ADmetric AI
Request access

Changelog

What we shipped, week by week

We publish every meaningful change. If something here matters to your team, your account manager can walk you through it on the next review call.

  1. June 26, 2026v3.14Feature

    Autopilot v3 — multi-objective rebalancing

    • Autopilot now optimises against CPA, ROAS and reach simultaneously using a soft-constrained solver.
    • Per-platform floors and ceilings so finance never wakes up to a runaway TikTok line.
    • Explainability drawer on every automated move — see exactly which signal triggered the shift.
  2. June 19, 2026v3.13Improvement

    Finance exports the way your CFO actually wants them

    • Native QuickBooks and Xero exports with VAT line items per platform.
    • Invoice PDFs are now reproducible — re-issue from any historical period without recompute drift.
    • New batch reconciliation view for agencies running 10+ client workspaces.
  3. June 11, 2026v3.12Performance

    Pacing decisions in under 12 minutes, end to end

    • Telemetry pipeline rewritten on a streaming backbone — median signal-to-action latency dropped from 41 to 11 minutes.
    • Connector polling backoff is now adaptive per platform, eliminating Meta rate-limit warnings for high-spend accounts.
  4. June 3, 2026v3.11Security

    SOC 2 Type II audit window opened

    • Type II observation period started June 1, 2026. Type I report available under NDA on request.
    • Audit logs now retained for 13 months on Business and Enterprise; 7 years on Enterprise+ with WORM storage.
    • New per-workspace IP allowlist and step-up auth on any balance-touching action.
  5. May 27, 2026v3.10Feature

    Spotify Ads connector — generally available

    • Full pacing, balance funding and reconciliation parity with Meta, Google, TikTok and Spotify.
    • Audio-specific creative health checks (silence detection, loudness normalisation warnings).
  6. May 20, 2026v3.9Improvement

    Forecast confidence bands you can actually trust

    • Probabilistic forecast switched from naive Holt-Winters to a quantile gradient-boosted model trained per workspace.
    • p10–p90 bands now calibrated weekly against realised spend; backtest error visible in the forecast header.