Scoring Retailer Display Photos with AI
A Power Apps solution that brings in retailer display photos, checks each photo against the campaign rules using Azure OpenAI, and ranks retailers by region so the brand team can choose the winners.
Industry
Time Required
Engagement Model
Solution
Current Phase
Technologies
A leading lubricants company in India runs display contests for its retailers. Field staff visit shops, take photos of the brand’s posters, tags, cans and standees, and send them through the Mobile App as Excel reports. Each photo then has to be checked against the contest rules before winners can be chosen in each region.
This case study explores how Zelite Solutions built a Power Apps solution that scores every retailer photo with Azure OpenAI and ranks retailers by region.
The client needed a solution to address:
- Each retailer can send up to 5 or 6 photos, and the full contest covers about 45,000 retailers and around 1.2 lakh photos.
- Every photo had to be compared by eye with the campaign’s reference images and rules.
- Scores needed to be the same for the same photo, with a clear reason for each one.
- Rules can change during a campaign, so earlier results had to be kept when photos are scored again.
- Photo reports arrive as Excel files, which had to be brought into one system without creating duplicates.
Zelite Solutions built Product Display, a model-driven Power App on Microsoft Dataverse with AI scoring powered by Azure OpenAI. Every campaign is set up by the brand team in the app, and the AI scores photos against whatever that campaign asks for, with no code changes.
- Dynamic Campaign Management: Each campaign has its own creatives, reference images, rules, points, qualifying score and number of winners per region.
- Rules Written in Plain Language: Each rule is a simple instruction for the AI, with its points and whether it is mandatory.
- AI Scoring Against Reference Images: Azure OpenAI compares every photo with the reference images; each photo is checked three times and the majority answer is used.
- Clear Results and Easy Rescoring: Every photo gets a score, a confidence level and a short reason, and a campaign can be rescored with one click.
- Ranking and Review: Qualified retailers are ranked within their region, and the brand team approves winners in the app.
- Time to score one retailer: 3-5 minutes reduced to about 40-60 seconds (2 photos).
- Campaign management: from Excel sheets to one system.
- Photo checking: manual checking by eye removed.
- Rescoring after a rule change: one click, with full history kept.
- Photo reports: picked up automatically from SharePoint, with no manual data entry.
The solution delivered significant improvements for the brand team:
- Consistent Scoring: Each photo is checked against the same rules and reference images, every time.
- Explainable Results: Every score comes with remarks and the photo where each rule was met.
- Traceable Changes: Earlier scores are kept when a campaign is scored again.
- No Manual Data Entry: Excel reports are imported from SharePoint without duplicate records.
- Faster Decisions: The brand team reviews a ranked list per region instead of every photo.
Before, there was no system for running campaigns - everything was handled by hand in Excel. Now it all happens in one app: the brand team sets up each campaign, the photos come in automatically, and the AI scores every photo with a clear reason, so winners can be approved quickly and fairly.
