Applying optimization algorithms to profits

As a technology company, we like to apply technical solutions to problems at all levels. Our board of directors even apply technical solutions to the problem of company direction.

Business can be thought of as an optimization algorithm: tweaking stuff and things in order to maximize company profits. Interestingly, we use a number of different kinds of optimization algorithm when dictating the direction of the company.

We begin new product lines based on experience but continue to optimize our products based on customer feedback, trying to solve the problems that are most important to our customers. For example, our F# for Numerics library started life as our second attempt at selling libraries to F# users (our first attempt was F# for Visualization) and we provided the features we thought would be most useful. Customers inevitably requested more features including both technical features like parallel matrix inversion with arbitrary-precision rational arithmetic but also non-technical features such as licences allowing them to bundle our library in their own commercial products. This approach of incrementally improving products is a kind of local optimization algorithm, like gradient descent. This is a low-risk approach that yields small improvements in profits.

Over the years, we have been commissioned to write many books and reports and have published many books ourselves. We often use content from one successful book as the inspiration for the next. For example, our 2004 book OCaml for Scientists was surprisingly successful and became the inspiration for our 2006 book F# for Scientists and more recently Visual F# 2010 for Technical Computing. This is a kind of evolutionary algorithm because the next generation of books are derived from the previous generation and share some of their "DNA" with them. This is a riskier approach with variable results. We take a lot of risks and expect a significant proportion of our new products to fail but every now and then we come up with a new product that ends up earning as much revenue as all of the others combined.

At this level of abstraction, computational algorithms can be used for all sorts of weird and wonderful applications. Doubtless we all use them without even realising it!