Case Studies
Automating loyalty campaigns with machine learning
Turning receipt processing from a manual marketing operation into a scalable digital workflow
Sector
FMCG and retail marketing
Client
Global consumer brand
Challenge
Cashback and loyalty campaigns required significant manual processing, third-party fees and fragmented consumer data
Solution
Automated campaign platform combining receipt recognition, machine learning, campaign rules and cashback processing
01
The challenge
Cashback campaigns were digital for the customer and manual behind the scenes.
Consumer promotions often look simple.
Buy a product.
Upload a receipt.
Receive a reward.
Behind that experience, brands may still need to:
Collect receipts
Read purchase information
Check campaign eligibility
Validate products
Calculate cashback
Manage exceptions
Process payments
Report on campaign performance
Much of this work was traditionally handled manually or outsourced to third-party campaign platforms.
That created two problems.
The campaigns were expensive to operate.
And valuable consumer and purchasing data often remained with external intermediaries.
The client wanted a more direct and scalable model.
02
What needed to change
The objective was larger than recognising text on a receipt.
The entire campaign operation needed to work together.
The system needed to understand:
Which campaign was active
Which products qualified
What the customer had purchased
What price they had paid
Which reward rules applied
Whether the receipt had already been used
How much cashback was due
Which cases needed manual review
The technology needed to turn an image into an operational decision.
03
What we built
One platform for configuring and running loyalty campaigns.
WeAreBrain helped build a reusable application for consumer loyalty initiatives, including cashback and reward campaigns.
The platform combined:
Campaign configuration
Define the products, rules and reward mechanics for each promotion.
Receipt recognition
Read information from uploaded receipt images.
Machine learning
Recognise and match relevant purchase information.
Cashback calculation
Apply the campaign rules to the products and prices found on the receipt.
Human review
Allow teams to inspect and correct results where the system is uncertain.
Campaign management
Give internal teams visibility over campaign activity and results.
The result was not simply an OCR feature.
It was an end-to-end campaign operation.
1. The consumer uploads a receipt
A participant buys a qualifying product and submits an image of their receipt.
Receipts are difficult source material.
They differ by:
Retailer
Layout
Paper size
Print quality
Product description
Formatting
Image quality
Some are photographed.
Some are scanned.
Some are long rolled receipts.
The system needs to turn all of that variation into structured information.
2. Image recognition reads the purchase
Turn a photograph into usable transaction data.
The platform uses image processing and optical character recognition to analyse uploaded receipts.
Machine-learning technology then helps recognise and interpret the information.
The system can identify relevant purchase data and connect it to the campaign.
The original solution was designed to process receipts from a wide range of retailers and supermarkets.
3. Match products to campaign rules
Reading the receipt is only part of the job.
The system also needs to determine whether the purchase qualifies.
A matching algorithm connects the information found on the receipt to:
Campaign products
Actual purchase prices
Promotion rules
Reward conditions
This allows the platform to calculate the appropriate cashback for the consumer automatically.
4. Handle uncertainty
Automation needs a review path.
Receipt recognition is not perfect.
Poor photographs, damaged paper, unusual retailer formats or ambiguous descriptions can create uncertainty.
The platform therefore includes an internal management interface where employees can review and correct results.
That creates a practical automation model:
High-confidence result
Process automatically.
Uncertain result
Send for review.
Human correction
Use the approved result to continue the campaign workflow.
The aim is not to pretend every case can be automated.
It is to make manual work the exception.
5. Keep campaign rules flexible
Different campaigns have different mechanics.
A reusable platform needs to support variation without requiring a new application every time.
Brand and marketing teams can configure campaign initiatives around different products, rules and rewards.
The technology becomes reusable infrastructure for multiple campaigns rather than a one-off promotional microsite.
6. Own the consumer relationship
Automation also changed the data model.
Using external cashback marketplaces can reduce the direct relationship between the brand and the consumer.
Building the campaign platform internally gave the organisation greater access to first-party campaign information.
That could include insight into:
Participation
Purchasing behaviour
Products
Campaign performance
Consumer engagement
This makes the platform valuable beyond processing receipts.
It becomes part of the brand's consumer-data infrastructure.
04
The workflow
Consumer purchases qualifying product
Receipt is uploaded
Image is processed
Text and purchase data are recognised
Products are matched against campaign rules
Cashback is calculated
Uncertain cases go to human review
Approved reward is processed
Campaign data becomes available to the brand
One workflow replaces several separate manual processes.
05
The result
30,000+ receipts processed across the first three campaigns.
The platform produced measurable improvements during its initial rollout.
30,000+ receipts
More than 30,000 receipts were successfully processed during the first three campaigns.
~80% automatic correct recognition
The system achieved an automatic correct-recognition judgement rate of approximately 80%.
90%+ lower internal processing costs
Internal handling and processing costs were reduced by more than 90%.
Lower third-party costs
Moving the capability in-house also reduced fees and commissions associated with external affiliate and cashback platforms.
These metrics make the business case clear.
The value did not come from using machine learning.
It came from removing manual work from a high-volume process.
06
Why this matters
Marketing automation can go much deeper than email.
Marketing automation is often associated with:
Campaign emails
Lead nurturing
Ad optimisation
CRM journeys
But many marketing operations still contain physical and administrative processes.
Receipts.
Coupons.
Claims.
Rewards.
Product registrations.
Proof of purchase.
Rebates.
These workflows can involve thousands or millions of small transactions.
That makes them good candidates for intelligent automation.
07
The AI was only one part of the system
Receipt recognition required machine learning.
But the complete business solution also needed:
Campaign configuration
Business rules
Image processing
Data matching
Consumer interfaces
Internal tooling
Human review
Reward calculation
Reporting
That distinction matters.
A model can recognise a receipt.
A product turns that capability into a working loyalty programme.
AI services used
AI Agents
Turning receipt submission, recognition, validation and reward calculation into one automated campaign workflow.
Software Engineering
Building the machine-learning, image-processing and software systems behind the platform.
Data Infrastructure
Converting unstructured receipt images into structured consumer and transaction information.
Product Design
Creating simple consumer participation alongside usable internal campaign-management tools.





Still running high-volume customer processes by hand?
The best automation opportunities are often hiding inside repetitive operational work. We build systems that combine AI, business rules and human review to turn those processes into scalable products.