The Software House was asked to develop an AI-powered OCR application that would automate document processing and eliminate the manual data entry bottleneck that was limiting Pension Lab's growth.
Pension Lab had already cut client onboarding from 3 months to 3 days in an earlier project with The Software House. But there was still a major bottleneck: manually processing each document took 15 minutes. With thousands of users waiting to be onboarded, this wasn't sustainable. The company needed AI automation to keep up with their ambition to scale by partnering with major pension providers.
Partnership goal:
→ Build an AI-powered document processing system that would dramatically reduce manual data entry time and allow Pension Lab to scale their onboarding operations efficiently.
Pension Lab
Pension Lab provides a single, unified view of all retirement pension balances that people paid into across their professional careers. Users can track pension schemes from the past, analyze current retirement spending, and consolidate their pensions. Founded by Scott Phillips, the company developed one of the first pension dashboards of this kind and obtained FCA Authorization in 2019.
INDUSTRY
Fintech
COUNTRY
United Kingdom
SERVICE PROVIDED
AI Development
Challenge
Pension Lab was on a rapid growth trajectory, but faced a critical operational bottleneck. Each user's personal or business account information required document processing. Even after an earlier optimization project that introduced new APIs and templates, manually parsing each document still took 15 minutes. With thousands of users to process and more pension providers signing up, the team was spending hours on repetitive data entry.
The challenge wasn't just about speed. Pension Lab needed a solution that could:
-> Process documents in seconds rather than minutes to handle growing user volumes
-> Extract 15-20 fields accurately from various document types and formats
-> Handle Letters of Authorization (LoA) from multiple pension providers, each with different formats
-> Maintain data security and ensure sensitive personal information stayed protected
-> Scale efficiently without requiring proportional growth in manual processing staff
-> Integrate smoothly with existing workflows and systems
Solution
The Software House had been working with Pension Lab for several years and understood their business model deeply. When Pension Lab asked if AI could solve their document processing challenge, the team already had ideas ready.
We kicked off with our GenAI Rapid Prototyping Sprint™ – a workshop process that analyzes business requirements, data considerations, and architecture in the context of AI implementation.
Within one week, we delivered the first Proof of Concept demonstrating that the solution could extract 15-20 document fields in 40 seconds. This POC validated that an AI-powered OCR approach could realistically deliver the 95% time reduction Pension Lab needed.
“Many companies try implementing AI but fail to deliver it or don’t get any real business value from it.
I believe we hit the mark with The Software House on the first try. A 95% process time reduction is our new record.”
Scott Phillips
Founder & CEO at Pension Lab
Process
The Software House assembled a specialized team focused on AI and Python development to build this solution.
1. OCR and AI-Powered Extraction
We built the foundation using Python to create an OCR system that could read scanned documents and extract data. The team evaluated different OpenAI models to find the most cost-efficient option that still delivered the speed and accuracy needed. The winning combination processed documents in 40 seconds while extracting 15-20 fields accurately.
2. Model Fine-Tuning for Different Document Types
Pension providers each use different document formats and Letter of Authorization templates. Rather than forcing a one-size-fits-all approach, we fine-tuned the OpenAI model for each provider's specific document structure.
3. Data Anonymization and Security
Financial services require strict data protection. We built a feature that anonymizes sensitive personal data before it reaches OpenAI's processing layer.
4. Workflow Automation
Beyond extraction, the application needed to fit into Pension Lab's existing processes. We added automated document forwarding and summary generation.
We designed the architecture to be decoupled and flexible, which gives Pension Lab room to refine and expand the text analysis capabilities as their needs evolve. The serverless approach using AWS Lambda and Step Functions means the system scales automatically based on document processing volume without requiring infrastructure management.
Outcome
Pension Lab reduced document processing time from 15 minutes to 40 seconds – a 95% improvement that eliminated their biggest operational bottleneck.
→ 95% reduction in document processing time
→ 15-20 fields extracted automatically in 40 seconds through AI-powered OCR
→ Scalable onboarding infrastructure that allows Pension Lab to partner with major pension providers
