The HMRC Connect project was a particularly hot topic because it is an example of a successful ‘Big Data’/Data Analytics system implementation. At a time when HMRC staff have been cut by 40 per cent or more, the Connect application has garnered at least £2 billion in previously lost tax revenues, and figure that is growing by the day.
It has achieved this by ‘blending’ data from 28 different sources - from both within and outside HMRC - and linking together data from these sources that relate to particular entities, such as individual taxpayers and postal addresses. Analyses can be run on these details to detect anomalies that suggest fraud is taking place, such as infeasible numbers of people claiming to be resident at the same small flat. The leads identified are followed up by conventional investigation teams. Not all the leads will reveal wrong-doing, but the information leads to drastic cuts in the time wasted on searching blind alleys.
The concept for the HMRC Connect system was initially tested by the building of a pilot application using data located in Surrey. This study was spectacularly successful, and persuaded the powers that be to extend the approach to the country as a whole. The option of treating the pilot as a throwaway prototype and building a national system from scratch was considered.
However, the pilot system was found to be robust enough to be used as the basis for the broader application. Mike did not explicitly label the approach used as ‘agile’, but it was clearly and incremental and iterative way of working.
Are there broader lessons that can be drawn from this project? Project Eye thinks that, firstly, there are certain conclusions that should NOT be jumped to. It is tempting to compare HMRC Connect with the Universal Credit programme. This programme also ‘blends’ data from different applications - in this case to calculate and consolidate the social security benefits due to the needy. It is true that a key benefit of the Universal Credit application is seen as an improvement in the detection of benefit fraud.
But while for HMRC Connect an imperfect system can still generate beneficial outcomes in the more effective use of the time of HMRC investigators, the main thrust of Universal Credit is the calculation of the correct benefits for families and individuals in need. It is important that here, as far as possible, every automated calculation is as accurate and timely as possible, not just a percentage of them.
Other insights gleaned from this presentation include the way that the technologies deployed were generally existing ones and that the successful implementation depended on collaboration between different specialists including data management experts, statisticians and AI specialists, taxation experts and IT developers.
One thing that Project Eye wonders about is the degree to which the successful projects, particularly agile ones, depend on a pool of domain and technical expertise already existing. Work can start and be progressed quickly where people have a good idea of where they are going.
In the case of HMRC it was not the case that no-one had ever before thought to use data mining to detect anomalies - Project Eye recalls hearing of such an application many years ago at Inland Revenue, which for reasons that are incomprehensible to Project Eye, was known informally as the Welsh Farmers system.
Because of the sensitivity of some of the information relating to this programme, copies of the overheads used at the meeting are not available. However a list of sources of information about HMRC Connect has been provided for those interested in following up the issues raised in this fascinating talk.
Learn more at:
Accounting Web - How HMRC keeps a watchful eye
Computing - Connecting the dots at HMRC
HMRC given power to obtain payment card transaction data from merchant acquirers
Bdaily - Business News - For businesses something new is on the cards - HMRC
