Lifetimely: Anomaly Detection

Anomaly Detection

At a glance

Anomaly Detection identifies dates when connected ad-platform spend, attributed revenue, or conversions fall outside the range Lifetimely expects for that platform. Use it to prioritize dates and campaigns for investigation—not as proof that a campaign caused a change or that a change was harmful.

Availability and data freshness

Availability and prerequisites: Attribution entitlement, an override, or an active 90-day trial is required; see shared Attribution eligibility. The report needs data from a connected Facebook, Google, or TikTok source to return platform events. Selected ad accounts and campaign exclusions are applied from the store's marketing profile.

Freshness and history: The fallback range starts one year ago and ends yesterday in the store timezone. A remembered report range can replace that fallback.

On this page

• Use the report controls
• Interpret the cards, chart, and table
• Understand the data contract
• Investigate an event
• Resolve missing or different results

Use the report controls

Open Attribution > Anomaly Detection (/attribution/anomaly_detection).

Choose which events to search

Time period changes the dates searched for anomaly events. Choose custom dates or Today, Yesterday, This week (Sun - Today), Last week (Sun - Sat), Last 7 days, Last 28 days, Last 30 days, Last 90 days, Last 6 months, Last 12 months, This year, or Last year. “Last” rolling presets end yesterday; Today and This week (Sun - Today) include today.

The fallback is one year ago through yesterday. Lifetimely remembers a selected time period in the browser and can apply it when this or another report opens. The page does not provide report-level platform, metric, comparison, sensitivity, confidence, or campaign filters.

Inspect one date in the chart

The chart has controls independent of the report-level Time period:

• Date: Select the date around which to request the detail time series. The detail API, rather than the report's selected range, determines how much surrounding history is returned.
• Platform: Choose Facebook, Google, or TikTok. The events API can also return TikTok GMV Max and Combined rows, but the current chart selector does not list those two options.
• Metric: Choose Revenue, Spend, or Conversions. This changes the line, expected range, markers, and the column used to sort a chart-level Drill Down link; it does not change which events appear in the cards, table, or CSV.

Selecting an event card or its eye action updates the chart to that event. If more than one metric is anomalous, the initial chart metric follows this priority: Spend, then Revenue, then Conversions.

Interpret the cards, chart, and table

• Event cards: The page displays at most the first four events returned by the API. Each card shows all three metrics' rounded change from their seven-day averages; anomalous values are emphasized.
• Events table: Each returned event has Date, Platform, Spend, Revenue, Conversions, and actions. Only metrics flagged as anomalous receive a percentage badge. The current table has no search, sorting, or pagination controls.
• Expected-range chart: The line is the selected metric, the shaded band is the API-returned lower-to-upper expected range, and a visible point marker identifies a flagged date. Hover to compare the value and range. The current chart formats every tooltip as currency, including Conversions; treat conversion values as counts despite that display.
• Drill Down: The scan action opens Drill Down in a new tab for the event's exact platform and date, sorted descending by the selected or first flagged metric.
• Export CSV: Export CSV requests the anomaly export with the active report date range, store settings, display currency, and timezone.
• Ask AI: The action receives the active date range and returned anomaly-report data. The currently selected chart date, platform, and metric are not explicitly added to the page context, so name the event in the question when asking about a particular chart.

Data contract: what goes into an anomaly

The report's visible grain is one returned platform-date event. A row contains the platform's spend, conversion value (shown as Revenue), and conversions for that date, their seven-day averages, and a yes/no anomaly result for each metric. An event can therefore contain one or several flagged metrics; unflagged companion metrics remain visible for context.

Where the data comes from

The request applies the current store's display currency, timezone, and shop settings automatically. The UI does not expose these as Anomaly Detection controls. Platform reporting timezone, conversion-date versus click-date assignment, attribution window, modeled conversions, currency conversion timing, account selection, and campaign exclusions can all affect which date and value are returned. See the shared Attribution terminology.

Percentage shown on cards and badges

The display calculation is:

percentage change = (event value − seven-day average) / seven-day average × 100

The UI rounds this result to the nearest whole percentage. If the seven-day average is zero, it displays 0% rather than dividing by zero. This display fallback does not explain how the detection service handles a zero baseline.

For Revenue and Conversions, an anomalous increase is green and a decrease is red. Spend uses the opposite convention: an anomalous increase is red and a decrease is green. These colors are directional shortcuts, not profitability conclusions. A spend increase can be productive, and a revenue increase can still be unprofitable.

Example investigation

Suppose Facebook spend is $1,500 on a date whose seven-day average is $1,000:

($1,500 − $1,000) / $1,000 × 100 = 50%

The card displays a 50% increase in spend. If the API flags it, the value is emphasized in red. Investigate before reducing budget:

1. Select the card and review Revenue and Conversions over the same chart window and expected ranges.
2. Open Drill Down to inspect Facebook campaigns for that exact date, sorted by Spend descending.
3. Check whether a planned promotion, launch, or budget change explains the increase.
4. Compare the date with Overview's Platform Data using the same platform, timezone, currency, and connected accounts.
5. Change budget or tracking only after confirming whether revenue, conversions, and business context support the alert.

Resolve missing or different results

No anomalies are shown

“No anomalies detected” means the API returned no flagged events for the selected period. It does not prove that source data is complete or that campaign performance is healthy. Confirm the date range, connected platform and ad accounts, campaign exclusions, and latest source import.

The percentage does not explain the alert

Cards and badges compare the event with a seven-day average, while the anomaly flag is supplied separately by the detection service. Do not infer the detection threshold from the displayed percentage. A zero baseline displays 0%, and the expected-range algorithm is not documented.

Anomaly Detection does not match Overview or Drill Down

• Compare against Platform Data, not Shopify Data or All Sales. Anomaly Detection uses platform-reported conversion value; Shopify Data uses Shopify Customer Visits and a Lifetimely attribution model.
• Match platform, exact date, display currency, store timezone, selected ad accounts, and campaign exclusions. Platform conversion windows, modeled conversions, and reporting timezones can still produce differences.
• Drill Down is campaign/ad-entity detail, while Anomaly Detection is a platform-date event. A combined event or a date with several campaigns will not necessarily correspond to one Drill Down row.

Next actions

• Inspect the event's campaigns in Drill Down.
• Compare channel totals using the same platform source in Overview.
• Verify selected ad accounts and campaign exclusions.