This website uses cookies. By navigating around this site you consent to cookies being stored on your machine

Machine learning and analytics - without data scientists!

Chief Product Officer, BAE Systems Applied Intelligence
The importance of domain knowledge in good data science cannot be overstated. To extract value from investment it must be targeted data, that is useful and pertinent to you, and that can be turned into valuable information.
Machine learning and analyticsThe importance of domain knowledge in good data science cannot be overstated. However, to extract the added-value you might expect when investing in data science, acquiring any old data is pointless. It must be targeted data, that is useful and pertinent to you, that can be turned into valuable information.
 
And that raises the question: Useful for what?
The answer: decision making.
 
The problem with data science algorithms and software is that they automatically spit out more (and more) data, yet there are no guarantees as to how beneficial this might be. This is where domain knowledge comes into its own, translating data into meaningful information.
 

$10,000 for a hammer

 
It’s worth citing here the apocryphal story of a factory machine that broke down and halted production, at potentially enormous cost to the business.
 
An expert had to be called in to fix the problem. After a brief inspection, he seized a hammer and struck the machine, which immediately sprang back to life. The relief of the chief mechanic only lasted until he was presented with a $10,000 bill.
 
“This is outrageous!” he exclaimed. “10,000 dollars to hit a machine with a hammer? You spent ten minutes here. I’ll require an itemised bill for this.”
 
The expert pulled out an invoice sheet, wrote a few things down and presented the chief mechanic with the following:
  • Hitting machine with hammer – $1.00
  • Knowing where to hit machine – $9,999.00
  • Being online and making money – PRICELESS
 

It's about learn not churn

 
One household name that is clearly aware of the power and influence of domain knowledge is retailing giant Sears. Recently, it empowered 400 staff from its business intelligence operations to carry out advanced big data-driven customer segmentation. Traditionally, that kind of work would have been carried out by specialist big data analysts, probably with higher degrees, but things have moved on. What was once a matter of ‘churn’ (torrents of indeterminate data) has given way to ‘learn’ (identifying the data that has value – and could even be priceless). The Sears move is said to have created hundreds of thousands of dollars’ worth of efficiencies in data preparation costs alone.
 
Likewise, data science is a powerful tool but you need to know where to ‘strike’ –  how and where to apply it. And this requires domain expertise. Citizen data scientists come with domain knowledge, of course, but they may need support selecting and deploying the right tools from the data science toolbox.
 
Ultimately, when determining how to deploy data science, financial institutions must get this approach absolutely right to avoid regulatory fines or financial fraud losses. Only by having the precise data in the first place will they know they are setting out on that quest from the right place.
 
top
Neal Watkins Chief Product Officer, BAE Systems Applied Intelligence June 14, 2017