© 2018 by M.i. Media

Google Partner.png
LW_logo_london_rgb.jpg
  • Admin

Data Science - Can Machine Learning Still Benefit From Human Intervention?

Updated: Aug 28, 2019

Written by Sophie O'Brien

Artificial Intelligence. Big Data. Machine Learning. Data Science. The buzz words of the moment, but what do they mean for us in the advertising industry and just how useful might they be?


As this Venn diagram from leading analytics developer SAS ably demonstrates, it is all a bit of a minefield. You are looking at the universe of data science, in case that’s not totally obvious!



Reference: SAS Institute

Still lost? While these terms do actually mean different things, their core objective remains fundamentally the same: to harness the power of data and make decisions more quickly, more intelligently and with as little human interference as possible.


The applications are seemingly endless. In advertising alone data and technology is being used to automate TV buying, combat digital ad fraud and simplify the management of biddable campaigns. It’s a wonder any of us have jobs anymore. If the buzz is anything to go by, the advertising world as we know it will never be the same.


There’s no doubt that these advancements in technology are extremely powerful and as a result advertisers are increasingly turning to data science, whatever field it might be, to make decisions about their marketing investment.


However, as with all new technology, we need to be very clear on its potential limitations before we all just jump on the bandwagon and blindly follow what we are being told.

At M.i. Media we use sales data to enhance advertising performance every day. We therefore have a pretty good understanding of the most common problems companies encounter when trying to tap into this potentially rich vein of insight.


In our experience, many companies don’t have the infrastructure in place to accurately collect the right data in the first place.

It might sound straight forward (especially to companies who can fund a team of CRM specialists), but with a challenging economic climate and huge pressure on marketing teams to boost short term revenue, it’s quite often seen as too big and complex a project which is continually put off until tomorrow.


Many companies just don’t have the gigabytes or terabytes of data needed to power machine learning.

Larger businesses are inevitably going to benefit from more robust machine learning systems because they have “big data” available to build into a model. But smaller businesses may well struggle to fuel a sophisticated decision making process with sufficient data.


Blind faith in algorithms can be very expensive.

The computer is never wrong. Oh dear. We have news for you. It very often is! One of the most common problems we encounter is when businesses have stopped questioning the information they are being given. When a model tells you you’ve made £1,000,000 and your bank account tells you otherwise, it’s probably time to question the model!


Computers don’t understand the nuances of media buying

We use predictive TV modelling tools on a daily basis and continually find ourselves pushing back when we uncover insights that just don’t stack up. Unless of course we’re being too sceptical about the phenomenal impact that Horse & Country TV could be having on our client’s business!


Is the investment required actually greater than the potential commercial value?

Unsurprisingly many advertisers feel a million miles away from being able to jump on the “AI” band wagon. However, starting off on the journey to harness data and make more intelligent decisions may not be as intimidating and costly as it seems.


Rather ironically the answer, we feel, is human intervention. The power of data certainly can’t be questioned. But how we use that data is crucial in the decision making process and sometimes a bit of human rigour is exactly what is needed.


At M.i. Media we find the right solution for each of our client’s data needs, from data centralisation and cross-channel attribution through to econometric or revenue forecast modelling. By combining the power of both machine and human intelligence, we help our clients deliver substantial efficiencies in a timely and affordable way.