# Marketing Monitoring

> Detect, investigate, assign, monitor, and resolve meaningful changes across connected marketing data.

Canonical URL: https://marcenta.ai/docs/marketing-data-monitoring
Markdown URL: https://marcenta.ai/site-md/docs/marketing-data-monitoring

Marketing Monitoring continuously reviews supported metrics from connected data sources and turns meaningful changes into incidents. An incident gives your team one place to understand what changed, assess likely causes, examine supporting evidence, coordinate follow-up, and retain a record of the response.

Incidents are investigation signals. They draw attention to changes that may matter, but they do not by themselves prove business impact or establish a root cause.

## Before incidents can be detected [#before-incidents-can-be-detected]

Marcenta needs:

* an active connection for the relevant source;
* enough successfully imported historical data to establish a useful comparison;
* a supported metric and dimension combination; and
* detection processing to run after the source data is available.

A newly connected source may not produce incidents immediately. Initial imports, provider reporting delays, and limited history can all affect when a reliable comparison becomes possible.

## How Marketing Monitoring works [#how-marketing-monitoring-works]

Marketing Monitoring follows this sequence:

1. A supported connection imports data from the source platform.
2. Marcenta organizes supported metrics into daily observations, including an overall source total and eligible dimension values.
3. The system identifies significant dimensions so that monitoring remains focused on segments that materially contribute to performance rather than every low-volume value.
4. After enough history is available, Marcenta compares the latest observation with a robust baseline derived from recent comparable data.
5. A change must be large and statistically meaningful enough for the source's configured sensitivity before it becomes an incident.
6. Marcenta classifies the incident as a regression or opportunity, assigns a severity, and adds available context for investigation.

Monitoring currently evaluates daily data on weekdays. It accounts for the reporting delay of each provider and requires enough usable historical observations before evaluating a metric. This helps avoid treating incomplete recent data or a single unusual historical value as normal performance.

Repeated unusual observations can increase the urgency of an incident. Marcenta also applies minimum movement and volume requirements appropriate to the metric, so a mathematically large percentage change on a negligible value does not necessarily create an incident.

Detection runs during scheduled processing and can also be started with **Run detection** from the Incidents page. Manual detection starts asynchronous processing; results can appear after the initial request completes. If detection is already running for the workspace, Marcenta avoids starting a duplicate concurrent run.

Repeated or closely related signals may be consolidated into a parent incident. The parent appears in the incident list, and its related signals remain available on the detail page.

## Metrics monitored today [#metrics-monitored-today]

Marcenta monitors the following metrics when the relevant connection returns them and they are successfully imported. Availability can depend on the connected account, provider permissions, enabled products, and whether the source contains enough data.

| Source                | Supported metrics                                                                                                                                                                                                                                    |
| --------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Google Analytics 4    | Sessions, users, new users, engaged sessions, engagement rate, bounce rate, average session duration, conversions, and revenue                                                                                                                       |
| Google Search Console | Clicks, impressions, click-through rate, and average position                                                                                                                                                                                        |
| Google Ads            | Clicks, impressions, conversions, spend, click-through rate, and cost per click                                                                                                                                                                      |
| LinkedIn Ads          | Clicks, impressions, sends, opens, total engagements, other engagements, text-ad URL clicks, landing-page clicks, spend, cost in USD, conversions, video views, comments, likes, shares, company-page clicks, click-through rate, and cost per click |
| HubSpot               | Contacts created, deals created, created deal value, deals closed won, deals closed lost, won deal value, meetings booked, meetings completed, and form submissions                                                                                  |
| Salesforce            | Leads created, contacts created, deals created, created deal value, deals closed won, deals closed lost, won deal value, meetings booked, and meetings completed                                                                                     |

A metric appearing elsewhere in Marcenta does not automatically mean that Marketing Monitoring supports it. Reporting availability and monitoring availability are separate because a monitored metric also needs consistent daily observations, comparison logic, and enough history to produce reliable signals.

## Significant dimensions [#significant-dimensions]

A dimension describes the segment behind a metric—for example, a campaign, country, device, landing page, keyword, pipeline stage, or lead source. Monitoring only overall totals can hide an important change inside one segment, while monitoring every possible value can create noise from long-tail or very low-volume data.

Marcenta therefore monitors:

* the overall total for each eligible source and metric; and
* dimension values that are significant enough to represent a meaningful share of activity.

Significance is refreshed as connected data is synchronized. For most sources, Marcenta prioritizes dimension values according to their contribution to traffic or conversions. Paid-media and CRM sources can retain a broader set of useful non-demographic values, while high-cardinality demographic data is narrowed to the values carrying the most activity.

Depending on the source and the data available in the connected account, monitored dimensions can include:

| Source                | Dimensions that can be evaluated                                                                                |
| --------------------- | --------------------------------------------------------------------------------------------------------------- |
| Google Analytics 4    | Source, medium, campaign, channel group, country, region, city, device, and landing page                        |
| Google Search Console | Landing page, search query or keyword, country, and device                                                      |
| Google Ads            | Campaign, ad group, keyword, search term, device, country, source, medium, and channel group                    |
| LinkedIn Ads          | Campaign, ad or creative, job function, job title, seniority, company size, industry, company, and country      |
| HubSpot               | Pipeline, pipeline stage, lead source, owner, configured source properties, campaign, landing page, and channel |
| Salesforce            | Pipeline, pipeline stage, lead source, and owner                                                                |

Not every listed dimension will appear for every workspace. A value must be present in imported source data and meet the applicable significance requirements. A dimension can become eligible or cease to be eligible as its contribution changes over time; this does not affect monitoring of the overall source total.

## Get a metric monitored [#get-a-metric-monitored]

There is currently no separate switch for selecting individual metrics. A supported metric begins participating in monitoring when all of the following are true:

1. Its source is connected and the connection is healthy.
2. The connection has permission to retrieve the relevant data.
3. A completed sync has imported the metric into Marcenta's daily monitoring data.
4. Enough eligible historical observations and meaningful volume are available to establish a baseline.
5. The observation is either an overall source total or belongs to a significant dimension.
6. Scheduled detection runs, or a user selects **Run detection** after the sync completes.

Changing sensitivity controls how readily supported movements create incidents; it does not add an unsupported metric or force a metric with insufficient data to be evaluated.

If a supported metric is not being monitored, verify the source connection, permissions, latest successful sync, provider reporting delay, and available history. If the metric is not in the supported list above—or you need a custom metric—contact Marcenta to request connector and monitoring support. Adding a metric to a report alone does not enable incident detection for it.

## Regressions and opportunities [#regressions-and-opportunities]

Marcenta classifies a detected movement as either:

* **Regression** — the metric moved in an unfavorable direction; or
* **Opportunity** — the metric moved in a favorable direction that may deserve attention or further investment.

Direction depends on the meaning of the metric. An increase in conversions can be favorable, while an increase in cost per click, bounce rate, average search position, or lost impression share can be unfavorable. Do not interpret every increase as an opportunity or every decrease as a regression.

## Severity [#severity]

Severity ranges from **Very low** to **Very high**. It helps prioritize the review queue using the size and significance of the detected movement.

Severity is not a financial-impact estimate. A high-severity change can have limited business impact, while a lower-severity movement in a strategically important segment may still require immediate attention. Consider spend, conversion volume, pipeline impact, business timing, and data quality before deciding what to do.

## Review the incident list [#review-the-incident-list]

Open **Incidents** from the Marcenta navigation. Each row can show:

* severity and data-confirmation state;
* open or resolved status;
* regression or opportunity type;
* a plain-language title;
* metric and affected dimension;
* connected source and observation date; and
* observed and baseline values.

A **+N related** indicator means the incident contains additional related signals. Open the incident to inspect them.

Use the toolbar to:

* search incident titles;
* filter by status, date, source, type, or severity;
* switch between regressions and opportunities;
* show or hide table columns under **View settings**;
* sort or resize supported columns;
* reset active filters; and
* select the number of incidents shown per page.

The available source filter is based on sources represented in the current incident data.

## Run detection manually [#run-detection-manually]

Use **Run detection** when fresh source data has been synchronized and you do not want to wait for the next scheduled detection cycle.

1. Confirm the required connections are healthy and their latest sync has completed.
2. Open **Incidents**.
3. Select **Run detection**.
4. Wait while the detection job processes in the background.
5. Review the refreshed incident list after processing completes.

Running detection does not force a provider sync and does not guarantee a new incident. It analyzes the supported data currently available in Marcenta. Repeatedly running it against unchanged data should not be used as a substitute for fixing a delayed or failed source sync.

## Understand unconfirmed data [#understand-unconfirmed-data]

Some providers revise or finalize recent data after its first import. An incident based on data still inside the source's normal finalization window can show **Unconfirmed**.

Treat an unconfirmed incident as provisional:

1. review the observation and determine whether immediate risk justifies action;
2. compare the result with the source platform;
3. allow the provider's reporting window to complete where appropriate; and
4. return after the next sync and detection run.

The incident can change or be resolved after more complete provider data arrives. Avoid making irreversible budget or tracking changes solely from a provisional signal.

## Read an incident [#read-an-incident]

The incident heading summarizes the metric, direction, percentage change, and affected dimension. The summary metrics show:

| Field    | Meaning                                                |
| -------- | ------------------------------------------------------ |
| Observed | Value recorded for the detected period                 |
| Baseline | Comparison value used to contextualize the observation |
| Change   | Percentage movement from the baseline                  |
| Duration | Time elapsed since the incident was observed           |

The metadata panel identifies the date, dimension, metric, severity, status, assignee, source, reporting grain, and percentage change. Use it to confirm that you are investigating the intended source and segment before acting.

## Use the trendline [#use-the-trendline]

The trendline places the detected observation in historical context. Select the available 7, 14, 30, or 90-period view to determine whether the movement appears isolated, sustained, recurring, or part of a wider trend.

The baseline reference and highlighted anomaly window help distinguish the detected period from surrounding observations. A chart needs sufficient compatible history; **Not enough history to show trend** does not mean the incident record is invalid.

Do not rely on the chart alone for causal conclusions. Check tracking changes, campaign edits, budget changes, seasonality, promotions, website releases, source-platform notices, and business events.

## Review the analysis [#review-the-analysis]

The **Analysis** tab can contain:

* a concise explanation of what happened;
* the most likely cause supported by available context; and
* evidence that supports or limits that interpretation.

AI-generated analysis is an informed interpretation of the available data, not a verified causal finding. Validate material conclusions in the connected source and with the people responsible for the affected campaign, website, or process.

Analysis can be pending while supporting data is retrieved or generated. Missing analysis does not prevent you from reviewing the observed value, baseline, trendline, metadata, or related signals.

## Expand the investigation with Marcenta Agent [#expand-the-investigation-with-marcenta-agent]

Select **Expand investigation** to open Marcenta Agent with the incident context. Use it to ask follow-up questions, challenge an initial explanation, compare likely causes, or identify the evidence needed before action.

Useful questions include:

* “Which segment contributed most to this change?”
* “Could this be caused by incomplete source data?”
* “What should I verify in the provider before changing the campaign?”
* “Summarize the evidence for and against the proposed cause.”
* “What is the lowest-risk next step?”

Keep questions tied to evidence available to the workspace. If the necessary data is not connected or has not synchronized, the agent may not be able to verify the answer.

## Review recommended actions [#review-recommended-actions]

The **Actions** tab presents suggested next steps when available. Actions can include an expected timeframe, impact level, and supporting description.

Treat recommendations as proposals for review. Confirm permissions, ownership, current provider state, and business impact before executing changes. The action list does not automatically modify an advertising, analytics, search, or CRM account.

Select the add-task action beside a recommendation to create a task for continued work. The task and incident have separate lifecycles: creating or completing a task does not automatically acknowledge or resolve the incident.

## Review related signals [#review-related-signals]

The **Related Flags** tab lists signals consolidated under the current parent incident. Each related signal can show its severity, metric, dimension, observed value, and percentage change.

Use related signals to determine whether the issue is isolated or appears across multiple metrics or segments. They provide supporting context; they should not be counted as independent incidents when estimating the size of the review queue.

## Assign a responder [#assign-a-responder]

Use the **Responders** panel to assign or reassign the incident to a workspace user. Assignment identifies who owns the investigation; it does not change the incident status.

When AI Monitoring is available, the incident can instead be assigned to **Marcenta Agent**. The agent reviews the incident during subsequent monitoring cycles and records its observations. Depending on the evidence, it can continue monitoring, resolve a recovered incident, or escalate it for human review.

The **AI Monitoring** tab shows monitoring dates, decisions, confidence, and recorded observations. A monitoring decision remains an analytical judgment and should be reviewed according to your organization's approval and risk policies.

## Acknowledge and resolve [#acknowledge-and-resolve]

Use **Acknowledge** when someone has reviewed the incident and accepted responsibility for investigating it. Acknowledgement indicates active ownership; it does not mean the issue is fixed.

Use **Mark Resolved** when:

* the underlying issue has been corrected;
* the opportunity has been evaluated and closed;
* the signal was explained by an expected business event; or
* updated data shows that no further action is required.

Record the conclusion in a note before resolving important incidents. The current incident page does not provide a reopen action, so verify the outcome before marking it resolved.

## Notes and activity history [#notes-and-activity-history]

The **Activity** panel preserves status changes, assignment events, monitoring updates, and team notes. Add a note when you confirm a cause, contact an owner, make an external change, create follow-up work, or decide that no action is required.

A useful resolution note states:

* what was investigated;
* which evidence was checked;
* the conclusion or root cause;
* what changed, if anything; and
* what should be monitored next.

Do not paste credentials, access tokens, private customer data, or other secrets into incident notes or agent conversations.

## Configure detection sensitivity [#configure-detection-sensitivity]

Open **Settings → Detection** to set the sensitivity for each supported connected source.

| Sensitivity | Expected behavior                                                    |
| ----------- | -------------------------------------------------------------------- |
| Low         | Fewer incidents, focused on larger deviations                        |
| Medium      | Balanced incident volume and sensitivity                             |
| High        | More sensitivity to smaller movements and potentially more incidents |

Sensitivity changes affect future detection behavior. They do not reclassify historical incidents or guarantee that every business-relevant change will be detected.

If incident volume is too high, first check for tracking changes, duplicate imports, incomplete data, or broad source-level disruption. Lower sensitivity only after confirming that the additional signals are not useful. If expected incidents are missing, confirm connection health, data history, supported metrics, and completed detection processing before raising sensitivity.

## Recommended triage sequence [#recommended-triage-sequence]

For a material incident:

1. Confirm the source connection is healthy and recently synchronized.
2. Check whether the incident is marked **Unconfirmed**.
3. Verify the metric, dimension, observation date, baseline, and reporting grain.
4. Review the trendline and related signals.
5. Read the analysis and distinguish evidence from hypotheses.
6. Compare the same period in the original provider.
7. Check tracking, attribution, naming, currency, timezone, campaign, budget, and website changes.
8. Assign a responder or create a task for follow-up.
9. Record the conclusion in the activity history.
10. Acknowledge or resolve the incident as appropriate.

## Troubleshooting [#troubleshooting]

| Symptom                                    | What to check                                                                                                               |
| ------------------------------------------ | --------------------------------------------------------------------------------------------------------------------------- |
| No incidents appear                        | Active source connection, completed initial history, supported metrics, successful sync, and completed detection processing |
| Recent incident is marked Unconfirmed      | Provider finalization window and whether a newer sync is available                                                          |
| Observed value differs from the provider   | Date boundary, timezone, attribution, currency, reporting grain, filters, and provider data revisions                       |
| Too many incidents appear                  | Detection sensitivity, tracking changes, duplicate data, source-wide disruptions, and whether related signals are grouped   |
| Expected incident is missing               | Data availability, baseline history, source support, sensitivity, metric direction, and detection completion                |
| Trendline is unavailable                   | Sufficient compatible history for the selected metric and period                                                            |
| Analysis is missing or pending             | AI processing state, available source context, and whether the incident still has sufficient evidence                       |
| Recommended action is not appropriate      | Validate the source data and business context; record a different conclusion rather than executing it automatically         |
| Manual detection shows no immediate result | Detection runs asynchronously; wait for processing and refresh the list                                                     |

<Callout type="warn">
  Marketing Monitoring identifies changes that deserve investigation. It does not automatically prove causation or authorize changes to spend, targeting, tracking, CRM records, or customer data.
</Callout>