Why most companies are too stupid to get value from AI
This is a guest blog post by Chris Lynch, executive chairman and CEO, AtScale.
I am not known for mincing my words. It is a fact that most companies are too stupid to get value from AI/ML. Too stupid to get value from data science. Too stupid to get value from their data. But I don’t mean stupid as in “low intellect.” I mean it how Forest Gump meant it, “Stupid is as stupid does.”
The simple fact is that most organisations are not organising themselves, their business operations, or their data assets in a way to get any real value from enterprise AI or data investments. Furthermore, they waste their time whining about a lack of smart data scientists when they should be focused on technology investment, organisational alignment, and process change to integrate data-driven culture into the fabric of their business.
I don’t mean to oversimplify the challenge. But as Chairman of the board at DataRobot and Executive Chairman and CEO at AtScale, I have seen what smart teams do. There are a few basic priorities that leaders need to set in order to put their organisations on a path out of data stupidity.
Manage data as an asset
The applications that modern businesses run on generate massive amounts of data – from inventory to customer to sales to product to supply chain to employees to financial to engineering data. Stupid companies manage data as a cost centre. Smart companies think about building data assets that can fuel decision making and innovation.
Data assets are built from raw data but represent models of the business that can be used to answer questions. Data assets are designed by architects and modelers who understand the business and understand how decisions are made by leaders. Data assets are dynamic in that they are continually refreshed with new data flowing in from applications.
Help make data speak the language of decision makers
“Lack of data skills” is such a common refrain in modern business it is cliché. Stupid companies funnel all their data questions to a small group of experts with data skills. Smart companies adopt a semantic layer for data and analytics that makes data speak the language of decision makers.
Instead of interacting with raw data with obscure naming conventions and complex normalisation schemas, data consumers work with defined business metrics and analysis dimensions. The concept of a semantic layer applies to both classic business intelligence and to AI use cases, where it is referred to as a feature store. A recent article from McKinsey that’s worth reading talks about “Adopting a Smart Data Mindset,” where data teams focus on curating smart data assets that are designed for AI use cases.
Ask AI to do things it is good at
We are a long way from an Enterprise AI Singularity that will just tell us how to be more competitive. Stupid companies don’t know how to formulate questions that AI can answer. Smart companies focus on augmenting analytics and business processes with the insanely fast statistical modeling techniques that AutoML platforms bring to the table.
Predictions, forecasts, anomaly detection, and pattern recognition algorithms can augment routine business processes. Decision makers still need to make decisions. But they can make better, faster decisions when guided by clear views of historical data (BI) augmented with AI-generated insights. Inventory planners get better predictions of demand. Customer success managers identify signals of potential customer churn. Marketers get new ideas for adjusting advertising mix. Sales delivers better forecasts to finance.
The key is asking the right questions of the data and focusing AI on the right use cases.
Get AI-generated insights in front of decision makers
Data science teams are churning out insights in the form of predictive and prescriptive analytics. Stupid companies look at AI-generated insights as a separate category of data to be analyzed in forecast meetings. Smart companies integrate AI-generated insights into business process tools, letting decision makers consume predictive and prescriptive analytics the same way they manage their day to day. This means publishing AI outputs in the same dashboards used for analyzing historical data or, even better, integrating AI-generated insights into business applications.
The concept of a semantic layer is again critical to make the new data generated by AI understandable to decision makers.
Act smart, be smart
In my experience, most companies still fall into the stupid category when it comes to getting real value from AI. The geniuses at McKinsey just wrote about the role of the CEO in setting a strategic vision for integrating AI. I agree it starts with vision, but it must be followed by action. By working with some of the most forward thinking, truly smart organisations, I’ve seen firsthand that the journey toward AI empowerment is gradual and made up of discrete, conscious steps. Money and time are spent deliberately to position teams and individuals to make progress. The smart companies that do this well are destined for greatness.