An update to a blog I wrote back in 2011 titled “Two Thousand Models – And Counting!”
It’s been two years since I last checked, so I just queried my laptop to see how many model files I currently have. Not surprising, at least to me, the count is now 6,103. That is SIX THOUSAND different optimization-mathematical programming-prescriptive analytics models, or however you prefer to classify any model designed to solve a business problem by maximizing or minimizing some criteria.
The list of models span many industries: agriculture, banking, cement, chemicals, consumer products, defense, education, energy, food & beverage, forest products, healthcare, logistics, general manufacturing, mining, oil & gas, retail, steel, telecom and many others.
Ok, I admit that many are derivative models, or models based on prior models. The number of truly unique models is much fewer – maybe only a thousand, give or take a few hundred. Still, a derivative model doesn’t mean just a copy or “throw away” never to be used again. That would completely minimize their importance and disregard the time and effort spent on each one.
For example, a new model is often named something like ‘2014 base plan’. Once refined and validated, it usually leads to many additional targeted, “what-if” scenario models like ‘2014 base plan with plant 5 closed’ and ‘2014 base plan with Industrial product prices +10pct’. And those models can evolve into other models, as necessary. ‘2014 base plan’ might eventually become ‘2015 base plan’, and so on.
Compared to a base model, a scenario model always requires some additional modification, whether it’s changing the actual model structure (constraints and/or variables); data (objective function, coefficients, or limits), or both. Creating a scenario often means new import and export tasks, queries, dashboards, and reports. This all takes time, effort, domain knowledge, training, and capabilities. None of this could happen without a well-designed platform and an established process.
What I’m referring to is a radically different approach to prescriptive modeling; a unique process that has proven to bring substantial value to our customers over many years. A process where a consulting partner using a small team of subject matter experts and modelers can build, analyze, and use a fully working model in just 4-5 weeks. A whole new universe of short and medium term projects is available that was not possible using prior tools (even Excel). And, it allows our partners to share models as templates so their users can easily learn/interpret and expand what others have done before.
Let me describe this another way: all 6000+ models in my possession differ only due to customization (using a drag-and-drop visual UI); configuration (using wizards); and data (using nearly any available source). For a truly solution-based approach, this is the only way it can work. Rules required to create the underlying equations must be part of the base software and not custom-coded and stored as part of each model’s definition. This is a critical point because any model, even one created years ago, can still be upgraded to the latest software version. Indeed, I have models built over 10 years ago that are still relevant and useful today!