Scale fail: when modelling is crucial (and surprising) | The Dyson Blog

My last blog on scale took me on a journey from Mitochondria - the microscopic batteries of life - to human evolution, to railways and to factories.

This blog will focus on the latter and particularly with the size and scale of unit operations in production plants. I will reiterate here that I remain highly curious that most economic discussions focus on economies of scale and thus constantly encourage scale increase and growth. They talk little about the diseconomies of scale – a potentially dangerous blind-spot. In all systems there are performance curves which demonstrate operational ranges where the most beneficial outcomes are achieved, these are always constrained at both ends. As density of traffic increase on a motorway, the volume throughput of cars increases. Too high, and the car throughput spectacularly dives – something commuters know all too well. If we take one or two paracetamol tablets our pain is lessened, ten and we suffer potentially fatal liver damage. Too many cooks spoil the broth is a colloquial statement of the same system understanding.

In an area I know a little about and with a degree of understanding of the economics, I decided to create a simple economic model to test this idea.

I modelled how the scale of unit operations would change the overall financial return (NPV) of a portfolio of new pharmaceutical products. I started with the basis that each new product would follow one of four trajectories of sales or demand. The first would be that it hit a base demand profile, the second 1.5 x base demand, the third 0.5 x base demand and the fourth did not get launched or sold at all or 0. The demand curves were normalised curves.

To simply represent a portfolio of products I allocated a probability to each of the four trajectories.

Then I looked at scale. I used a few API (chemical synthesis) projects to normalise a base scale which I named [1]. I then looked at relative scales around this. The smallest being [0.2], the largest being [1.5]. I used standard factors to adjust the capital cost in a non-linear fashion i.e. halving the scale does not half the cost or time for each unit to be deployed; the capital per unit volume actually increases.

Lining up scale with demand:

Finally, I changed the conversion costs for each scale based upon rules of thumb factors. More batches mean more operator interventions, more laboratory tests, thus multiplying the variable cost component of conversion cost.

Having set up the fundamentals of the model I needed to run a simulation which allowed each scale to be applied to each of the demand trajectories; setting when investment decisions would need to be made to ensure product supply did not fall behind the demand. This gave an investment profile for each scenario. To do this I needed help from an expert, Stamos Stamatiadis at Bryden Wood.

Before I reveal the results, which you have probably already looked at, I must make a few comments. Although the model covers several key aspects of the implications of economies of scale, it is simple compared to the actual complexities of decision making in the industry. Each product and portfolio and company is different. What I find interesting,though, is it confirms how production unit scale in some circumstances could have very significant impacts on financial returns over the lifetime of a portfolio of products and where there appears to be little impact. It is also worth saying that the results have not been modelled where there are lower risk profiles than those of bringing new pharmaceutical products to market.

Let me start with a first curve. Modelled on a scenario where the revenue value of the new products was very high (‘Blockbuster Drugs’) you can see that scale has little relative influence over the NPV at different scales. There is a difference, smaller scale has a lower return and somewhere the scale can be optimised, but changes are relatively small and would suggest that sticking with an installed or familiar scale may make sense, if you are confident that most of your products will be blockbusters. There is a natural upper limit to the scale, and this maybe should be understood to prevent “traffic-jam” consequences.

Looking at low value products, scale appears to be potentially critical to securing a positive cashflowover the longer term. This makes sense as over-investing in larger scale, when the potential returns are low, would be hugely detrimental.

May be the middle ground is more representative of the real situation that companies face, though in reality there is a mix. This graph paints the picture that scale has financial significance.

The model does not consider that smaller scale can be made more efficient through step changes in automation which are already being developed. Equally, it does not account for the fact that making a larger number of repeat units itself leverages economies of scale and the unit price for small scale could well reduce over time. This is the dream of new nuclear, where the aim is that smaller multiple units will deliver both a time and cost improvement.

The conclusion for me is that scale absolutely matters and it needs to be carefully and thoroughly considered and modelled not just for one product but for the unpredictable portfolio of products that will come. Without this, the reality is that billions of dollars and pounds of cash will be lost. In an era of changing product platforms and portfolio this will impact product viability, accessibility and return on investment all with significant socio-economic impacts.

Humans came to prominence because they were a smaller and significantly more adaptive, there is a lesson in here somewhere.