Introducing AI Insights: Anomaly Detection That Explains the 'Why'

Why Anomaly Detection Matters for PPC

If you manage more than a handful of ad accounts, you already know the morning routine: open each platform, scan for anything unusual, cross-reference yesterday's spend against targets, and hope nothing slipped through overnight. It works until it doesn't.

The problem is scale. A single Google Ads account might have dozens of campaigns, each with hundreds of keywords. Multiply that by three platforms and ten client accounts, and you're staring at thousands of data points every morning. Manual monitoring doesn't fail because people are careless — it fails because the volume of data exceeds human attention.

Most teams discover anomalies reactively. A client emails about unexpected charges. A monthly report reveals a campaign that burned through its budget in the first week. A quality score degradation that started six days ago finally shows up as a CPC increase today. By the time you notice, the damage is already done.

Anomaly detection flips this dynamic. Instead of scanning dashboards for problems, the system scans for you and tells you what matters. The question isn't whether something changed — it's whether that change is meaningful enough to act on.

How Blueprint's AI Insights Work

Blueprint runs 16 specialized anomaly detectors against every connected ad account. Each detector focuses on a specific metric category and uses statistical analysis — z-scores, moving averages, and trend detection — to determine whether a data point is genuinely anomalous or just normal variance.

Here's what happens under the hood:

1. Data ingestion: Blueprint syncs performance data from Google Ads, Microsoft Ads, and Meta Ads on a regular schedule. Each sync updates the time-series metrics that feed the detectors.
2. Baseline calculation: For each metric, Blueprint builds a rolling baseline using historical data. This baseline accounts for day-of-week patterns, seasonal trends, and account-specific norms.
3. Anomaly scoring: Each data point is compared against its baseline. Deviations are scored using z-scores — a measure of how many standard deviations a value falls from the mean. Higher z-scores mean higher severity.
4. Severity classification: Anomalies are bucketed into severity levels so you can prioritize at a glance. A 5% CPC fluctuation is noise. A 40% overnight spend spike is critical.

Every detected anomaly generates an insight that includes the anomaly type, severity level, affected account and campaign, a description of what changed, and a recommended next step. These insights flow into your dashboard's Morning Pulse feed, sorted by severity so the most urgent items sit at the top.

The entire process runs locally on Blueprint's infrastructure. No ad account data is sent to third-party AI services during detection. This is statistical analysis, not machine learning — it's fast, predictable, and transparent.

The "AI Explain" Feature

Numbers tell you what happened. They don't always tell you why.

That's the gap "AI Explain" fills. Available as a PRO-tier feature, AI Explain takes a detected anomaly and generates a natural language explanation of the likely cause. It's powered by Gemini 2.5 Flash, chosen for its speed and reasoning quality.

When you click "Explain" on any insight card, Blueprint sends only the relevant context — the metric type, the deviation, the time range, and surrounding campaign data — to generate an explanation. You get a paragraph that reads like advice from a senior PPC strategist, not a data dump.

"Your CPC for the 'Brand — Exact Match' campaign increased 34% on Tuesday. This likely coincides with the new competitor bidding activity detected in your auction insights over the past 72 hours, combined with a Quality Score drop from 8 to 6 on your top three keywords."

AI Explain is entirely optional. The anomaly detectors work without it. If you never click "Explain," no data is sent to any AI model. For teams with strict data handling policies, the local-only detection still delivers full value.

What Gets Detected

Blueprint's 16 detectors cover the metrics that matter most to PPC performance. Here's what they watch:

• Spend anomalies — Sudden increases or decreases in daily or campaign-level spend that deviate from historical patterns.
• CPC spikes — Cost-per-click jumps that signal competitive pressure, bid strategy issues, or quality score degradation.
• CTR drops — Click-through rate declines that may indicate ad fatigue, audience mismatch, or creative degradation.
• Conversion rate changes — Shifts in conversion rate that point to landing page issues, tracking problems, or audience quality changes.
• Impression share shifts — Lost impression share due to budget or rank, helping you understand visibility trends before they impact revenue.
• Quality score degradation — Keyword-level quality score drops tracked over time, catching erosion before it compounds into higher CPCs.
• Budget depletion rate — Campaigns spending faster than expected relative to the remaining budget period, an early warning for month-end overspend.
• Cost-per-conversion outliers — Individual campaigns or ad groups where cost-per-conversion has moved significantly outside normal range.

Each detector runs independently, which means a single campaign can trigger multiple insights simultaneously. If your CPC spikes and your quality score drops on the same keyword set, you'll see both — and the correlation will be obvious.

Built for Privacy

We built AI Insights with a privacy-first architecture because we know how sensitive ad account data is. Here's the design:

• Local analysis by default: All 16 anomaly detectors run on Blueprint's infrastructure using statistical methods. No ad account data is sent to external AI services during the detection phase.
• Optional AI, explicit consent: The "AI Explain" feature only activates when a user clicks "Explain" on a specific insight. It's a deliberate action, not a background process.
• Minimal context sent: When AI Explain is triggered, only the relevant anomaly context is sent — not your full account data, not your client list, not your billing information.
• No data used for training: Blueprint's agreement with Gemini ensures that your data is not used to train or improve AI models.

Your ad data stays on Blueprint. AI is an assistant you invite in, not a surveillance layer running in the background.