product and variant, tested against what had actually sold.
• The live site - navigation, collection structure, product pages and the discovery journey, reviewed alongside the data rather than separately.

In total, more than 147,000 individual records.

The method matters more than the volume. We built cohort curves to separate genuine retention from the flattering effect of new-customer growth, matched historical revenue against the current catalogue to see what was still purchasable, and analysed baskets by composition - single-item against multi-item, single-brand against cross-brand - to test whether the marketplace promise was showing up in what people actually bought.

What we found

All figures, revenue data, order volumes, brand-level performance and account details remain confidential. What follows are the strategic findings only.

The first order was solved. The second was the whole problem.

Acquisition was working. Repeat purchase was not, and by a wide margin against the norm for homeware. The cohort curves ruled out the usual excuse: the earliest cohort repeated at roughly three times the current rate, which proves the proposition can hold people. This was not a product problem or a market problem. It was the absence of any systematic reason to come back.

The timing finding made it actionable. Where second orders did happen, most landed within a month of the first, and a customer who had gone quiet for three months was statistically gone. The window to earn order two is far shorter than most retailers assume - and it is precisely the window in which most stores send nothing but a dispatch email.

Two businesses were sharing one checkout

The order data contained two distinct businesses. One is a high-frequency accessories shop - tableware, prints, smaller decorative pieces - which drives the order count. The other is a low-frequency, big-ticket furniture business, which drives the bank balance. They share a storefront and almost nothing else.

They need opposite treatment. The first is a repeat-purchase and gifting engine, where frequency is the goal. The second is a considered purchase where finance, delivery clarity and follow-up decide the sale. Averaged together, they produce metrics that describe neither - and marketing built on those averages speaks to neither customer.

The one-roof promise was not showing up in baskets

The strategic pitch of a curated marketplace is everything for your home, chosen well, in one place. The basket data told a different story: most orders left with a single item, and among multi-item baskets, only a minority mixed more than one brand. Customers were shopping a brand they had found, not the marketplace itself.

This is a merchandising gap rather than a proposition failure. The natural co-purchases were colour-pair twins and same-category duplicates, not the cross-brand, room-building combinations the model is built to deliver. The assets to fix it - room edits, styled inspiration, look-based discovery - already existed, but were being used as inspiration content rather than as basket mechanics.

Breadth is not the same as curation

A large majority of the active catalogue had never recorded a single sale, while a meaningful share of historical revenue traced to products no longer purchasable - proven sellers delisted, mostly when brands departed.

For a curated marketplace, that combination is the sharpest possible finding: rigorous about which brands come in, far looser about what happens to the shelf afterwards. An uncurated catalogue dilutes collection pages, bloats the product feed, and buries the products that actually sell. Vendor churn is normal in this model - but letting proven sellers leave without a replacement in the same niche is a decision with a measurable cost.

The highest-intent customers were getting the least follow-up

Customers ordering physical samples - fabric, wallpaper, paint - are the clearest statement of purchase intent in the dataset. They have measured a wall and shortlisted a colour. Yet this was the least-worked audience on the books, with no dedicated follow-up sequence at all, and the small number who did convert went on to spend many times the site average.

The pattern generalises well beyond homeware: wherever a low-value, high-intent step sits in front of a high-value purchase, that step is usually under-served precisely because it looks small in a revenue report.

The catalogue data was gating the growth channel

Product feed health rarely reaches a board conversation, but it decides what is possible in paid and organic search. Here, identifier coverage across active variants was low enough to limit or disqualify listings in Google Shopping - the channel where big-ticket homeware demand is most effectively monetised - and the large majority of products lacked search descriptions.

The strategic point is not technical. Ambition in acquisition was capped by a data quality problem nobody owned, and no amount of budget would have moved it.

A baseline, not just a verdict

The audit closes with a scored baseline across seven commercial dimensions - demand and momentum, retention and lifetime value, basket and merchandising, pricing discipline, catalogue health, operations and post-purchase, and site experience and discovery.

The value is in the repeatability. Each dimension maps to the findings that move it, and every one is re-measurable from the same exports at a re-audit. That turns a set of recommendations into something testable: not whether the work felt worthwhile, but whether the number moved.

Who this applies to

The Roost's model is specific, but the pattern is not. If any of these describe your business, the same findings are likely sitting in your data:

• You sell many brands, or many categories, through one storefront and report on them as one business.
• Your mix spans frequent low-value purchases and infrequent high-value ones, blended into a single average order value.
• Acquisition is performing and you assume retention will follow, without having measured it by cohort.
• Your catalogue has grown faster than anyone has curated it.
• A low-value, high-intent step - a sample, a swatch, a consultation, a quote - sits in front of your biggest purchases.
• Paid search and Shopping underperform, and nobody has checked whether the product data allows them to work.

In each case the headline numbers look reasonable, and the averages are hiding the two or three things that would actually change the business.