One of the challenges I see within the development of models, analytics and the technical engineering space, is how to keep an initiative or assignment focused and efficient. Dealing with mostly technical people, myself included find it all too easy to think of more things to look at or some other nice way to accomplish a task.
A challenge I find with junior or graduate resources who are starting out, is the tendency to solve or develop a solution to a problem in one go. There are inevitably issues with any model or tool that is developed, utilising a structured approach is a must – even if it is a set of items to implement.
In the software development space, there is a concept of a Proof of Concept (PoC) and Minimum Viable Product (MVP). The PoC is about testing a concept quickly to see if it is feasible. In most cases PoC’s are developed within 1-2 weeks and is used within a fail-fast approach. An MVP is where a software tool is developed to have the minimum or basic or core set of requirements that can be tested with users – this concept can be readily used in the model development space for physics based models or those developed from data as such as Machine Learning or Large Language Models.
In the consulting analytics and modelling arena we are typically utilising tools or frameworks which are tried and tested as we want to be able to deliver on time and budget. Thinking about the way we tackle an assignment or problem in steps also provides the ability to show progress – as clients can become nervous if they don’t see something to give them comfort that their project will be delivered on time and meet their requirements. This approach also allows for the key items to be tackled first, and the more optional nice-to-have items to be added towards the end of the project.
Another question to ask, is how good is good enough? I have had clients who are chasing an academic answer to an engineering question. Sometimes confirming that the basic solution is more than adequate and proves that the solution being considered is suitable. Does refining the model to get a better accuracy actually change the decision/outcome that a particular analysis is being undertaken for?
I hope these thoughts and insights give you some areas to investigate and look for optimisations in your work. Always happy to discuss some more, as there are some good workshop approaches to apply early in these types of projects.
Comments are closed