The Strategy Vector
If you think you can separate your digital and business strategy, then you are probably a bit behind the curve!
The idea of this post is to understand
- Can we represent business strategy as a Vector? ( i.e.an array of numbers that may be multi-dimensional)
- Why would we want a strategy vector?
- Is any company using such an approach?
The "geeky" answer to these questions is [1,NaN,1]. The Strategy Vector is an "ubercool" approach used by companies to balance the hard objectivity of analytics or data science with soft subjectivity of "gut feel" or business focus.
What is a Strategy Vector?
Simply put, the Strategy Vector (SV) is a set of weights or importance assigned to each output area. As an example, you calculate cross sell propensity for a customer to take up a particular product e.g. Car Loan or Personal Loan amongst other products. The X-Sell vector for Customer1 across all products is [0.88, 0.95,....,0.15]. The bank's strategy for this quarter is to promote Car Loans over any other loan while taking affordability into consideration. Then the strategy vector may look like [2,1,..., 0.5] i.e. Car loans are twice as important as Personal Loans due to bank strategy. This strategy vector weight would be multiplied with analytics output to generate final propensity. Hence if "customer needs" derived probability is in line with bank's strategy, those customer offers get a boost while if a customer need is very strong, that can also cross the strategy threshold and figure in final output. It introduces a deliberate and controlled bias towards any strategic area. There are many forms, uses and derivations that can be utilized to find out the strategy vector but that would be going into too much detail. In case you want to turn it off, SV = [1,1,1,..., 1].
Why use Strategy Vector?
There are 3 primary uses of Strategy Vector
- As an incubation approach [analytics change management]
- As a safety net
- To give a strategic sales push or use a short term Product Centric approach without drastically sacrificing Customer Centricity.
Is it being used?
Yes, it is used in a few places ideally when there is a bit of resistance to accept a predominantly analytics driven approach. So the product or customer segment manager may say "fine I will use your model output provided they do not disrupt my business as usual too much". Then some deliberate bias is introduce to instill his confidence in the model and then gradually SV is set to [1,1,..].
The other place where it was used was when the channels started complaining they are not getting enough volume of leads to work with. It does not matter if they are getting quality leads so that net conversion is higher; there is still a mindset that needs to justify channel capacity driven by daily volumes (spray and pray approach). Rather than setting a lower model output threshold, use of SV gives a far more sophisticated tool to gradually manage paradigm shift of quality vs. quantity and over / under capacity utilization issues especially for specialized skill based channel where multi-skilling takes time.
Lastly, campaign managers ask "how to handle the company push on selling quartz filament electric heaters over oil ones. The product team is way behind their targets and that area needs special attention". You already know what to do there!
All above were transactional examples where SV is fairly dynamic and localized. As a business strategy if you want to explore a new business area for a long term, you can set a global SV which needs to be used throughout all modeling outcomes to give a new business push.
Though I have seen local SVs being used often, I have not yet seen an overarching SV used throughout. It should be possible and only a matter of time before it jumps from theory into practice. What do you think? Let me know...