Rolling Out Size Recommendations Across Brands

Portfolio retailers rarely have one sizing problem — they have five, ten, or twenty of them, one per label, each with its own fit block, its own size chart conventions, and its own returns history. Rolling out a size recommendation engine across a house of brands without a plan usually produces either a generic tool that fits no one well, or a pile of one-off integrations that nobody can maintain. This is the playbook for doing it right.

Why multi-brand size recommendation rollout is different

A single-brand merchant can hand-tune one model against one fit identity. A house of brands cannot use that approach at scale — the operational cost of rebuilding logic per label, per market, per season grows faster than the headcount to manage it. At the same time, brands inside a portfolio are rarely interchangeable: a denim label with a slim, developed-on-live-models fit and a basics label with a relaxed, size-inclusive fit cannot share one recommendation curve without one of them looking wrong to customers.

The tension is structural. Group technology teams want one platform, one API, one dashboard. Brand merchandising teams want their fit DNA respected, because fit is often the single biggest driver of repeat purchase and the single biggest driver of returns when it's wrong. A rollout plan has to resolve that tension deliberately, not by accident.

The cost of getting this wrong

Fashion retail groups report return rates of 20-30% in apparel, and size/fit issues are consistently cited as the leading cause according to industry data compiled by the National Retail Federation. When a shared engine is rolled out without per-brand calibration, the failure mode is invisible until returns data comes in three months later — by then, a full season of customers has already had a bad fit experience with a brand they may not try again.

Architecture: one engine, per-brand configuration

The pattern that scales is a shared recommendation engine with brand-level configuration layers, not a shared model. Concretely:

This mirrors how most multi-brand retail groups already structure supply chain and inventory systems: one ERP, brand-specific SKU logic on top. Applying the same separation to sizing keeps engineering effort centralized while letting merchandising own the parts that actually touch the customer relationship.

Fit profiles as the unit of brand identity

Each brand should own a fit profile — the specific relationship between body measurements and garment measurements that defines "true to size" for that label. A slim-fit brand and a relaxed-fit brand measuring the exact same customer should legitimately recommend different sizes, and the shared engine needs to treat that as a configuration input, not an exception to work around.

Getting this right depends on having enough size options in the first place; see how to build a size assortment that matches your customers for how assortment and fit-profile work feed each other.

A phased rollout sequence

Rolling out simultaneously across all brands is the single most common mistake. A phased sequence controls risk and builds internal proof points.

Phase 1: pilot on one brand with clean data

Pick the brand with the most complete size chart, the most returns history, and the most cooperative merchandising team — not necessarily the flagship or biggest-revenue label. The goal of phase one is validating the engine's accuracy against real returns data, not maximizing immediate revenue impact.

Phase 2: expand to brands with similar fit logic

Group brands by fit family — slim/tailored, relaxed/casual, size-inclusive — and roll out to the next brand within the same family as the pilot. Configuration changes should be incremental at this stage, which keeps the governance overhead low while the team builds a repeatable rollout checklist.

Phase 3: onboard structurally different brands

Brands with very different fit logic (activewear vs. outerwear, for instance) or different markets need dedicated fit-profile work before go-live. This is also the phase where localization matters most — a brand selling into multiple regions needs region-specific size chart conventions, not just translated labels. See how to localize sizing for international markets for the mapping work this requires between US/UK/EU conventions.

Phase 4: steady-state governance

Once all brands are live, the rollout becomes an ongoing governance function: quarterly fit-profile reviews per brand, shared dashboards for group-level returns reporting, and a change-management process for updating any brand's fit profile without needing an engineering release.

Governance: who owns what

A common failure after rollout isn't technical — it's organizational. Nobody owns fit-profile accuracy once the initial project team disbands. Set this up explicitly before launch:

• Group platform team owns the engine, integrations, and uptime.
• Brand merchandising leads own their fit profile and size chart, with a defined review cadence (quarterly is typical).
• Group analytics owns the shared returns/fit-accuracy dashboard so leadership can compare brands on a common metric without flattening their fit identities into one number.

McKinsey's research on retail analytics maturity notes that portfolio retailers who centralize data infrastructure while decentralizing decision rights on customer-facing configuration consistently outperform those who centralize both — a pattern documented by McKinsey & Company that applies directly to sizing governance.

Choosing the underlying platform

Not every size recommendation vendor is built for multi-brand deployment. Before signing, confirm the platform actually supports brand-level fit profiles as a first-class configuration, not a workaround, and that group-level reporting doesn't require exporting data into a separate BI tool by hand. The core evaluation criteria are the same ones that apply to a single-brand deployment, just multiplied — see how to choose a size recommendation solution for the full checklist, and weight multi-tenant configuration and per-brand reporting heavily if you're buying for a portfolio.

Questions to ask vendors specifically for portfolio rollout

Ask how fast a new brand can be onboarded once the platform is live elsewhere in the group — this should be weeks, not a re-implementation. Ask whether fit-profile changes for one brand require any downtime or code change for others. And ask for a reference customer that runs multiple labels on the same platform, not just multiple markets for one brand.

Measuring success across the portfolio

Track both group-level and brand-level metrics, and don't let one obscure the other. Group-level return rate reduction proves the investment case to leadership. Brand-level fit-profile accuracy (recommendation-to-keep rate, size-exchange rate) proves the engine is actually respecting each label's identity rather than dragging every brand toward an average.

Frequently Asked Questions

How long does a multi-brand size recommendation rollout typically take?

A pilot on one brand usually takes four to eight weeks including data preparation and validation against returns history. Full portfolio rollout timing depends heavily on how many distinct fit families exist across the brands, but a phased approach across four to six brands commonly runs six to twelve months end to end. Structurally different brands, such as those with very different fit logic or new markets, extend the timeline the most.

Should every brand in the portfolio use the exact same size labels?

No. Size labels should follow each brand's existing convention and target market, since forcing a shared label set onto brands with different customer bases creates confusion rather than consistency. The recommendation engine and underlying fit logic can be shared even when the labels displayed to customers differ completely between brands.

Who should own fit-profile accuracy after the initial rollout project ends?

Ownership should sit with each brand's merchandising lead, supported by a group platform team that maintains the shared engine and reporting infrastructure. Without an explicit owner, fit profiles tend to drift out of date as brands introduce new fit blocks or categories, which quietly degrades recommendation accuracy over time.