Parallel World. For data partners

The difference

Commanded behaviour vs revealed behaviour.

Every major data supplier in the AI training and evaluation market sells the same shape of input: structured responses from paid annotators. Useful, but not what users actually do. Parallel World captures the opposite.

What's on the market today

Commanded data

Paid annotators answering questions in a sandbox
Synthetic scenarios reverse-engineered from product hunches
One-shot tasks with no follow-through
Sources unclear; copyright and AI Act exposure variable

What Parallel World captures

Revealed behaviour

Real users in real apps, on real tasks they chose
Side-by-side moments across Claude, ChatGPT, Gemini, Perplexity
Regenerate, switch, copy out, abandon. Captured cleanly
Consented and paid at source, with full audit trail

What's in the data

Behaviour your synthetic stack can't generate.

Parallel World captures the moments where a user actually makes a choice. Across models, across sessions, across months. The training, evaluation and reliability signal your team has been inferring indirectly.

Cross-model side-by-side

Same person, same task, multiple AI surfaces open. Which one they pick. Why they switched.

Failure moments, labelled

Hesitation, regeneration, abandonment, copy-and-exit, manual takeover. The signal your evaluators currently infer.

Real workflows end to end

Full browser sessions across research, checkout, support and tool-to-tool journeys. Not stitched fragments.

Longitudinal continuity

The same users tracked across months as your models, and the web around them, evolve. Drift becomes measurable.

Data categories at launch

Five categories. Live or imminent.

Each category ships with sample records, a schema, k-anonymity thresholds, and a median record age. Partners can scope a pilot against one category or across several.

01

E-commerce behaviour

Retail, DTC and marketplace journeys. Product comparisons, dwell time on price bands, intent scores.

user_id_hashed · category: "Footwear / Running" · price_band: "$80-140" · competitors_viewed: 3 · intent: 0.82

02

App usage patterns

Mobile session frequency, feature engagement, cross-app switching, time-of-day archetypes.

cohort_id · 14 sessions / 7 days · avg_duration: 4m12s · pattern: "commute"

03

Cross-platform journeys

Multi-device attribution and tool-to-tool handoff. The full path from research to choice.

journey: "Mobile → Web → Ext" · outcome: converted · lift: +18%

04

Cross-model AI use

The same user on Claude, ChatGPT, Gemini and Perplexity on the same task. Which they pick. Why they switch.

task_type: "research" · models_tried: [claude, chatgpt, perplexity] · chosen: claude · regenerations: 2

05

Engagement and discovery

Topic affinity, creator follows, content velocity, social engagement archetypes by cohort.

cohort_size: 12,408 · topic: "ai-tools" · follow_rate: 6.4% · velocity: "high"

Two ways to start

A scoped first pilot, not a marketplace pitch.

The first commercial relationship is a small, dated, defined pilot. There are two shapes ready to deliver.

Shape A

Behaviour data pack for AI testing

A consented dataset around one workflow family. Steps, tool context, outcomes, and a short taxonomy of failure points (hesitation, abandonment, correction, retry, takeover). Designed to slot directly into your evaluator calibration or judge training pipeline.

Best fitEvaluator platforms, agent reliability teams, AI security and red-team partners.

Shape B

Browser session corpus

Full web sessions across research-to-decision and tool-to-tool journeys. Designed for benchmarking browser agents, measuring policy compliance, or fine-tuning agent memory and planning layers.

Best fitBrowser-agent infrastructure, coding agent vendors, enterprise agent platforms.

Who this is for

The teams already buying data of roughly this shape.

If you sell software that tests AI systems, builds browser or coding agents, secures runtime AI, or supplies behavioural data to the labs, Parallel World is a clean upstream source for the inputs you already need.

Eval & reliability

Judge models, eval calibration, agent trace analysis

Agent infrastructure

Browser agents, coding agents, multi-tool orchestration

Security & red-team

Behaviour-grounded scenarios, runtime policy testing

Insight & intent

MarTech holding companies, insight houses, financial services

Provenance and procurement

Built for the procurement questions you already ask.

Consented at source

Every record begins with the user opting in, at the granularity of data type. Consent is auditable per record.

Paid at source

The user is compensated for the data you buy. No scraping. No silent collection. Cleaner than the alternative routes when procurement asks where the data came from.

Differential privacy by default

Hashing, k-anonymity thresholds and differential privacy noise applied before any record leaves our infrastructure. Configurable per category.

EU AI Act compliant

Designed against Article 53 training data disclosure requirements. Hosted in France. Suitable for partners with European deployment exposure.

Delivery and freshness

Live ingestion. Choice of pipe.

Median record age

Under six hours from user activity to partner-accessible record. Continuous ingestion, not batched dumps.

Delivery options

REST API, Snowflake share, or S3 drop. Schema versioned. Webhook events for ingestion-rate partners.

Pilot to production

Pilots scoped to a single category and a fixed volume. Production access opens against signed LOI with a named contract.

Live today

Chrome extension live across the major LLM surfaces. Android mobile in build. iOS in evaluation. First partner pilots scoping for Q3 2026 delivery.

The team

The people behind Parallel World.

Founders and operators with category-defining track records in consumer products, AI infrastructure, regulated finance and digital assets.

Livio Bisterzo

Founder & CEO

Serial entrepreneur with multiple exits across consumer products, technology and sustainability. Founder of Hippeas (acquired) and Green Park Brands.

Angelo Salvetti

Chief Strategy Officer

25+ years building and scaling ventures in technology, telecom and entertainment. Created over US$500M in enterprise value through two exits and a Nasdaq IPO. Based in Brazil.

Andrew Morfill

Chief Operating Officer

Senior technology and security executive in fintech, Web3 and digital assets. Former CSSO at Komainu and Global Head of Cyber Defence at Banco Santander.

Rory Anderson

Chief Marketing Officer

Experienced company builder who has scaled multiple VC-backed startups globally. Former CGO at Partanna, raised $27M and expanded internationally. Strategist across climate tech, infrastructure and finance.

Arno Frantz

Founding Consultant

Digital asset and derivatives specialist with 15+ years in quantitative trading, market structure and risk systems. Co-founder of NimbleAlpha. Deep expertise in perpetual futures, options and DeFi.

Talk to us

Book a 30-minute pilot conversation.

Tell us what you're testing, training or evaluating. We'll come back with a scoped pilot shape and a sample slice within 48 hours.

Real user behaviour.Consent at source.