Most of our local government clients have strong in-house analytics and business intelligence teams, who do most of their demand projection and strategy analysis, often making heavy use of Excel, which is a great workhorse tool for many problems. When we start working with them on complex services such as SEND, one of their considerations is whether they really need to work with us using more complex tools, or whether they can do it themselves in-house using their existing skills and resources.

This post explores two main reasons why it is difficult to project future demand for complex services, or analyse how that demand might change in different policy scenarios, using your own team and using standard “in house” tools and approaches.

Complex services need demand simulation, not trend projection

It is very difficult, perhaps impossible, to do simulation models in Excel, and simulation rather than trend analysis is generally needed in complex services. For example, in a typical Special Educational Needs service, there may be six standard categories of need, which can be met in four types of setting. The pupils with these need/setting combinations are of all ages from 0-25, generating hundreds of possible unique combinations of age/need/setting for a pupil. A trend projection method has to either ignore all of these distinctions, and do a reasonably good high-level projection of just a couple of groups – which is not very useful to the SEND leader who needs to plan accurately – or somehow work with these hundreds of subgroups, making the projections extremely unreliable due to the small numbers involved. A trend-based method doesn’t cope with issues such as ‘bumps’ in population which will age through the system over time, nor with the very common situation of only having a couple of years’ history to project from.

All of these challenges suggest that for a complex service, with multiple categories of need and with limited historical data, it makes much more sense to take existing datapoints about the real population, and then simulate the future pathways that each member of that population might take, as well as what individuals might join the service in future. This type of simulation method tends to be more accurate on a macro level, but more importantly can give the detailed breakdown of types of demand that the service head needs in order to make good decisions.

Using external experts brings across ideas and learnings from elsewhere

If you do have the in-house ability to build these kinds of simulation models, then we’d still argue that you should engage with external experts or with peers on some level.
This is for two main reasons:

• Our experience is that the devil really is in the detail, in building models for these complex services with many nuances – implementing an initial model is one thing, but adding in all the safety checks, special cases, and data quality adaptations needed is a lot more work. We’ve been working on our open-source SEND codebase since March 2017, we’ve made 527 updates so far, and we’re still making improvements; we’d encourage any local authority who is building their own model to explore this code and borrow its ideas, rather than having to re-discover all the things we’ve worked through already
• Understanding what sensible projection results might look like, and how they can vary under policy changes, can be difficult without the reference point of other authorities’ experiences; for example, at the moment many authorities’ SEND Further Education figures are undergoing rapid growth due to raised national age limits, so it’s helpful to compare with others what the ‘reasonable’ level of rapid growth might be, vs what is actually more than will happen. Working either directly with peers, or leveraging their experience via external experts, can ensure you reuse as much as possible from others’ experience, as well as setting your own direction.

These two major factors, complexity and need to learn from elsewhere, mean that we usually suggest it makes sense to work with external experts (like us!), and that if you have plenty of technical capacity in-house and want to build it yourself, you should still engage with the wider community to save yourself a lot of effort.