Department has purchased three licences of automated ‘social media listening’ tool that will offers the ability to search for specified terms and flag up thousands of individuals’ posts each day
In a bid to clamp down on benefit fraud, the Department for Work and Pensions has signed a deal for automated technology that can analyse and flag up millions of social posts made by citizens.
On 18 December, the department entered into a one-year deal with Capita for the use of its Cosain software, which is an “AI-driven automated social-media listening and OSINT (open source intelligence) platform”, according to the product’s listing on GOV.UK.
Newly published commercial documents reveal that the tool will be deployed to provide “an open-source search and alert platform to provide the intelligence needed to allow DWP to identify and manage fraud”.
The text of the contract indicates that the DWP will have use of three licences for the product, and each of these will offer “alert monitoring functionality” and will give users the ability to “build their own queries” using up to two terms or definitions from an existing database. For each licensed version, up to 1,000 posts per day for each of the query terms will then be flagged up to anti-fraud officials, the contract states.
Over the course of a full year, this equates to a potential total of 2.2 million alerts being issued for online posts identified by the technology.
As with almost all public sector commercial engagements, the document goes on to provide details of the data-protection implications of the deal.
The contract states that the data subjects affected will be “members of the public” that use social media and that “in theory, any sort of personal data could be collected if it has been posted on an open internet site”. But the expectation is that “the vast majority of personal data will be basic information such as names and usernames”.
1,500
Number of social media accounts DWP flagged for removal between July 2019 and May 2022
2.2 million
Potential number of posts that will be flagged to the department annually by Cosain software
£70m
Planned DWP investment from 2022 to 2025 on ‘advanced analytics to tackle fraud and error’
£1.3bn
Amount the department intends to save this year through anti-fraud measures
“The processing takes place on an ad hoc basis; the client enters the request and, dependant on the information requested, and availability of this information, it is passed onto the client,” the document says. “The intended nature of the data processing is the collection, temporary storage and dissemination to the client of requested data. This data is automatically deleted after 30 days as there is no requirement to keep it for longer than this period.”
PublicTechnology contacted the DWP enquiring as to what information the software might be tasked with looking for – for example, whether it might search for evidence of someone working while claiming out-of-work benefits – and how this would then be used in the context of anti-fraud work.
In the response, the department indicated that its fraud investigators would use all available information in pursuing a case – including public online social posts.
A DWP spokesperson added: “The government has a duty to spend taxpayer’s money responsibly which is why we are cracking down on fraud with a variety of measures to save £1.3bn this year.”
‘Potential sources of evidence’
The department’s 600-page guidance document for fraud investigators suggests that social media sites including Facebook and Twitter “may identify potential sources of evidence” for officials – particularly for cases of fraud related to “periods of absence abroad that the claimant has not disclosed to the DWP for the purposes of assessing benefit entitlement”.
In Fighting Fraud in the Welfare System, a policy paper published in May 2022, the department claimed that, in recent years, it has seen a rise in the use of social platforms for the online sale and sharing of “information on how to defraud public services including the welfare system”.
“It would take years to read just the posts from just one day, therefore manually monitoring all these sources is both inefficient and hugely time-consuming”
Cosain product guidance
“[This] can range from promoting fraudulent approaches, scamming claimants, as well as selling how-to manuals,” the paper says. “There was a surge in the promotion of ways to defraud DWP on social media when coronavirus easements were put in place due to the streamlined application process, and we now see this threat persisting.”
In light of the this surge, the anti-fraud plan said that the department is “working more closely with technology providers and, since July 2019, we have referred approximately 1,500 social media accounts to providers for removal”.
Such close work will be furthered by the deal for the use of the Cosain software – which procurement records suggest represents the first time the DWP has deployed automated tech that can monitor social networks at large scale.
Product guidance created by Capita indicates that the technology was developed with the support of OSINT experts with experience of working in law enforcement – for whose use the software is primarily intended.
“Online social media is full of discussions, opinions and debates across various social media platforms, forums, and blogs, and it would take years to read just the posts from just one day, therefore manually monitoring all these sources is both inefficient and hugely time-consuming,” the guidance adds.
The DWP’s deal for the use of the technology is valued at £77,760 – equating to £25,920 for each of the three annual licences. This figure could double if the department chooses to enact an optional one-year extension.
The department has previously indicated that it plans to spend £70m over a three-year period on using AI and automation to tackle fraud. But civil society groups have criticised the lack of information available on how the technology operates with campaign group the Public Law Project warning that the DWP could be “sleepwalking into an AI disaster”, in this piece written for PublicTechnology.
