# Improving Accuracy in a Marketing Event Pipeline

> TechSupport Team supported AdSignal’s Kafka, dbt and BigQuery marketing attribution pipeline. Over four months, the client’s BigQuery records show event accuracy rising from 87% to 99.2%. Pipeline lag fell from 22 minutes to 90 seconds in Datadog.

- Publisher: TechSupport Team
- Canonical page: https://techsupport-team.com/case-studies/marketing-attribution-pipeline-accuracy/
- Updated: 2026-11-16

Client: AdSignal · SaaS & B2B Software · US marketing attribution provider

## Client and engagement facts

- **Client:** AdSignal
- **Industry:** SaaS & B2B Software
- **Business context:** US marketing attribution provider
- **Delivery model:** Data engineering support
- **Application:** Marketing attribution with multi-source event pipeline
- **Supplied geography:** US

## Application stack and tools

### Application stack

- Python
- Kafka
- BigQuery
- dbt
- GCP

### Tools used

- dbt
- BigQuery
- Datadog
- Jira
- Google Ads API

Related services: [Application support](https://techsupport-team.com/services/application-support/)

## Why Support Was Needed

Campaign reports disagreed with Google Ads. The engineers traced the mismatch to upstream schema drift, dropped events during Kafka rebalances and a deduplication join that mishandled nulls.

## Technical Work

- Rebuilt the dbt test suite to detect four schema drift patterns.
- Fixed dropped events during Kafka consumer rebalances.
- Changed the BigQuery deduplication join to handle null values.

## Representative Incident

A downstream customer changed its tracking setup mid-engagement. The engineers recorded the impact and worked around the change in the pipeline.

## Results and how they were measured

Table: Reported results and their measurement scope

| Metric | Baseline | Result | Units | Measurement window | Source and scope |
| --- | --- | --- | --- | --- | --- |
| Event accuracy | 87% | 99.2% | % | 4 months | Client BigQuery |
| Pipeline lag | 22 min | 90 s | Baseline: min; Result: s | 4 months | Client Datadog |
| Discrepancy tickets | 31/mo | 2/mo | /mo | 4 months | Client Jira |

## Evidence and Limitations

The four-month comparison excludes one customer-side tracking change. The case records a two-week parallel run before the production switch.

## Parallel Validation Before Cutover

The engineers compared the new pipeline with the Google Ads API in a two-week parallel run before switching production.

## Scope a Similar Engagement

Name the event sources that disagree and how you currently measure accuracy and freshness. [Request a support plan](https://techsupport-team.com/contact/) or explore [application support](https://techsupport-team.com/services/application-support/).

## Related pages

- [Application support](https://techsupport-team.com/services/application-support/): The related service for this engagement.

## Contact

Request engineers from TechSupport Team or ask a question: https://techsupport-team.com/contact/
