LAB LIVE
SCANNING414 CAMPAIGNS
ANOMALIES15 FLAGGED
AVG LIFT+4.10× ROAS
Back to cases
SL-ZA-263REV.ACLASSIFIED · CLIENT-VIEW
AI PLATFORM · RUSSIA · YANDEX.DIRECT

ZvenoAI

200+ AI models via unified API · B2B + B2C growth system

AI · SaaS
[00]DIAGNOSIS
One product, three incompatible audiences — B2B teams, API developers, and B2C users — all driven into a single funnel. Analytics stopped at registration, making it impossible to distinguish quality traffic from noise. No structural separation by intent or use case.
Registrations from ads
680
78% of all new users
Balance top-ups from ads
137
68% of all paid conversions
Ad groups (search)
20
intent-separated
Keywords
1,300
B2B + B2C combined
Funnel stages tracked
4
reg → API key → top-up
Audiences separated
2
B2B and B2C
[01]SITUATION

ZvenoAI gives access to 200+ AI models through a single OpenAI-compatible API, with payments in rubles and support for both individual and team use cases. The platform serves two structurally different audiences: developers and businesses integrating AI into workflows (B2B), and individual users running personal AI tasks (B2C).

The task was not to launch ads — it was to build a growth system for a complex product: separate the audiences, build campaign structure, craft differentiated messaging, and set up analytics that tracked real user actions inside the service, not just registrations.

[02]AUDIENCE SEPARATION

The first structural decision: B2B and B2C cannot share a funnel. Mixing use cases — from personal chat to enterprise API integration — produced messaging that fit neither audience well.

B2B
  • ·API access and integration
  • ·Team and workflow automation
  • ·Business use cases
  • ·Developer scenarios
B2C
  • ·AI chat and personal assistant
  • ·Writing and content creation
  • ·Individual task scenarios
  • ·Day-to-day AI access
[03]CAMPAIGN STRUCTURE — 20 GROUPS, 5 INTENT CLUSTERS

Search campaigns were rebuilt around demand intent, not product features. 1,300 keywords across 20 groups, segmented into 5 clusters — each with its own ad copy, landing path, and UTM tagging through to registration and balance top-up.

CLUSTERINTENT SIGNALAUDIENCE
API & ModelsDirect API access, model comparison, OpenAI alternativeTechnical / B2B
Business tasksAI for teams, content, automation, workflowsB2B
IntegrationAPI integration, SDK, compatible APIB2B / Dev
MultimodalityImage, voice, multi-modal queriesB2B / B2C
B2C scenariosAI chat, personal assistant, writing helpB2C
[04]ANALYTICS ARCHITECTURE — BEYOND REGISTRATION

Registration alone was not a useful optimization signal for a freemium AI platform. Four Metrica goals mapped the full activation funnel, making it possible to evaluate campaigns by user quality, not just traffic volume.

01
Registration
Top of funnel
02
API key created
Activation signal
03
Top-up opened
Intent to pay
04
Balance topped up
★ Primary quality signal
REGISTRATIONS FROM ADS
680
of 867 total · 78% share
BALANCE TOP-UPS FROM ADS
137
of 203 total · 68% share
[05]KEY INSIGHTS
01Registration is a weak signal — balance top-up is the real one

For a freemium AI platform, a registration tells you nothing about user quality. Balance top-up is the first monetization signal. Building optimization around top-up rate, not registrations, changed what “good traffic” meant across every campaign.

02B2B and B2C cannot share a funnel

Different use cases, different messaging, different conversion logic. API access and team workflows need entirely separate landing paths from personal chat use. Mixing them into one funnel diluted conversion at every stage.

03Intent segmentation makes optimization tractable

20 groups by demand type — API, models, integration, business, B2C — meant that each cluster could be written, pruned, and evaluated independently. Structural precision replaced guesswork about which queries were delivering actual value.

04UTM → Metrica loop turns analytics into optimization

Without tracking through to balance top-up, only traffic volume was visible. Full UTM coverage across all campaigns and groups revealed which exact campaign-group combinations produced paid users — not just visitors.

05Weekly dynamics reveal the effect of structural changes

Comparing week-by-week dynamics in February and March showed that improvements to campaign structure, semantics, and ad copy had a measurable impact on top-up volume — not just on clicks. It made the connection between work and result visible.

[06]OUTCOME

A complex AI platform with mixed audiences was turned into a structured, measurable growth system. Audiences separated. Campaign architecture rebuilt around intent. Analytics extended beyond registration to the first monetization event. The result: 680 registrations and 137 balance top-ups from ads in the tracked period, with weekly dynamics showing improving user quality as the structure matured.

Structure: 2 audience tracks · 5 intent clusters · 20 groups · 1,300 keywords
Analytics: 4 Metrica goals · UTM to balance top-up level
Share of new users from ads: 78% registrations · 68% balance top-ups
Channel: Yandex.Direct Search

Client name used with permission. Report ID is internal. Metrics reflect a specific measurement period — results vary by product maturity, funnel setup, and market conditions.