Benefits department has worked with core supplier to create a system based on an LLM which can read scans of paper documents and then flag up risks or general themes
The Department for Work and Pensions has deployed artificial intelligence to help review tens of thousands of letters every day and detect potentially vulnerable citizens that may be in need of additional support.
The Whitemail Insights and Vulnerability Scanner is an automated tool that “reads circa 25,000 scanned documents [and] letters received from citizens per day and flags people who may need urgent assistance”, according to an algorithmic transparency record newly published on GOV.UK.
“People are at the centre of this work, and a caseworker will always decide if extra support is needed.”
DWP spokesperson
The document states that the benefits derived from the technology include “reduced time in manually sorting out mail, particularly those that require immediate attention [and] increased transparency in information leading to timely and targeted/specialist support to claimants”.
The AI platform has not replaced any previous manual processes as “the existing business-as-usual casework processes are still in place – this is a new and additional service”.
The tool, which was developed and trained using an – unspecified – LLM, is comprised of two elements, including the Whitemail Insights component. This is designed to scan paper documents and detect “core themes”, allowing DWP staff to direct correspondence to the right specialised benefit team.
The Vulnerability Scanner functionality, meanwhile, “analyses the content and context of ‘Whitemail’, [which is] physical post/mail for which we have no automatic handling rules – loosely everything that is not a DWP-issued form – to identify a potentially vulnerable customer”, the transparency record says.
“The solution also provides the rationale for identifying a vulnerable customer using DWP’s prescribed themes, for example mental health [or] suicide,” the document adds.
Other potential vulnerabilities that the software is designed to look out for include evidence of issues with alcohol or drug abuse, financial hardship, or indicators of self-harm or domestic abuse.
Once post arrives at the department, it is then scanned, with text converted to a “machine-readable format” and personal data automatically redacted.
The document is then first passed through the Vulnerability Scanner which, once it has completed its assessments, “creates an actionable shortlist of potentially vulnerable people” which is then passed on to DWP caseworkers.
“This list includes a reference to the original source scanned image that a human can call up, review and then decide whether an intervention is appropriate or not,” the transparency record says.
The vulnerability referral has “no personal information… included, [but] each scanned document has a unique ID that can be used to trace the customer record in systems separate from the two AI tools”.
The algorithm record says: “Procedurally speaking, DWP staff handling the claim or engaging with a potentially vulnerable customer are expected to assess each document and the situation on a case-by-case basis. The department’s numerous benefit lines follow overarching and claim-specific standards to provide effective service to their customers and these standards are not influenced by inputs from the two tools.”
The document adds: “If the document does not indicate vulnerability, it is then relayed to the Whitemail Insights solution and labelled according to the relevant document category/theme – for example change of address [or] change of bank.”
25,000
Number of scanned paper documents assessed by the AI tools each day
Nine
Number of themes, such as change of address or bank, that the ‘insights’ tool is designed to detect
£49m
Potential value of deal with Accenture to support the ‘Garage’ tech innovation facility
2.5%
Proportion of DWP’s annual tech budget that is now dedicated to AI, according to recent figures
These two examples are one of nine possible themes the insights tool is intended to detect. As well as identifying these topics in each individual piece of correspondence, the system also provides the department with “general volume-based trend information” on themes that dominate the DWP’s postbag at any given time.
In response to enquiries from PublicTechnology, a DWP spokesperson said: “We are using new technologies to deliver the department’s services efficiently and improve customer experience, particularly for those with vulnerabilities. The Whitemail Insights and Vulnerability Scanner helps identify vulnerable customers who may be at risk by scanning paper correspondence, considerably speeding up our ability to act. People are at the centre of this work, and a caseworker will always decide if extra support is needed.”
‘Exhaustive governance’
The tool – details of which come in light of those for a similar tool deployed across the DWP’s online Universal Credit journal service – was developed by the AI and automation-focused Garage initiative on which the department partners with consultancy Accenture. The two parties last year signed a deal potentially worth almost £50m for “DDaT Specialists for the Garage delivery factory and live service teams”, according to online procurement records. The transparency record indicates that “on this piece of work, the developers were mainly Accenture resources, led by an Accenture delivery lead, working under DWP project managers”.
Despite the supplier taking the lead, it is understood that the DWP owns the intellectual property of the software. While the AI system is already fully “performing in production with no interaction by the public or DWP staff”, it is still “undergoing iterative improvements to serve the needs surfacing from end user feedback on an ongoing basis”.
Both the scanner and insights tools are hosted in an Amazon Web Services environment “with no connectivity to or from the internet and [where] all services communicate via private link”, the record says.
The DWP indicated that data gathered by the tool is not retained for any longer than is deemed necessary.
The transparency document concludes: “The tools were subjected to exhaustive governance by various long-standing departmental functions that assessed performance and security readiness, as well as compliance with standards/ethics around service delivery and protection of citizens’ personal information. No key risk was identified at any stage throughout the development and post-live phases.”
