ANONYMOUS MAIL SERVICE

AI-POWERED PRIVACY PROTECTION FOR CHILDREN'S LETTERS

Client

We were approached by a software development lead who was one of the most well-known and largest mail carriers in the world. They had developed and been running every year a cherished holiday initiative that allows children to send Santa letters, which were ultimately fulfilled by generous donors who would send them the requested gifts. Every day, a team of “flaggers” on the carrier's side would sort through hundreds of letters containing kids' wishes and personal stories. The flaggers would read the letters and flag any personally identifiable information (PII) that had to be covered up to protect the child’s identity.

Protecting children's information was a top goal within their request but they also wanted to see how they could use Artificial Intelligence and OCR technology to automate and detect some of this PII in the children’s letters. As their data protection officer stated, "Protecting this information is not just our job; it's our solemn responsibility." They required a solution that could meet their rigorous standards while at the same time allow them to process an enormous volume of handwritten letters from children that could include drawings as well.

Challenge

The client faced significant challenges with their annual letter program:

Privacy Concerns: They had to make sure that the protection of children's personally identifiable information (PII) was in compliance with regulations and details didn’ slip out to the public.
Manual Review Processing: Their existing process of manually reviewing each letter was slow, labor-intensive, and prone to errors.
Scalability: The increasing volume of letters each year made it difficult to process them efficiently.
Rising Costs: The expense of employing seasonal workers was growing annually as they were spending upwards of $440k in seasonal salaries alone.

The manual process of identifying and flagging personal information in kid’s handwriting was painfully slow and inefficient. People were making mistakes, and each one could expose a child's private information or the carrier to unwanted exposure. As the influx of content grew, the existing system struggled to keep up.

The client asked if it was possible to design a system that was capable of handling various types of personal information related to children and making sure there was a strict adherence to data protection laws. All these had to be achieved without altering the original content so the Xmas meaning of the request stayed intact.

Strategy

We developed a comprehensive solution that combined innovation and effectiveness. We approached the problem from a few different angles, using cutting-edge OCR models, clever image processing algorithms, and AI-powered PII detection.

Our tech toolbox included:

Advanced OCR models trained on hundreds of thousands of handwritten samples
Image processing algorithms that could spot personal info in “fonts, ”photos and drawings
AI and machine learning models for scanning every document for personal info
We didn't just use these technologies off the shelf. We tweaked, tuned, and sometimes completely overhauled them to fit our specific needs. Our AI got smarter with every letter it processed, learning from its mistakes and getting better at spotting tricky bits of personal information and making the life of the reviewer that much easier.

Unique features of our system included:

Real-time PII detection and flagging
Adaptive learning
Handling diverse content
"PII Shield" - an automated redaction tool

/ ANONYMOUS MAIL SERVICE

Solutions

Technologies

PHP
Python
Go
OCR
AI
Machine Learning
Image Processing
Cloud Computing

Our solution addressed each of the client's challenges:

Privacy Protection: We implemented advanced AI and machine learning models to accurately detect and flag PII in both handwritten and image content.
Automated Processing: Our system replaced the manual review process with an efficient, AI-driven solution that could process letters at a much faster rate.
Scalability: The cloud-based system was designed to handle increasing volumes of letters, easily scaling up during peak seasons.
Cost Reduction: By automating the process, we significantly reduced the need for seasonal workers, cutting operational costs.

There was a secure bridge between the old and new systems, that made sure the data flowed smoothly without any hiccups.

Challenges we encountered and overcame included:

Deciphering varied handwriting styles
Identifying personal information shared in unexpected ways
Handling large volume spikes during busy seasons

Each challenge pushed us to think creatively and dig deeper into our tech toolbox.

Results

Our cutting-edge solution gave the client the ability to manage sensitive information related to children and dramatically cut their seasonal overhead costs around this program. The results were remarkable:

Accuracy rate increased from 95% to 98%, significantly reducing the risk of exposing children's personal data and making far less work to review the letters that were flagged.
Processing speed increased from 100 letters per hour to over 3000 that could be reviewed and then processed once a flag was detected.
The AI agent system caught and flagged 99.9% of potential PII, up from 92% with the old manual process and we trained it to remember what to look for and catch so it could remember this going forward
Operational costs were reduced by 80%.
More letters could be processed than before.

We learned a bunch about how to glue different AI solutions together and how to use AI agentic networks to automate manual work. The system continues to learn and improve, providing long-term value and growth support.

Review

“Through the innovative deployment of our Wippy AI system for this customer, we've transformed their annual children's letter program and protected their data while dramatically reducing costs. The solution expanded the program's reach, allowing the customer to fulfill more children's wishes efficiently during the holiday season.“