When you don’t know what to do…
  • 3rd April, 2018
  •  

When you don’t know what to do…

By Steven Sidley

Machine Learning and Artificial Intelligence or AI have a hype problem. C-Suite executives read about this magical new technology and send frantic missives down the line - ‘If our competitors get hold of these tools, we are toast! Buy some machine learning. Start AI projects! Make it happen! Don’t delay!’ And so they are faced with the oft-repeated solution-looking-for-a-problem bewilderment. 

We've engaged with a number of corporate clients who find themselves stuck in the middle of this mire, including those who have assembled entire teams of technologists and anointed these departments as ‘Data Science’ or ‘Advanced Analytics’ or even the grandiose sounding, ‘Intelligence’. They soon find themselves attending conferences, taking deep learning and statistics courses and wrangling the bewildering array of free and not-so-free tools in hastily assembled tests and light pilots.  

And after showing absolutely no business value after months or even years of fits, starts, strategies and tactics and grand plans, failure is rationalised away. A manager, called to account, will stammer excuses about the corporation having no ‘data culture’, or blame it on the intransigence of old technologies or creaky architectures or lack of executive champions. Or worse. The manager may even produce a fancy visualisation of a sales report and claim victory, silently aware that such reports have been available since the medieval days of what was then called Business Intelligence. 

This is not to say that the solution-without-a-problem is necessarily doomed to fail. We remember well the rush to connect to the Internet, or to connect to broadband, or to mobile long before the true advantages of these technologies found their most productive homes within the enterprise. 

At Ixio Analytics we’ve devised a careful strategy for those who may feel they are stumbling blindly in the dark. The scaffold looks like this:

1 - Data Discovery

Many enterprises (including IT departments) have little structured understanding of where data is kept across the corporation, for what purpose it is used, who owns it, whether is is tethered to other data in the enterprise, and how accessible it is. Without a clear schema to map this high level view, the company is largely data blind. 

2 - Opportunity Analysis

While many enterprises have a well-articulated and prioritised view of company drivers (revenue, costs, profit, assets, etc), they rarely have a clear view of what data exists (or could be collected) to inform these drivers, and whether that data is clean, current or sufficient. An analysis of drivers and their relation to data can lead directly to opportunities that data science approaches  can then exploit.  

3 - Representation

Once an opportunity or opportunities have been identified, solution implementation begins with the often underestimated task of representation. This may include not only data cleansing and re-housing, but some of the techniques formalised in data science, such as labelling, taxonomies, linking and segmentation. 

4 - Optimisation 

The optimisation step is where the science takes hold, and where specialised skills are required. Whether the techniques are driven by inverse deduction, evolutionary learning, back propagation, statistical modelling or similarity/vector modelling, it is unlikely that these resources are widely available in the enterprise. This requires deep specialist skills, and these usually exist outside of the enterprise. 

5 - Evaluation

Ultimately, evaluation is measured by which company drivers have increased along identified metrics and along forecast timelines. Without this business-level adjudication, the solution is incomplete. 

Finally, the steps above require both domain specialists and generalists, both technologists and strategists. Given that the magic field of data has suddenly found its moment in the sun, and bearing in mind that most enterprises were not driven by data from their inception, there needs to be an understanding that some structural re-invention, time and money will be necessary to truly turn data into fuel. 

So talk to us if you want to introduce machine learning and AI in to your enterprise, but are not quite sure where to start. We’ve been there, and we know the route. 

Share This Artcle :

Comments