ArveLabsContact

Engine

One engine,every product.

AI made building cheap. Knowing what to build next, what to charge and where to find the people who want it is where products are now won. The engine is a system of AI agents that answers those questions, product after product.

PRODUCTSPRISMFITFLOWCOOKIEJARPAGEVERAFOLDFOCUSUNFOLDWALLSUNHEATEDTHE ENGINEONE SHARED MEMORYARVEAI OPERATORSIGNALLABSTORECREATIVEMEMORYOUTPUTSEXPERIMENTSSTORE LISTINGSLAB NOTESNEXT PRODUCTWHAT WE LEARN SEEDS THE NEXT PRODUCT

Products

PrismFitFlowCookieJarPageVeraFoldFocusUnfoldWallsUnheated

The engine · operated by Arve

Arve
SignalLabStoreCreativeMemory

Outputs

ExperimentsStore listingsLab notesNext product

↺ What we learn seeds the next product

Figure 1. Every product reports into one engine. Arve operates its five systems; what they learn becomes the starting point for the next launch.

AI-native

Run by agents,decided by people.

Every system below is operated by AI agents working from one shared memory. A person approves anything that leaves the studio.

A typical day in the studio07:00
  1. 07:00PlanOrchestrator agentReads every product’s numbers and picks the day’s work, product by product.
  2. 09:00BuildCoding agentsShip features, fixes and store builds from written briefs.
  3. 12:00MeasureAnalyst agentsPull revenue, funnels and search ranks through official connectors.
  4. 15:00CreateCreative agentsDraft listings, screenshots and ad variants ready to test.
  5. 17:00ApproveA personDecides what ships, spends or publishes. Nothing goes out without it.
  6. 18:00RecordArveWrites the verdict into the lab notes and the shared memory.

Systems

Five systems.One loop.

Arve operates all five day to day. A person approves anything that ships, spends or publishes.

Signal

One event language for the whole portfolio.

Every product reports onboarding steps, paywall views, trials, renewals and cancellations the same way. A news product and a fitness product can sit in the same table and be compared honestly - which is the only way a portfolio learns anything at all.

  • A shared event schema across products
  • Analyst agents read revenue, behaviour and store data daily

Lab

Pricing, paywalls and onboarding are never finished.

Changes ship as experiments: a hypothesis, a metric, a stopping rule. Results are judged against category benchmarks drawn from more than a hundred thousand subscription businesses, not against our own hopes.

  • Agents draft the hypothesis and the variants
  • Benchmarks set per category, not per mood

Store

Search is the cheapest distribution there is.

Keywords, titles and screenshots are researched and tracked market by market, and localised only where the numbers justify it. A listing is treated as a product surface, not an afterthought.

  • Agents research keywords and write localised listings
  • Rank tracked per country, every week

Creative

Borrow what works. Never copy it.

We study the formats that already win in each category, brief creators on the angles our own numbers can support, and buy media only after the organic signal is clear. Every creative has a budget and a kill rule.

  • Generative tools produce variants in minutes
  • Paid spend unlocked by evidence, not enthusiasm

Memory

Every result ends in a written verdict.

Wins, losses and dead ends are recorded in one place with the evidence attached. A new product does not start from zero; it starts from everything the portfolio has already paid to learn.

  • Every agent reads the same record before it acts
  • New products inherit proven defaults

Gates

Four gates.No shortcuts.

Every product passes the same gates in the same order. None is skipped, however promising the early signs.

GateOpens whenUnlocks
IMeasureSubscriptions, funnel events and attribution report end to end.Any paid traffic
IIMonetisePaywall and prices set from category evidence; billing grace switched on.Pricing experiments
IIIAcquireConverts at 4% or better by day 35 and holds a 4.5★ rating.Apple Search Ads
IVScaleOne paid channel above break-even, two weeks running.Creators and short video

The launch kit

Eleven standards every product must meet

Measure
K1Subscriptions liveK2Paywall platformK5Funnel eventsK6Ad attribution
Monetise
K3First-session paywallK4Benchmark pricingK7Billing grace period
Retain
K8Cancel flow
Search
K9Store page standard
Learn
K10Weekly data join
Ship
K11Review pack

Evidence

Benchmarks,not hopes.

Prices, paywalls and trials start from public evidence, then move only when our own experiments say so.

Download to paid, day 35

Median · RevenueCat, State of Subscription Apps 2026

2.0%All apps
2.1%Freemium
2.9%Health & Fitness
10.7%Hard paywall

A hard paywall converts about five times better than freemium, with the same first-year retention. We start there.

Trial to paid, by trial length

Median · RevenueCat, State of Subscription Apps 2026

≤ 4 days25.5%
5–9 days37.4%
17–32 days42.5%

In motion

The lognever stops.

Every action the engine takes is written down - what changed, where, and what we expect to happen next.

Read the lab notes
arve · studio log replay
  1. 11 OCTfoldfocus

Instruments

App Store ConnectRevenueCatPostHogSuperwallSupabaseSonarApple AdsDataFastResendNotionHiggsfieldClaude

We build what makes us different and rent everything else.