How Brand Partner Index Surveys ~5,270 European Fashion Brand Partners via AI Voice Agent

The supply-side of European fashion e-commerce generates more than €50bn in annual GMV across Zalando, ASOS, About You, Otto, and Amazon Fashion. The brands that supply this inventory, from Adidas and Nike down to the 7,000 smaller brand partners, have opinions, grievances, and allocation decisions that shape platform outcomes. Until recently, no systematic, independent, recurring dataset captured these signals at scale.

This post explains how Brand Partner Index collects that data.

Why this data didn't exist before

Collecting honest sentiment from 5,000+ brand operators per quarter has historically required human callers at €50–200 per interview. At that cost, the research stays artisanal: 50–300 interviews per quarter, single-vertical, single-geography, one-off rather than longitudinal. Hedge funds and platform operators would pay for the data in principle, but the economics of production made it non-viable.

AI voice agents changed the economics by 25–100×. Per-interview cost dropped to €1–2. A 5,270-brand quarterly panel, previously a €500,000+ production cost, became a €10,000–€15,000 marginal cost per wave. That step-change in economics is what makes BPI possible.

Panel universe construction

The Brand Partner Index universe covers ~5,270 unique brand partners across Zalando, Otto, and ASOS as of the Q1 2026 panel frame. Universe construction involved:

1. Source aggregation. Brand partner lists were assembled from platform partner portals (where publicly accessible), industry directories, and brand self-identification (where brands had listed their platform distribution). No scraping of private seller databases was performed.
2. Deduplication. Brands appearing on multiple platforms are counted once in the universe; the interview covers all platforms they sell through, enabling cross-platform comparison.
3. Contact enrichment. For each brand in the universe, decision-maker contacts (Head of E-Commerce, VP Wholesale, Head of Channel Strategy) were identified from LinkedIn, company websites, and trade registries. A phone number with confidence score is associated with each contact.
4. Stratification. The universe is stratified by brand-revenue tier (large, mid, small), platform mix (Zalando-only, multi-platform, ASOS-heavy), and geography (DACH, UK, Southern Europe). This stratification enables cross-tabulation in the delivered dataset.

Interview methodology

Brand Partner Index is a qualitative sentiment panel, not a Likert-scale survey. The product would be worthless if it were one. A 1-10 score on "payment terms satisfaction" tells a buyer nothing they couldn't already infer from public filings.

What buyers pay for is the operator's actual words about what is happening, and what is changing, and why. Each interview is built around open-ended prompts that surface specific operational reality:

• "How have payment terms changed over the past two quarters? Walk me through what shifted."
• "Where do you see the platform's commercial relationship with brand partners heading?"
• "If you had to reduce allocation across one of your platforms next season, which one, and why?"
• "What's the single thing the platform is doing today that you'd want corrected?"
• "Compared to ASOS / About You / Amazon Fashion, where does this platform actually win, and where does it lose?"

The agent has the latitude to follow up. If a brand says payment terms are getting worse, the agent asks how much worse, when it started, whether it has affected the brand's planning. The output of every call is a transcript of that conversation, not a tick-box form.

AI agent status is disclosed upfront per GDPR Art. 22. Only contacts in brand-side commercial, wholesale, or e-commerce roles are eligible. Average call duration: 6–10 minutes.

What gets extracted

After each call, the transcript flows through a structured extraction pipeline. The point of the pipeline is not to compress sentiment into a number — it is to make ~5,000 unstructured conversations searchable and comparable:

1. Transcription via Whisper or comparable streaming STT, with per-segment confidence.
2. Topic tagging. Each segment is tagged with topics that came up: payment terms, returns friction, marketing co-op, channel allocation, competitor mentions, etc. A single call typically touches 4-7 topics.
3. Sentiment polarity per topic. For each topic the operator discussed, an extracted polarity (negative / neutral / positive) plus directionality (worsening / stable / improving). This is the closest the dataset gets to a "score," and it sits next to the verbatim it was derived from, never in isolation.
4. Verbatim retention. The actual quote that supports each topic + polarity is retained. Buyers can search "all brands describing returns friction as worsening over the past two quarters" and read the operator's actual words.
5. Audit trail. Every extracted field carries a transcript ID and a confidence score. A buyer who wants to verify can pull the transcript and read the segment.

What the dataset looks like

The delivered dataset is transcript-first, not row-per-score. Each interview produces:

• The full anonymized transcript (typed text + timestamped audio segments).
• A topic-tagged structure with verbatim quotes attached to each tag.
• An extracted sentiment polarity + directionality per topic.
• Stratification metadata: brand tier, platforms used, geography, role of contact.

A buyer querying the dataset is not pulling a column from a spreadsheet. They are pulling operator quotes filtered by platform, topic, sentiment direction, and quarter — backed by transcripts they can read and audit.

The institutional-tier dataset includes the full verbatim archive plus aggregate views: topic frequency by quarter, sentiment-direction shifts wave-over-wave, and cross-platform comparisons grounded in quote evidence.

Publication tiers

Brand Partner Index uses a tiered release model:

The institutional tier is what institutional buyers subscribe to at an annual license fee. The participant tier creates a response-rate incentive (brands that receive their own benchmark are 5–10× more likely to respond to the next wave).

Why independence is non-negotiable

Brand partners will not report honestly about Zalando on a Zalando-commissioned survey. The commercial relationship prevents truth. The structural analogy: Coca-Cola publishing a blind taste test against Pepsi. Independent researchers fill the structural gap, and the best-case outcome for Zalando is to become a subscriber to the data it cannot credibly produce itself.

Brand Partner Index is published by Auraqu, Inc., which has no commercial relationship with any measured platform. Auraqu's only commercial relationships are with data subscribers, who pay for the independence.

Accessing the data

Brand Partner Index Vol. I preview is targeting Q2 2026. Founding-subscriber pricing is available now for institutional buyers.

View the publication at brandpartnerindex.com or contact us to discuss a data subscription.