70-80% of AI projects in IT organizations fail. Here’s why.

"If the only tool you have is a hammer,
you will start treating all your problems like a nail.”
– Abraham Maslow

While the quote has many variations and has been attributed to several folks, the underlying message is simple – over-reliance on one tool, process, or technology leads to cognitive blind spots.

Ever since GenAI gained mass popularity, everyone seems to be trying to force-fit AI to solve their problems. Even before the launch of ChatGPT, IT organizations have tried to introduce AI with a low success rate.

Around 70-80% of AI initiatives fail. The core problem often comes down to giving in to the hype and taking a “hammer” approach with AI while missing the nuances.

From a mindset perspective, it’s important to remember these two seemingly conflicting tenets outlined by Charles Araujo in a recent chat I had with him. I paraphrase:

1. AI is a means to an end. The end goal is more important than a specific tool or technology.
2. The hype will compel your competition to try and adopt AI. Since they’re not standing still, neither can you.

Since you’re reading this, I’ll assume that you already buy into point #2. IT organizations that are not looking at business problems and opportunities and asking themselves “where can we leverage AI to solve for these?” will be at a disadvantage once the dust settles.

There are a ton of potential applications of AI in ITSM, including:

• Driving self-service adoption
• Deflecting mundane and repetitive tickets
• Improving productivity and efficiency across the board
• Improving user experience at scale without hiring more support agents
• And so much more

What are the potential roadblocks?

If AI in ITSM has so many apparent applications, why aren’t they happening?

In a McKinsey interview, McKinsey senior partner Harry Robinson talks about why 70% of transformations (in general, not just AI) fail. 

The three main reasons highlighted by Robinson include:

• Inadequate leadership aspiration
• Skills and capability gaps
• Neglect of procedural elements

We conducted a study on the state of AI in IT. One of the questions we asked IT professionals was about the barriers to AI adoption. We found these to be the most popular barriers:

• Customer data security
• Additional cost
• Inaccuracy or inconsistency
• Lack of expertise within the team
• Governance and compliance

How can you increase the odds of success?

Based on all these factors, we can distil down a few things to keep in mind when planning your next AI initiative.

1. Pick a problem that’s a business priority

Leadership buy-in is important to ensure that the project gets not only the resources, but also a top-down push when you’re met with lateral resistance. 

A project involving a relatively new technology, like AI, is bound to run into challenges and failed experiments. If you’ve set out to solve a problem that’s critical to the business, you can expect sustained availability of the resources and leadership backing to coast through them.

2. Understand that changing user behavior is hard

There are exceptions to this rule and it fundamentally comes down to whether the net change in friction is positive or negative. People are inherently lazy and if technology allows them to be more lazy, they’re likely to “adopt” it without much resistance.

Of course that’s a broad-brush generalization, but all major technological leaps – from electricity to computers and from Amazon to Netflix – have allowed humans to get more output with less effort. The pandemic has now given rise to what’s being called the “introvert economy” which has exacerbated this further. Use it to your advantage.

ON THAT NOTE 
At Atomicwork, we believe that asking end users to visit portals (or remember different email IDs for IT, HR, etc.) to access support is flawed.
 
We think you should bring support to where they already are – tools like Microsoft Teams and Slack.

Ask our team of AI-based ITSM experts for a demo.

3. Start nurturing talent early on through hiring and training

One roadblock that both McKinsey and our research found is the lack of the right skills and talent. Since AI teams haven’t been a thing for too long, finding the right talent can take time. AI is a broad term that encapsulates Machine Learning, GenAI, Computer Vision, and more. Prioritize hiring and training for the bespoke skills you need.

Also, have a “build vs buy” discussion early on. You might not need to spend a ton of time and money on hiring and nurturing talent unless you’re hoping that the AI-based solution gives you a unique competitive advantage.

4. Recognize that users are prioritizing their own goals

We see “shadow IT” as inherently negative, but that’s because we see just one side of the story. Users don’t subvert IT guardrails to buy technology just because they can. They do so, because the techology provides value or helps them solve a problem. Work with them to understand this value and help them solve their problem with technology.

There is a strong tailwind when it comes to end users adopting AI – around 75% of end users are using free AI tools for their work already. Use this to your advantage rather than clamping down on users, which will only end up driving users to bypass IT guidelines.

5. Educate users on the importance of governance and compliance

In line with the previous point, it’s important for IT to be seen as enablers rather than roadblocks. One way to do this is to educate users on the importence of following protocol.

In addition to communicating the rules, help them get the maximum benefit from the AI offering through training and development. Work with your HR or L&D partners for this.

6. Change manage the program with the end user in mind

The IT version of change management addresses the technical side of the planning, rollout, and if required, rollback process. Another important part of rolling out an org-wide change is the human aspect.

Have a pre-launch, launch, and post-launch plan. During the days leading up to the launch, communicate the benefits of the technology to the end users and show them how their lives would be better because of it. Post launch, give them a channel to share feedback and show them how you're addressing or acting on it.

7. Track progress and continually improve

Finally, make sure that you have the analytics and tracking in place to monitor the performance of your initiative. GenAI is yet to evolve to a 100% accuracy and hallucinations are still a big concern for IT and users alike. To avoid eroding the trust of end users, it’s important to track inaccuracies that make them revert to the usual method.

If GenAI is a big part of your offering, remember that it’s only as good as the data and knowledge base that you feed it. Qualitative feedback and quantitative data will help you improve the efficacy of the data and the quality of responses over time.

Conclusion

Nothing like sharing another implementation stand-out example to conclude!

Unilever utilizes GPT API to help their agents smartly compose responses to customer messages and create product listings. Unilever's AI tool effectively discerns consumer preferences, significantly reducing the time needed for agents to compose responses by over 90%.

Seems like magic, doesn’t it? This is precisely what AI in ITSM can help achieve!

To put it plainly, AI in ITSM isn’t really about AI. It is about using AI effectively to achieve clear-cut business goals. AI is a great medium, given its immense potential and promise.

It is important for organizations to do a deep evaluation and get started with smaller IT projects that have AI at the core to rake up quick wins, learn from them, and then move on to bigger and more ambitious projects.

In his book ‘Shoe Dog’, Phil Knight, the legendary founder of Nike, says, “Life is growth. You grow or you die.”

True words indeed!