Risk model to detect fraud

The request was partially successful.

Dear Department for Work and Pensions,

Your 2021-22 accounts, (https://assets.publishing.service.gov.uk...) published on 7 July 2022, refer at paragraph 48 of the Report of the Comptroller and Auditor General (“the report”) (https://www.nao.org.uk/wp-content/upload...) to a “risk model to detect fraud in Universal Credit advances claims” (“the model”).

The model uses a machine learning algorithm to conduct an “analysis” of “information from historical fraud cases” and “predict which cases are likely to be fraudulent in the future”. The Department states that “cases scored as potentially fraudulent by the model are flagged to caseworkers, who then prioritise the review and processing of such cases accordingly".

Please provide the following information on the model:

1. Are all claims that are deemed to ‘look fraudulent’ by the model and so referred substantively investigated for possible error / fraud by a person?
a) Or, does a person first (by way of a preliminary decision-making stage) consider a referral from the model and decide whether or not to undertake a substantive investigation for possible error / fraud?
b) If so, what proportion of claims identified and referred by the model are ultimately substantively investigated?
2. What proportion of claims that are deemed to ‘look fraudulent’ by the model are suspended?
3. A breakdown of claims identified by the model as being potentially fraudulent by the claimant’s form of nationality, immigration status and sex.
4. The proportion of claims that the model identifies as being potentially fraudulent that are subsequently reinstated (i.e. what is the error rate of the model?).

In the event that you determine some of the information I have requested to be exempt from disclosure, please redact exempt information with black boxes, instead of snipping or excerpting, and please state which category of exemption you believe applies to the information.

If it is not possible to provide the information requested due to the cost of compliance limit identified in s.12 FOIA, please prioritise information relating to questions (1) and (3).

Yours faithfully,

Katarzyna Figiel

DWP freedom-of-information-requests, Department for Work and Pensions

Thank you for contacting the Department for Work and Pensions (DWP).
 
This is an automated confirmation that your request for information has
been received by the DWP Freedom of Information mailbox and will be
processed accordingly.
 
Please note that this mailbox is for Freedom of Information (FOI) or
Internal Review (IR) requests only – If you have submitted a non FOI or IR
related email then we may not be able to action it.
 
 
Timescales for responding
 
Freedom of Information (FOI) Requests
 
If your email is a valid FOI request, as per Section 8 of the FOI Act
2000, you can normally expect a response within 20 working days.
 
Internal Review (IR) Requests
 
You can normally expect a response within 20 working days.
 
We will respond to all valid FOI and IR requests within our statutory
timescales, if we are unable to do so then we will contact you to explain
why. Please note that we are not able to process your FOI or IR request in
any shorter timescales that you may indicate.
 
 
Receiving a response from us
 
Please note that email FOI and IR responses will be issued from
[email address]
 
We recommend that you add this address to your email contacts otherwise
the response may be treated as Spam or Junk by your email account
provider.  
 
 
Further Information
 
Should you have any further queries in connection with your FOI or IR
request please contact us.
 
More information on the DWP can be accessed on gov.uk here –
[1]https://www.gov.uk/government/organisati...
 

show quoted sections

References

Visible links
1. https://www.gov.uk/government/organisati...

J Roberts left an annotation ()

Some people may be interested in this:

Kate Osamor: [142793]

'To ask the Secretary of State for Work and Pensions, how many benefits claims suspended by the risk review team have been (a) closed and (b) remain suspended.'

Tom Pursglove:

'As of 11th February 2023, the cumulative total of customer cases suspended by the Risk Review Team is 188,119. Of these, 71,496 have been closed and 109,402 remain suspended. The remaining 7,221 have been reinstated following review.'

https://questions-statements.parliament....

ICO 19/1/23

'As part of this inquiry, we consulted with a range of technical suppliers, a representative sample of local authorities across the country and the Department for Work and Pensions.
...

In this instance, we have not found any evidence to suggest that claimants are subjected to any harms or financial detriment as a result of the use of algorithms or similar technologies in the welfare and social care sector. It is our understanding that there is meaningful human involvement before any final decision is made on benefit entitlement. Many of the providers we spoke with confirmed that the processing is not carried out using AI or machine learning but with what they describe as a simple algorithm to reduce administrative workload, rather than making any decisions of consequence.'

https://ico.org.uk/about-the-ico/media-c...

Related request:

https://www.whatdotheyknow.com/request/r...

freedom-of-information-request@dwp.gov.uk,

1 Attachment

Dear Katarzyna Figiel,

I am writing in response to your request for information, received 24th
February.

Yours sincerely,

DWP Central FoI Team

J Roberts left an annotation ()

The Guardian 3/9/23

'The UK government risks contempt of court unless it improves its response to requests for transparency over the use of artificial intelligence (AI) to vet welfare claims, the information commissioner has said.'

https://www.theguardian.com/politics/202...

This appears to be the related DN (IC-193012-J8T2):

https://ico.org.uk/media/action-weve-tak...

Paragraph 31 spells out what advice the DWP should provide:

'What elements of the request could be complied with within the appropriate limit;

Whether introducing a timeframe would bring the request within the appropriate limit; and

Whether removing the emails from the scope of the request would sufficiently refine the request.'

Q4 relating to emails:

'4. Any reports, presentations, emails providing the results and analysis of the trial in 2021-22 testing its ability to detect fraud in dvances claims already in payment.'