Everyone’s talking about artificial intelligence, but what are the practical steps you need to take to get it to work for your business? Assaf Baciu, SVP of product, Persado looks at the questions organisations have to ask if ask if they’re aiming to use AI in marketing.
One question businesses ask themselves is, “are we ready for AI”? Based on the number of solutions providers touting fancy new “AI” capabilities my Google alerts throw up, the number of businesses who think they’re ready is growing. But what is it that they’re ready for?
There’s a lack of clarity around AI: Star Wars commercials, new Blade Runner movies and the progress of IBM Watson are spurring excitement. Yet this media eagerness to create hype – coupled with founder’s eagerness to drum up venture dollars – is creating a murkiness that tech giants like Adobe and Salesforce are capitalising on. Businesses have to begin separating hype from reality, by asking questions about how intelligent the solutions flooding the market really are.
Before and after adding any AI solution to the business, organisations should ask the following three questions to properly assess its value.
1. Is the “AI technology” actually new?
Back in spring Salesforce unveiled Einstein, the AI capability that’s claimed to make Salesforce’s platform better at identifying valuable customers in the sales funnel. Yet recommendation or next-best-action engines have been identifying high value clients for years – so what exactly has Einstein brought that’s new? Is it the ability to learn to correctly identify a valuable customer with no human assistance? Or is it just an improved algorithm? The former would be truly impressive, but I have a healthy scepticism about whether it’s possible. Einstein also claims to rank leads by value and suggest the best time to approach them – something which any programmer with an elementary understanding of machine learning could design a tool for.
In a similar vein, last November Adobe launched Sensei as “a unified AI and machine learning framework”. Adobe claims sensei will help businesses discover new “look-alike” audiences, but most good data management platforms provide that as standard. It also claims the ability to suggest messages that will resonate with a customer. Yet deep, personal and sometimes painful experience has taught me that this demands investment over years, with millions of samples used in thousands of experiments. How long has Sensei been studying this subject, and how far along the learning curve is it?
2. Have AI capabilities been developed with a specific business case in mind?
There’s currently no “plug-and-play” broad AI solution that adds value straight out of the box. Machine learning algorithms need time to learn: spending months, if not years, ingesting data to refine models and serve specific business needs.
For instance, IBM’s Watson API is undoubtedly impressive and has tackled questions from weather reporting and cancer diagnosis to how to win at Jeopardy! But these capabilities didn’t appear out of the blue – they needed a tonne of data feeding into Watson first. Something like Watson cannot be thrust at any new problem that appears unless the requisite contextual learning has taken place. If I were to take Watson’t horizontal API and apply it straight to marketing messages, I’d be lucky to get any actionable insights. More likely, any insights I get will be completely wrong. for example, Watson’s API would qualify “Attention please. Our offer ends today.” as “sadness.” How sad does that really sound to you?
3. How do you know the AI solution improves on what was already there?
The Harvard Business Review has noted that, while machine learning can help generate unique value insights, those insights will “often fail miserably if you try to apply them to something new, or, worse, they may degrade invisibly as your business and data change.” Testing AI recommendations against your current systems to ensure the solution is working as planned, and not giving potentially damaging results, is critical. Ensuring machine-learning technologies are actually benefitting operations demands changes to process and organisation that few businesses realise they need to make.
Without actively measuring results and holding the machine-learning models accountable, you can never be certain of AI’s ROI. Awareness of outputs is critical: when you consider how many “AI solutions” are just existing technology in a new box, keeping track of what they’re telling you and the value they bring is even more important.
By Assaf Baciu
SVP of product