It used to be all about Mobile First, now it’s about AI-first. Google now calls itself an ‘AI first’ company.
“How do you decipher the truth from puffery? Are there any shortcuts to really understand where AI is best applied in your business?”
I’m not a data scientist, but I spend my days talking to users and buyers of AI technology who are befuddled and increasingly cynical about the hype. Here is my 3-step guide to cutting through the noise.
Most products are using standard statistical techniques like regression. Without any machine learning baked into the technology, they are just matching tools. There are efficiency gains, but no ‘smarts’ and no learning in technology.
For example, in recruitment a stock-standard AI product would merely find you people with the same profile as those you have already hired, matching applicant profiles to hired profiles. CV parsers do this kind of thing. Now, that can be helpful to you if all you want to do is a short-cut to the same profile fast, and it can definitely save your recruitment team time.
But unless you know that these characteristics also match performance, you will not make a difference to your organisational outcomes.
If your Sales Director tells you that every new hire over the last year is hitting or exceeding budget, then absolutely keep using that tool. If she tells you that a third or more are underperforming or leaving the business, your AI tool is merely amplifying that bias and doing quantifiable damage to your company’s bottom line.
If you want both efficiency and business bottom-line impact – your AI needs to have machine learning baked in.
Takeaway: If the organisation selling you an AI tool has no Data Scientists, there is no machine learning in the product.
If I imagine a Maslow’s hierarchy of AI it would look like the below:
First up you need to have the data.
About 50% of companies don’t pass this threshold, but assuming you have it the next step is understanding the data context: What is the business problem you are trying to solve?
Next is the housekeeping. This means consolidating, cleaning, categorising and cataloguing the data. And then finally the optimisation is at the top – this is where the magic happens. Optimisation is the last mile and is what gets you to the big savings, but you need everything underneath it in order first.
In recruitment a genuinely smart AI tool with machine learning baked-in works best in these conditions:
Takeaway: It’s critical to have a solid understanding of what you’re trying to fix, and the means to measure the changes you’re making. Ask yourself why you are considering AI if you can’t quantify the problem, to begin with.
AI is about optimisation. For credit card companies it’s detecting fraud quickly. For online retailers better product recommendations. In recruitment, it’s finding the best new hires in a massive group of rookie players.
In each case, you are optimising for efficiency and accuracy, as the cost of getting it wrong is huge.
It means trusting the technology to find the patterns. You have to suspend theory, and your assumptions, a lot. You feed in a large amount of relevant and unbiased data and the machine learns on its own, finding the patterns. It is looking for the ‘signal in the noise’. Humans are unpredictable and more often than not unreliable.
The current hiring processing by humans is extremely resource-consuming and the result is not always satisfying. Using AI will free up your time if you allow it to, improving efficiency or outcomes, often both. But AI built just off CV data only adds bias and we’ve all see how badly that ends.
Takeaway: Predicting human decision making is not easy and not quick. The only way to get to the ‘answer’ is to start now and expect this to be a journey.
Get our insights newsletter to stay in the loop on how we are evolving PredictiveHire