Most (80%) of Enterprise AI projects fail.
Most of those failures are due to four simple things that I have described in “Four simple questions to prevent AI project failures” which may or may not be themselves do to failures in leadership.
Only about one in five of the project failures is killed by one of the leadership problems listed below. They often make things less valuable, more costly and late. But each one of these is rarely the primary cause of failure.
Question 1: Is your implementation team working on the same problem that the leadership has tasked them with?
This isn’t a question of malfeasance, simply of confusion. It is a very common problem in IT projects in general, but seemingly less frequently fatal on AI projects.
Question 2: Would this be easier to do without AI?
Unfortunately real examples: Extracting data that the company already had. Large Language Models doing the work of common pre-existing libraries that the company already had access to.
Replacing perfect results delivered the same day with person-hours of work, with so-so results delivered in under a year with less than 6 person-years of effort.
Question 3: Are the time estimates entirely unrealistic?
Decades ago, Gorden Bell observed that a mature organization is generally late and over-budget by about the same amount for each project of a particular type. He called this that organizations “Fantasy Factor.”
With more pressures, less experience and larger differences than typical from your normal work, are your estimates much more unrealistic than usual?
Question 4: Is your team more interested in getting another job than the success of your project?
Perhaps they are focusing on more marketable skills or different tools than would be best for your project.
Question 5: Have you skimped on data engineers?
Data engineering is a distinct skill using separate tools and techniques from the AI and data analysis tasks. It is common to have to few (or perhaps zero) data engineers leading AI scientists and data scientists to have to do these tasks. It takes them longer than it would data engineers. The results are less reliable. And it makes them unhappy and more likely to jump ship to another company that doesn’t make this mistake.
Question 6: Is you using the wrong development method?
I’m a big fan of test driven development (TDD), agile, rapid-prototyping and other methodologies that benefit from rapid feedback and deterministic processes.
Personally, I’d probably like to apply parts of those to many projects in your organizations.
But for many AI and Data Science tasks where and experiments may take hours or days rather than seconds, its a bad fit.
Is your AI saddled with a generally wonderful methodology that just isn’t wonderful for their project?
Question 7: Are you hiring the wrong people?
There is no consistency in job titles. Nor in degree names. What are you doing to see that the skills/interests match the jobs?
This is not a question of people lying on their resumes, etc. That’s a problem everywhere.
This is a problem of legitimate confusion.
Question 8: Have you picked the right problem?
There are two common ways to pick the wrong problem.
The first is answering a very short term need. It’s generally a large investment and a lot of lead team to build an enterprise AI solution. It should be something that will be valuable for a long time.
The second is focusing on something of prestige rather than value. While more often a problem in academia, it infects business as well. Projects that turn into wonderful presentations but have no real use for the company.
The four easier questions
If you haven’t already asked the four easier questions, you really should. They are easier, higher leverage and much more likely to be what kills your project. Kills it dead. They are described here: Four simple questions to prevent AI project failures.
Want to Avoid AI Project Failure? Let’s Talk.
If you want to dramatically reduce your time, cost, and risk, let’s discuss how your AI project can actually succeed.
✔ It’s free.
✔ It’s fast.
✔ It’s easy.
Schedule a time to talk. Only you can prevent catastrophic failure.
