AI Creative Quality Tracking for a Global Marketing Agency
DBB Software extended a global digital marketing agency's existing creative analytics platform with impression-weighted quality tracking, period comparisons, and account-to-creative reporting.
Industry
Technology
Service
AI Development
Team
1 Full-Stack Engineer
Project State
Jul 2026 - Ongoing
Country
Not disclosed
About the Client
The client is a global digital marketing agency. Its client portal brings the agency's analytics and AI products together on Google Cloud, with BigQuery as the data warehouse. One of the portal's apps is a creative analytics platform that scores video ad creatives, mostly for YouTube, and links each creative to its delivery data from DV360 and Google Ads.
The Client’s Initial Request
The agency wanted to integrate its existing creative-analysis method into the client portal, replacing standalone analysis and the manual exports behind each readout.
Historical Creative Quality Tracking
Track impression-weighted creative scores over selected periods, compare historical performance, and account for differences in creative formats.
01
Campaign Drill-Down and Export
Analyze scores from account to campaign, ad set, and individual creative, with component-level breakdowns and matching CSV exports.
02
Reporting Performance
Make the reporting screens fast enough to use during live client discussions.
03
Account-Level Data Isolation
Restrict access using Firebase authentication, account-level isolation, least-privilege service accounts, and signed URLs for creative media.
04
AI Assistant Access
Provide AI assistants with controlled access to the platform's creative analytics data.
05
Solutions we delivered
DBB Software integrated the agency's existing creative-analysis engine into its portal and developed the analytics, reporting, and data-access features around it.
AI Creative Scoring Integration
Our engineer integrated Gemini-based video creative scoring into the platform using Vertex AI.
Scores follow Google's ABCD framework and 5 creative quality dimensions: visual, sound, action, emotion, and relevance.
The analysis also extracts on-screen text, voiceover, and visual features from each creative.
Scoring a single video takes approximately 34 seconds, requiring adjustments to execution timeouts and parallelism.
Impression-Weighted Daily Scores
We built daily aggregate scores that combine AI-generated creative ratings with actual advertising impressions.
Delivery data comes from DV360 and Google Ads at creative, campaign, ad set, and day level.
Scores are weighted by impressions, with more frequently displayed creatives contributing more to the aggregate score.
A nightly job generates daily scores for accounts, campaigns, ad sets, and creative groups. Historical scores are preserved as recorded.
Historical Period Comparisons
The reporting interface supports comparisons across custom date ranges and predefined periods, including last month, last quarter, and year to date.
Users can compare scores against the previous period or the corresponding period in the previous year.
Custom calculations align periods day by day, including leap years, and display differences alongside historical scores.
Account-to-Creative Reporting
We developed reporting that lets users examine scores from account level down to individual creatives.
Each level includes a breakdown table for the next level, following the account - campaign - ad set - creative hierarchy.
CSV exports match the calculations displayed in the charts.
Separate chart lines for individual scoring components have been built and are in review.
Portal Performance Optimization
Performance improvements focused on the main portal view and its data retrieval processes.
Reduced the main view from 85 requests to 8.
Retained complete creative datasets because rankings and summaries require the full analysis population rather than paginated subsets. The full load of these lists went from 782 KB to 342 KB.
The Cloud Run API uses parameterized BigQuery queries and short-lived caching for repeated requests.
Private Media and Account Access
We implemented private storage for creative media, with access provided through signed URLs. Additionally, Firebase authentication and account-level scoping restrict users to their own account's data.
AI Assistant Access via MCP
Our engineer developed an MCP server that allows AI assistants to access the platform's creative analytics data. The implementation includes an AI access page, tool listing, and token management with scoped, revocable access. This functionality is built and currently in review.
Results Achieved
4.6-second main view load
Reduced from 15.2 seconds, with requests dropping from 85 to 8 and other tabs loading faster.
486 validation checks, 0 mismatches
Score figures matched an independent recalculation from the raw tables in every check.
29 videos scored for under $1
A complete creative corpus was processed using Gemini on Vertex AI, with the figure covering model fees only.
Daily creative quality tracking
Impression-weighted scores support historical comparisons and drill-down from account level to individual creatives.
Add AI Scoring to Your Analytics Product
DBB Software builds AI scoring and analytics features into existing products, with the figures checked against the source data.
Contact Us
"Most of our work starts with a 30-minute call where someone describes a product they're trying to ship and one part of the engineering picture they can't get around.
If that's where you are, let's set one up; I'll tell you straight whether we're the right fit.”
Mina Morkos
Business Development Manager
Want a similar outcome for your team?
Ask our AI assistant. It can pull related case studies, talk through the approach, and put you in touch with the team if you want a deeper conversation.