28 Apr, 2026

• Pace's pacing AI now sees how its recent recommendations actually played out. Each time Pace runs a pacing optimization, the prompt now opens with a "your recent track record" section: what was predicted last time, what actually happened to spend, conversions, and ROAS, and whether the prediction was confirmed or contradicted. Over weeks, the AI gets a calibration signal that lets it adjust its own confidence and stop repeating the same misses.
• Predictions are now structured and falsifiable. When Pace makes a budget recommendation, the AI also emits a specific forecast (for example, "expecting ROAS to go up 15–25% over the next 7 days") that gets stored alongside the change. A daily background job measures what actually happened and writes the verdict (confirmed, contradicted, or inconclusive) onto the change. Forecast bands are capped at 30 percentage points, so the AI can't game the loop with vague "between 0% and 50%" hedges.
• Pacing now reads what sparks already knows. When Pace decides what to do with budgets, it sees the durable account facts the agentic-sparks system has been pinning ("this account dips on Mondays," "branded campaigns out-convert generic 3:1") plus the most recent sparks observations for that account. Budget decisions stop re-discovering patterns sparks figured out weeks ago.
• Available across all four platforms. Google Ads, Meta Ads, LinkedIn Ads, and Microsoft Ads pacing all gained the same outcome-aware, sparks-aware prompts in this release.

The longer Pace runs on your account, the sharper its budget calls get.

• Sparks now investigates each account like an analyst. Instead of running one prompt and moving on, the AI pulls a high-level snapshot, drills into specific campaigns and keywords, checks what's changed recently, and only surfaces a finding when something genuinely deserves your attention.
• It remembers what it told you. Sparks tracks every prior observation per account and uses similarity search to recognise when a "new" finding is really a rehash. Fewer repetitive insights, more genuinely fresh ones.
• Stable knowledge compounds over time. Observations that hold up across multiple runs (like "this account dips on Mondays" or "branded campaigns out-convert generic 3:1") now stick as durable account facts that future sparks reference. The longer Pace runs on your account, the sharper its insights get.
• The Sparks tab now shows what Pace has learned. Open any account's Sparks tab and you'll see a "What Pace has learned about this account" panel listing the durable patterns the AI has confirmed, plus a recent-activity strip showing how often Pace checked the account this week and how many checks were quiet (Pace looked but found nothing worth surfacing). Silence now reads as active monitoring instead of absence.
• Each spark shows what it relates to. Open any new finding and you'll see up to three semantically-similar prior sparks the AI considered before writing this one, so you can see at a glance whether a pattern is recurring or genuinely new.