Hymans Robertson partnered with UIC Digital to harness the power of AI to solve one of pension administration’s most complex challenges, extracting and verifying key data buried within decades of unstructured member documents. The resulting AI-driven platform uses machine learning to read, interpret and cross-reference millions of records, enabling faster, more accurate data validation and transforming a previously manual, time-intensive process into an intelligent, scalable workflow.

(Data shown is for demonstration purposes only)

Anticipated Results:

• 90% reduction in manual effort

• 10x faster data discovery and validation

• Multi-million £ potential cost savings anticipated on large-scale cleanses

• Full compliance readiness for upcoming 2026 regulations

• Higher staff confidence and significantly less rework

 

The Challenge

Hymans Robertson, one of the UK’s leading pension and investment consultancies, administers over 200,000 pension scheme members across more than 70 clients. For decades, structured member data in their UPM system has been accompanied by millions of unstructured “backfiles” — scanned letters, forms, and PDFs dating back 40 years or more.
Historically, every backfile review was a manual process. Each record could take 15–45 minutes to review, with administrators opening hundreds of documents in search of a single missing or unverified data point. The work was described internally as “a very, very boring job” — repetitive, error-prone, and costly to clients.
The additional challenge is that new regulations, due to take effect in October 2026 required that member data be accurate and evidenced at all times, not only at the point of retirement. Current ,manual methods cannot scale to meet that obligation.

 

The Solution

Following a successful proof of concept that demonstrated the potential of AI to interpret unstructured pension documents, Hymans Robertson commissioned UIC Digital to deliver a full production system: the TPA AI Backfile platform.
The platform automatically reads, interprets and indexes key data from backfiles associated with each pension scheme member. Using OCR and large-language-model analysis, it identifies data such as dates, salaries, contributions and spouse details, highlights their exact location within the original document, and presents them in a clear, auditable interface.
UIC provided end-to-end delivery, designing and engineering both the backend AI pipeline and frontend application. The front end was built using Hymans Robertson’s existing in-house design system to ensure consistency with their digital estate, while UIC focused on the user experience design and technical architecture.

(Data shown is for demonstration purposes only)

 

 

How We Did It

1. Proof of Concept and Validation
UIC first ran a comparative proof of concept against multiple vendors, demonstrating that its AI model could not only recognise the relevant data items but also maintain a verifiable link back to the original page source. The clarity of this approach was decisive in Hymans Robertson’s selection of UIC as delivery partner.

2. Collaborative Design in Glasgow
A joint discovery and design workshop was held at Hymans Robertson’s Glasgow headquarters, bringing together members of the Data Journey and Administration teams; the “Doers” and “Checkers” who perform data cleanses day-to-day. Their insights into workflow pain points shaped the interface and defined practical success criteria: clarity, traceability, and reduced cognitive load.

3. Building the Platform
UIC’s multidisciplinary team engineered the full stack solution, integrating AI document processing, a secure Azure-hosted backend, and a responsive web interface. Authentication was implemented via Hymans’ single sign-on, with role-based permissions.
The system converts unstructured backfiles into a machine-readable form, surfaces key data points, and enables users to export verified data directly into structured outputs ready for upload into UPM. Each extracted value is accompanied by a direct link back to the document location it came from, creating a transparent audit trail.

4. Designed for Real-World Workflows
Rather than replacing human expertise, the tool assists it. The platform supports the established Doer/Checker process, allowing users to comment, review, and evidence findings collaboratively within a single environment. Future releases will extend this model to enable further automation and scaling across other schemes.

 

The Outcome

By combining AI analysis with thoughtful UX and rigorous engineering, the TPA AI Backfile platform transforms one of Hymans Robertson’s most time-consuming operational tasks into a streamlined, verifiable process.

• Early results from the first deployment anticipate:
• Record review times reduced from 30 minutes to around 3 minutes
• Cleansing cycles shortened from four months to six weeks
• Significant reduction in rework and human error
• A consistent, auditable data trail that strengthens regulatory assurance

Beyond efficiency gains, the platform also introduces a new standard of clarity and control, enabling Hymans Robertson to deliver faster, more reliable data-enhancement projects for its pension-scheme clients, while positioning itself as a market leader in AI-assisted pension administration.