5 Things That Keep an AI Project from Being Successful

Share it:

Many Artificial Intelligence initiatives are doomed from the start – but needn’t be. Here are five things that keep AI projects from achieving success.

1. No Strong Center

A successful AI project requires a well-defined center and clear goals. Why are you doing AI? What is the vision for the future? The inability to answer this question indicates no end-state vision. With a well-defined strong center, execution becomes subordinate to the goal and all paths converge. With a weakly centered AI initiative, the paths point in every direction. It becomes impossible to focus on doing one thing extremely well, so many things get done poorly.

2. Poor Leadership

AI needs leadership, not management. From vetting strategic opportunities to knowing what each day should prioritize, someone must carry the vision. Management isn’t enough, not even good management. Unfortunately, it is common to have the wrong person in charge. Organizations tend to either put a highly technical person in charge or a manager. Good AI leadership isn’t always the most technical person with a distinguished Ph.D.; a good leader is that individual who knows better than others what to do next.Consider this from the McKinsey Global Survey on AI: Respondents at AI high performers were 2.3 times more likely than others to consider their C-suite leaders very effective.

3. Lack of Urgency

Speed matters. Going fast is the best way to remove risk and improve competitive position. Everything about AI moves fast, and so must any organization intending to use it for a competitive advantage. The typical culture of “slow” that creeps into larger organizations simply can’t compete in the world of AI. There are too many fast-changing elements.

4. Bad Data

Tips 1-3 apply to any transformational initiative, but bad data is AI-specific. It’s not a question of whether you have bad data, because nearly everyone’s data is awful. With rare exception, nobody has good data from an AI perspective. The real factor is how one responds to a bad data situation. The typical approach is to pretend like the data is amazing, build an AI project around supervised machine learning (the easiest type), show some encouraging early results, then churn forever once the data quality issue catches you. Such projects have difficulty shipping meaningful innovation to production, and the cost usually outweighs the opportunity. However, bad data is only a showstopper if you don’t admit to it early and respond with evasive action. This is where good leadership really helps.

5. The Team

Another reason AI projects fail is the team. Every CEO likes to believe they have the best, smartest and hardest-working people – and they might indeed have them. But the makeup of an AI project team is critical; such projects are extremely sensitive to talent, skill set and work ethic. A common mistake is staffing an AI project with data scientists. Data scientists analyze data and produce reports. AI engineers, on the other hand, need to go beyond analysis, machine learning and Python. They are trained to build sophisticated architectures, develop complex offline training rigs, optimize computer hardware/software system performance, and run mind-numbing tests in monstrous repetition. While some data scientists graduate into AI engineers, most don’t. An AI engineering team needs people who are software systems engineers as well as neural network architects.
The promise of AI is real. While the widely cited case studies are largely still being written across each industry, assertive organizations are being extremely successful by avoiding these failure modes.