Decisions over dashboards
A chart that doesn't change a decision is noise. We measure the product by how often a recommendation is approved and how much rolled-back budget it reclaims.
— About
Performance intelligence, not another dashboard.
Most marketing tools end at the chart. Prooflytics starts there: explains what changed, recommends what to do, and writes back to the platform when you click Apply. Built for in-house marketing teams who are tired of reconciling four tabs every Monday.
Why we're building this
The job of a head of performance has shifted from running ads to explaining ads to the rest of the business.Two-thirds of the week goes into reconciling Google's "conversions", Meta's "results", the CRM's "revenue", and writing the same Monday email about why ROAS dropped.
Existing tools — Looker dashboards, agency reports, attribution platforms — produce charts. None of them write the email. None of them push the budget shift. None of them can tell you whether a creative is fatiguing or just having a slow week.
Prooflytics is the layer that produces the answer, not the chart. It reads the same data the dashboard does, classifies it through built-in expertise (creative lifecycle, attribution windows, anomaly diagnosis), and emits a daily brief at 06:00 — with one-click actions to apply.
We're a B2B SaaS for in-house marketing teams in the EU and US — performance specialists, growth leads, CMOs of mid-market companies running 6–8-figure ad budgets across Google Ads, Meta, LinkedIn, TikTok, GA4, and a CRM. Pay-per-service, four modules, bundle when ready.
Product principles
Show the work
Every recommendation is fact / interpretation / action / consequence. No black-box "AI says so". The user can challenge the interpretation and we re-explain.
Two front-rows, one source of truth
CMO and channel specialist look at the same data through different surfaces — a brief and a control panel. We refuse the trade-off where one role gets a worse view.
Read-only until you say so
OAuth read-only by default. Write actions are explicit, per-recommendation, with rollback. We don't auto-execute on your accounts. Trust is a default, not a setting.
No fine-tuning for memory
Agent learning is RAG over Postgres + pgvector, scoped per tenant. You can read every rule the agent follows. No magical model that "knows your business" — explicit memory you control.
Where we are
MVP — Campaign Intelligence and Weekly Performance Update. Both shipping with real customers. HADI Hypothesis Sprints and Competitor Intel are in active development.
Build is solo with a small ring of design partners providing weekly feedback. We talk to 15–20 heads of marketing in EU/US edtech, SaaS, and DTC every quarter. The roadmap is dictated by what design partners ask for next, not by intuition.
Want to be a design partner?
We trade free access for weekly 30-min calls. Limited slots. Email with your stack and what you'd want fixed first.