Page 71 - Reforming Benefits Decision-Making -(updated - August 2021)

in relation to fraud detection and prevention. The DWP should disclose
the data inputs and parameters of the system, and where possible, the
algorithms themselves, but at a minimum the criteria or indicators used
by the system. The DWP should also be clear about the constraints of
automated systems and what is, and is not, technologically possible. The
DWP should work with civil society organisations in order to establish
the most useful way to publish and present this information.
201
Standards, quality control, training and oversight

2.93 Throughout our research and consultations with users of the system and their
advisers and with the DWP themselves, it became apparent that in relation to
many issues that we identified there is a significant disconnect between what
DWP policy or ‘best practice’ is on the one hand and claimants’ experience of
benefits decision-making, on the other. In addition, there does not appear to
be a process or system in place to understand why decisions are frequently
being overturned on appeal and to implement changes as a result of lessons
learned.

Quality control, standards and oversight

2.94 The DWP has a Quality Assurance Framework which is a three-tier system –
the first two tiers report internally and the third to the NAO. It states that the
framework ensures that decisions made are legal, payments made are accurate
and claimants are notified of their entitlements and responsibilities and their
appeal rights. Tier 1 seeks to identify errors in individual decisions, tier 2
evaluates the effectiveness of processes and tier 3 evaluates whether processes
have resulted in reductions to levels of loss in DWP benefits through official

201 In this context the DWP should have regard to the Information Commissioner’s Guidance on
automated decision-making. Although the guidance applies to decision-making by automated means
without any human involvement, which is not something the DWP currently does, the principles and
practices in the guidance may be helpful in addressing this recommendation. For example using visuals
to explain what information is collected and why it is relevant to the process and having a set of ethical
principles to help build trust in the process (Information Commissioner’s Office, ‘Rights related to
automated decision making including profiling’, Guide to the General Data Protection Regulation
(GDPR)

62