A consent-first user-intelligence layer designed during Phase 1 (2020–2024), before AI emerged into the public conversation. The same dataset now serves three simultaneous beneficiaries.
Conventional commerce analytics serves the business. Conventional AI training data serves AI labs. User-research panels serve the user. Mazzaneh Analytics was designed before any of these categories were built around AI — which is precisely why the same architecture serves all three at once.
The design constraint was simple: users should benefit from sharing data, not be exploited for it. That single constraint, taken seriously across every module, produced a structured dataset that businesses can target on, that users earn from, and that AI systems can train on legitimately. None of these outcomes were retrofitted — they fall out of the original commerce design.
Mazzaneh's analytics methodology was designed during Phase 1 (2020–2024) — before LLMs entered the public conversation. The choices were made for commerce reasons: precision targeting at low waste, lawful consent capture, and user understanding without intrusion. The fact that this same architecture turns out to be exactly what AI systems need is structural, not a retrofit.
Before LLMs became the dominant analytical paradigm. The methods were chosen for commerce reasons, not AI reasons. That timing is the structural advantage — it means the architecture was not optimized around any specific AI vendor or technique.
Built around explicit user-stated signal, not behavioral inference. Inference-based shortcuts were not yet available, so the methodology had to ask directly and reward participation. That “limitation” is now the moat — consent-explicit data is exactly what regulation, users, and AI all increasingly demand.
The same architecture serves business analytics today AND happens to produce the structured behavioral data AI systems need for downstream tasks. That dual fit was not engineered post-hoc — it falls out of the original commerce design. The bridge is structural.
Every interaction in Mazzaneh produces value in three places at once. The user earns. The business targets. The AI learns. None of these are by-products — all three are first-class outcomes designed into the architecture from the start.
Conventional advertising shows a product to many users and hopes a small percentage match. Consent-first analytics shows the product only to users whose stated and validated preferences match. The signal-to-noise ratio shifts in favor of the business, and the structural waste of showing the wrong product to the wrong user is removed at the architecture level.
In most platforms, users are the product. Their attention is sold, their behavior is tracked, and the value of their data accrues to advertisers and the platform — not to them. Mazzaneh inverts that: the user is the customer of their own data. Each piece of profile signal earns income through Pulino. Each gamified ad through Board pays the user for engagement. Relevance is delivered as a service, not extracted as a resource.
Modern AI systems struggle with a structural problem: deep user understanding requires either inference (shallow, multi-session, regulatory exposure), opt-in memory (most users do not volunteer at depth), or third-party behavioral data (ambiguous, increasingly restricted). Mazzaneh produces a fourth path: data the user explicitly stated, validated through their own behavior, with permission to use. That is a category most AI training pipelines cannot legitimately produce at scale.
Mazzaneh is a super-app e-commerce platform — Phase 1 of the MZN ecosystem. The data flow from Mazzaneh into AI systems is not a generic data-licensing pipeline. It is an architectural bridge with three properties that conventional pipelines do not have.
Users answer questions about themselves to earn income on the platform. The consent is explicit, paid, and renewable. The signal is structured — not inferred from clicks, not scraped from behavior, not purchased from third-party brokers. Every data point has a user who chose to share it and was rewarded for doing so.
Stated preferences are validated through downstream behavior — search queries, purchase patterns, response accuracy on Board campaigns, time spent on Style Finder. Self-reported claims that do not survive behavioral validation are filtered out. The remaining signal is high-confidence: the user said it, and the user lives it.
The validated, consent-explicit dataset is structurally appropriate for AI fine-tuning, behavioral evaluation, alignment calibration, and personalization. Use is permission-bound from the user side and architecturally lawful from the platform side. AI systems get data they can use legitimately at depth that inference-based approaches cannot match.
In a single-purpose analytics product, data flows in one direction: collected, processed, displayed. In Mazzaneh, every module has a commercial purpose AND an analytics purpose AND an AI-feed purpose. Each module added increases the value of all the others — that is the compounding property. The dataset becomes denser, more validated, and more useful with every additional module participating.
Users answer profile questions to earn income. Each answer becomes a structured behavioral signal. Pulino is a commerce mechanism that produces analytics signal as a side-effect.
Gamified ad campaigns where users answer 4 questions about a product in 20 seconds. Comprehension validation is a commerce mechanism. The response patterns produce taste signal as a side-effect.
Local purchase requests with response time, location, and transaction confirmation. Radar is a commerce mechanism. The funnel data produces purchase-intent signal as a side-effect.
Visual preference selection across product categories. Style Finder is a UX mechanism. The pattern of selections produces aesthetic-preference signal as a side-effect.
Wardrobe and sizing inputs (~95% accuracy on size matching). User utility on top, structured fashion-domain signal underneath. The user benefits, the dataset improves.
The analytics layer feeds back into Board: campaigns are matched to user profiles automatically. The output of analytics becomes input to the commerce mechanism, closing the loop.
Search results filtered by user taste. Same query, different results per user, based on validated preference. The same architecture that targets ads also personalizes search.
The unified profile that emerges from all the above feeds HUAI fine-tuning, Zoyan personalization, and ZOE evaluation pipelines. The same dataset that drove the commerce side now feeds the AI side.
A systematic approach to understanding the user with their full participation. Each stage produces value for all three beneficiaries: the user earns, the business targets, the AI learns.
Users answer questions about their preferences, interests, and tastes. They answer voluntarily, because they earn rewards for doing so. Coverage spans personal traits and demographics, style and fashion preferences, shopping habits and price sensitivity, interests and lifestyle. Every answer is explicit, paid, and revocable.
Users see different product images and answer targeted questions to determine precise taste profiles. Visual preference testing, brand affinity scoring, multi-axis style classification (~80 dimensional taste profile), and threshold-based filtering (typical match threshold: ~80%). The user enjoys the experience — the system gains depth no inference-based approach can reach.
The analytics layer produces actionable user groups and dashboards for participating businesses. High-match user groups (matched by stated and validated preference), category popularity, characteristic profiles, and smart ad recommendations. Each output is anonymized at the identity layer — businesses see user characteristics, never names or phone numbers.
A structural comparison across four properties that distinguish consent-first commerce analytics from conventional approaches. Each named platform has strengths in its own design space; this comparison highlights where Mazzaneh occupies an architectural position other platforms were not built to occupy.
| Platform | Primary Signal | Consent-Explicit | Local Commerce | Behavioral Validation |
|---|---|---|---|---|
| Search-based platforms | Search intent | Limited | Limited | Indirect |
| Marketplace platforms | Purchase data | Limited | Limited | Yes |
| Social-graph platforms | Social signal | Limited | Indirect | Indirect |
| Mazzaneh Analytics | User-stated preference + validation | Explicit | Yes | Direct |
The analytics layer produces four kinds of output, each useful to a different beneficiary, all derived from the same consent-first dataset.
Users whose stated and validated preferences align with a given product profile above a configurable threshold (typical: ~80%). Filterable by occupation, hobbies, lifestyle, entertainment preferences, professional domain. Below-threshold users are filtered out at the architecture level — no mismatched-impression spending.
Anonymized profiles describing who actually buys what. Filter by job and profession, hobbies and interests, entertainment preferences, lifestyle segments. The dashboard never exposes user identities — businesses see characteristics (“User #9343: profession, taste, vehicle, follow patterns”), never names or phone numbers.
Don't just know who to target — get specific recommendations for advertising method, optimal campaign locations, messaging that resonates with each segment, and budget allocation suggestions. The output of analytics becomes the input to the commerce mechanism, closing the targeting loop.
A structured behavioral dataset suitable for fine-tuning, evaluation, and alignment work. Explicit-consent attributes, behavioral validation, multi-modal signal, pseudonymous identity layer. Available to AI partners under coordinated correspondence and partnership-tier engagement, not as a generic data product.
Three beneficiaries. One architecture.
The bridge was a structural consequence of the design, not a retrofit.
For businesses ready to leverage consent-first targeting, users curious about earning from their own data, or AI partners exploring data-architecture alignment — the conversation begins with the broader MZN ecosystem.