Mr. Michael Nolan:

The Deputy is 100% right. Most of our schemes come in on time and either on budget or under budget. In the first ten years of the organisation's existence, when we were still called the National Roads Authority, NRA, we were coming from a standing start and a low base. We did not have a stock of learning. We did not have a throughput or a portfolio of schemes to have learned from. By the early 2000s we had turned that around. In the first ten years, from 1994 until 2003 or 2004, we were in tune with international comparators. Our schemes generally ran 15% to 25% over budget. There were many such schemes in those days. Since then we have capitalised on those lessons learned and implemented new processes and ways to do work. We now generally come in approximately 10% under budget. It must be remembered, however, that we carry out schemes in portfolios. We do not insist that every scheme come in under budget. If budgets get too high, we have to keep them down.

We came up with this probabilistic kind of forecasting model for costs and cost ranges. We use outturn costs rather than estimated costs. We have built up a database of outturn costs. We also consider the base costs. Our engineering advisers and local authorities help us with those costs. We carry out a quantitative risk analysis of that and end up with a value of P50. This means that there is a 50% probability that a scheme will come in under budget, but also a 50% probability that it will come in over budget. We then carry out an analysis at P80. This higher figure means that there is an 80% chance of coming in under budget and a 20% change of going over. We test those two scenarios against low growth and high growth in demand. We draw up a matrix of low growth against high cost and high growth against low cost. We do all that when developing the business case. We test it. It is very probabilistic. This is the way such things are done internationally.

We also use a hybrid type of reference task forecasting. This is used in the UK, Australia, Switzerland and Denmark. We use the structured consistent data we have accumulated from previous outcomes. That works well in most cases. We have actually developed confidence curves, again using measures of 50% and 80%. We have used a hybrid form of that process and we are now going to formalise this reference task forecasting over the coming years. This is recognised internationally as another way of looking at projects. We build up the rates from our own inside perspective and then look at it from the outside and take account of all the unknown unknowns we can apply to a project. When one has a long history of delivering projects, one accumulates a large amount of data, which is really valuable.

Bespoke schemes like Dunkettle are complex. This is the only scheme like it we have done, so we do not have a long run of examples to leverage. That results in a higher level of risk. We have had a good run. We do not desire that every scheme come in under budget. If they did, it would mean that we had too much money reserved for those schemes, money which could be used for other projects. We bring it down a bit. We are happy if eight schemes in a portfolio come in on or under budget. The two that run over will be balanced out by the eight.

That is fine in a portfolio situation. That is what is used internationally and what we have been using for a long time with the National Roads Authority and TII and RPS, who all come together.