Over $30B have been spent by companies on Gen AI. But 95% of them fail to see any returns. A big reason for this is the bias towards top-line functions as opposed to high-ROI backoffice automation.

Replacing customer facing roles like sales/marketing/customer service or introuding features that are customer facing like virtual-try on for a fashion company etc. is definitely very enticing. It looks good on paper and on POC. But converting them from the pilot to production is extremely difficult. And most times this is where things get stuck and teams get demotivated and these toy applications never see any returns.

A better approach would be to automate a lot of the backoffice work. These are mostly deterministic. You can use AI in two ways:

  1. For tasks that don’t require any thinking and are pure logic, use AI to develop mini internal tools to make the employees more productive
  2. For tasks that require some decision making based on internal data, use AI to make those decisions (For eg: Which category to provide coupons, which ageing inventory to discount etc.)

These are not as sexy as customer facing AI tech, but they will provide immediate returns and verifiable returns. And as for getting the Engineering teams (which love making customer facing toy projects) to get onboard, I feel the below approach posted by a PM on Reddit is the way to go. I think if I got the engineers in my company to do some of the backoffice work for a week, they will completely revamp the entire internal tools we have.

Source