Seven Ways to Fail at AI
Seven Ways to Fail at AI
Recent forays into the marketplace have convinced me that there is a need to talk about AI failures. We are firmly in what Gartner calls ‘the trough of disillusionment’, the part of the hype cycle where a new technology fails to deliver the benefits everyone was hoping for. It happens every time something new and exciting comes on line.
Will the promise be realized? Yes, a lot of what we are anticipating with AI–good and bad–will happen. But because timing is everything, these things will not happen WHEN people believe they will. Also, it will not happen until companies become AI literate and put in the hard work. Meanwhile, there will be more wasted money, failed projects, programs and AI companies.
Be realistic, learn as much as you can from other’s failures and stay the course.
But if you WANT to fail–here’s what to do according to the evidence I’ve seen.
Market data and practitioner reporting indicate several structural ways to ensure an AI initiative fails to deliver value or maintain organizational support.
I’ll start with these basic failure modes. There are quite a few and most of them are even more serious.
Seven Ways to Fail at AI
Point your AI at big amorphous goals: Use AI to address undefined goals like knowledge worker productivity without specific workflows or baseline measurements. There is no point in making people 25% more productive unless they use that 25% to do something else. There is no provable ROI.
Don’t benchmark or create quantifiable outcome focused metrics: Failing to establish clear key performance indicators (KPIs) before implementation will make it impossible to prove ROI to stakeholders.
Implement or accept opaque, unpredictable pricing and licensing: Create or accept complex consumption-based charges that lack transparency and predictability. Confusion over licensing structures and consumption charges are a primary driver of decision hesitation and slower sales and implementation cycles, not to mention huge cost overruns.
Fail the CFO and procurement scrutiny test: Neglect to differentiate high-cost enterprise AI services from low-cost commodity tools. Guarantee that procurement scrutiny will intensify when a high-priced subscription is weighed against a $200-a-month alternative.
Assume or pretend that your data is AI ready: Do not factor in the costs of data preparation. Your data is a mess–everyone’s data is a mess. AI will help with data cleansing, governance and quality. It is not automagic, i.e., you still need people, and it is not free. Don’t act surprised at the 50% project cost overrun.
Fail to train your workforce: Start getting rid of people whose jobs will now be done by AI.
Forget everything you know about human nature: People respond in predictable ways to opportunities and threats, but don’t worry, that won’t happen this time. No one will use AI to look more productive, gain an unfair advantage or look for ways to evade change that they don’t want and may actually fear. People will gladly relinquish their spreadsheets and won’t ‘correct’ reports to match their targets.
Next week, I’ll be sharing more ways that you can guarantee failure with almost any AI project.
