There are lots of articles these days about the challenges of recruiting enough data scientists, predictive analytic specialists or data miner (whatever you call them). There’s not enough of them and it’s hard for most companies to compete with Google, Amazon, Facebook et al let alone political campaigns and startups. For most companies simply going out on the open market and trying to recruit in the face of this scarcity and competition is simply not going to work – it’s just too hard to recruit the skilled graduates they need. What to do instead?
There are lots of articles these days about the challenges of recruiting enough data scientists, predictive analytic specialists or data miner (whatever you call them). There’s not enough of them and it’s hard for most companies to compete with Google, Amazon, Facebook et al let alone political campaigns and startups. For most companies simply going out on the open market and trying to recruit in the face of this scarcity and competition is simply not going to work – it’s just too hard to recruit the skilled graduates they need. What to do instead? When you can’t recruit enough of a skill there are generally three things you can do:
- Outsource to specialist providers
- Train and develop your own staff
- Find a way to reduce your need for the skill
When it comes to advanced analytics there are certainly plenty of outsourcers out there with most big SIs adding analytic skills as fast as they can. If you see analytics as a critical skill however, and you should, outsourcing this skill to someone else is probably not a 100% solution. Add in the need to control at least some of the data involved pretty closely and having some in house capability should be in the plan. Developing and growing skills internally is certainly going to be part of the solution too but I would like to suggest that the third way, reducing demand for this skill without giving up the benefits of analytics, is the most practical and interesting to try. So how can you reduce your demand for advanced analytic skills without reducing your demand for advanced analytics?
Well, you can industrialize your analytic process so it becomes more scalable and production-oriented and less focused on individually hand-crafted analytics models. This means first using modern tools for the analytic tasks themselves that apply machine learning and other automated analytic techniques to reduce the need for analytic experts to do every step in the process. Stop having individual experts hand craft every script, every variable and take advantage of the modern tools that exist.
It also involves getting other people, people without the deep analytic skills that are so hard to come by, to participate effectively in advanced analytic projects. The analogy I like to use is complex IT projects. When a complex IT project starts it is true that IT architects and developers with deep technical skills will be involved. However much of the project definition, requirements elicitation and logical requirements design is done by folks without these skills. The ability of others to handle these tasks effectively scales those scare IT resources so they can handle multiple projects. Compare this with a typical analytic project where the first person assigned is an analytic expert and where this analytic expert is expected to do more or less everything – from developing business understanding to data analysis to model building and more.
This singular focus on analytic experts has several negative consequences:
- It’s hard to prioritize projects (and so focus your limited resources) because the resources you need to prioritize are the same ones who develop the initial value proposition for the projects you are trying to prioritize!
- It’s hard to scale to the number of projects you need because every new project requires advanced analytic skills – nothing can be done without access to an analytic expert.
- It’s frustrating for the analytic expert because they end up being brought into projects where there’s no clear statement of intent, no business objective and have to spend their time figuring this out rather than seeing how analytics might improve the business.
- It’s expensive because everything in every analytic project is being done by the most expensive people – analytic experts.
- It’s a sad fact that just because you are good at analytics you are not necessarily good at talking to people, making this first step of the project painful for all concerned.
What if the first steps in the project could be done by business analysts and business people with only minimal involvement from the analytics team? What if you could capture the business need and enough business understanding to see where and how analytics might help without these scarce analytic skills? What if every proposed project could develop a description of how it would improve the business that could be shared across business, IT and analytics team and compared between projects? Decision requirements modeling just such a tool.
Decision Requirements Modeling provides the techniques needed for developing business understanding of analytic projects. Established analytic approaches such as CRISP-DM stress the importance of understanding the project objectives and requirements from a business perspective, but most organizations do not apply a formal approach to capture this understanding in a repeatable, understandable format. Decision Requirements Modeling closes this gap. Using decision modeling techniques helps develop a clear business target – the decision to be influenced – as well as an understanding of how the analytic will be used and deployed, and by whom. A decision requirements model makes it clear how analytics can/should influence decision-making, where that decision-making fits in terms of business processes and systems, and which organizations will have to play a role in approving or using any analytic being developed. Constraints on decision-making like policies and regulations as well as the information available are likewise part of the model. Because these models are largely graphical, they can be developed and understood by business, business analyst, IT and analytics professionals alike.
Using decision requirements modeling gives you a standard way to define the requirements of your analytic project and one that scales scarce analytic skills across multiple projects, focusing them on critical analytic tasks and allowing effective prioritization and project management.