The worry about investing heavily in data and the people to use it is that we create a Rolls Royce engine that is not connected to the “wheels” of delivery.  This seems an odd thing to say as government departments are increasingly concerned to be (or at least say that they are) data-driven. Certainly, agencies that provide services to government increasingly complain about the burden of data collection (which creates an opportunity cost that limits their capacity to provide the service that they are being paid to provide).

Yet research by the Behavioural Insights team (BIT) [1] argues that in fact data is frequently nothing more than a support to the decisions predetermined by heuristics and political pressures, rather than a basis for decision making.  This returns us to the issue of data for insight not data for control purposes.  When data is used as a mechanism for control, metrics are chosen to fit a predetermined frame, and results assume causation.  In the development of policy this leads to confirmation bias.  In delivery, the data derived leads to both group reinforcement and feeds optimism bias.

So how then do we get to a position where data has greater influence on how public services are delivered?  Some of this is about the general lessons of how influence in systems work.  The BIT “Apples” mnemonic is helpful here. Quality of staff, top level support, academic alliances, even physical location, matter.  In particular, understanding the machinery of government is important. 

Beyond these technical elements, political nous and a healthy dose of pragmatism are essential. Advocates of a data driven approaches need to be prepared to demonstrate at a small scale how policy problems can be understood and solutions proposed using data.  In particular, we need to be ready to rigorously evaluate our own solutions and demonstrate where the work and where they don’t (and understand why the failures occurred).

We also need to acknowledge the limitations of a data driven approach.  Policies that are rational on their own terms can fail when they come into contact with political realities, while there are other policies of such symbolic importance in signalling political priority and “direction of travel” that their implementation is a success, even if there is little unambiguous evidence that they achieved their stated aims. 

We need to acknowledge this, but recognising “data isn’t everything” is not an argument that data is of value.  Above all though,  influence is about the ability to exist in different worlds, and recognise where and how the messy realties of life and pristine certainties of data models collide, but still be able to pull out the important and the useful.