Why AI Investing Fails

According to two recent Gartner reports, 85% of artificial intelligence (AI) and machine learning (ML) projects are failing, and only 53% are transitioning from prototypes to production. However, both reports show little sign of slowing AI investment. Many companies are planning to increase their AI investments.
ⓒ Getty Images Bank

Avoiding AI investment failures requires a bit of common-sense business thinking. Strong investment incentives include Fear Of Missing Out (FOMO) and the vain VC investment bubble of AI companies with large marketing budgets, and the recognition of the need to leverage AI-based decision-making and transition to data-driven companies.

It’s best to think of AI as an outdated capital investment rather than thinking of AI or ML projects as a one-time thing like database upgrades and new CRM adoption. Just like the way manufacturers justify buying expensive machines.

Manufacturers won’t treat machines as fancy new toys as many companies think about AI and ML. In purchasing decisions, floor space, spare parts, maintenance, staff training, product design, marketing and distribution channels for new and improved products will be considered. The same concerns must be faced when applying new AI and ML systems to enterprises.

Here are six common mistakes companies make when investing in AI and ML.

Put the carriage ahead of the horse

Jumping into an analytics program without knowing the problem to solve is a failure. If there are too many distractions, it is easy to become vigilant. Self-driving cars, facial recognition, and self-driving drones are the latest toys, and it’s natural to want to play with them. We must not forget the core business value of ‘making better decisions’ provided by AI and ML.

Data-driven decision making is not new. Claimed to be the world’s first ‘data scientist’, Ronald Fisher gave a short 10-page overview of the nature of data-driven decision making in his 1926 publication The Arrangement of Field Experiments. Operational Surveys, Six Sigma, and the work of statisticians such as Edwards Deming show the importance of analyzing data against statistically calculated limits as a means of quantifying process change.

In summary, AI and ML should be viewed as means to improve existing business processes, not new business opportunities. First, you need to analyze the determinants of the process and ask the question, “If you could improve this decision by x%, what would the consequences be?”

Ignoring organizational change

Difficulty in implementing change management is a major contributor to overall AI project failure. There is a lot of research showing that most AI transformation cases fail, and the technologies, models, and data are just a few. Also important is the attitude of employees who put data first. In fact, the shift in employee mindset may be far more important than AI itself. A data-centric business can be just as effective as using a spreadsheet.

The first step to a successful AI initiative is building the belief that data-driven decisions are superior to intuition or tradition. The efforts of citizen data scientists have largely failed. This is because business unit managers and executives stick to common sense, lack credibility in data, and refuse to hand over decision-making power to the analytics process. As a result, ‘grass-roots’ research activities and top-down initiatives often ended with simple hobbies, curiosity, and career building rather than turning into a real business.

If there is one hope, it is that research on corporate change and related issues is being conducted extensively. A company is a place where executives are tested. The ambition of executives cannot be obtained by giving orders from above. You have to change your mindset and attitudes gently and skillfully, typically slowly, and acknowledge that different people may respond differently to directives for desired behavior. There are generally four key areas: Communication, Leading the Way, Engagement, and Continuous Improvement, which are directly related to the decision management process.

Changing corporate culture around the AI ​​space can be particularly challenging, given that data-driven decision-making is often counter-intuitive. Building trust that data-driven decision-making is superior to intuition and tradition requires an element of ‘physiological safety’ that only the most advanced leadership companies have mastered. When it comes to physiological safety, there has been a lot of talk about the acronym ‘ITAAP’, which means ‘everything about a person’. Successful programs often dedicate 50% or more of their budgets to change management. In fact, I would argue that it should be closer to 60%. An additional 10% should be put into the project-specific workforce analyst program within the Chief Human Resources Officer (CHRO) office.

Putting in the Hale Mary Pass early on

A data culture cannot be built overnight. Therefore, you should not expect an immediate conversion effect to an analytics project. Successful AI or ML initiatives require experience with people, processes, technology, and appropriate supporting infrastructure. Such experience does not accumulate very quickly. After years of concerted effort, it took everyone for a long time before IBM Watson beat Jeopardy and DeepMind AlphaGo beat the human Go champion on ABC’s quiz show.

The reason many AI projects fail is because they exceed the capabilities of the company. This is especially true if you are trying to launch a new or sales item powered by AI. Because there are too many variables involved in building something from scratch to achieve success.

Dirty Harry’s ‘Magnum Force’ quote ‘People know their limits’ applies to companies as well. The myriad of business decisions made every day in large enterprises can be automated with AI and data. Overall, using AI to improve small-scale decision-making can provide a better return on investment. A company might be better off starting with a low-risk investment in AI and ML rather than risk-taking an attempt to improve an existing process. The press room may not know this, but the accountant will.

Even if you’re already successfully using AI to make data-driven decisions, improving an existing model may be a better investment than introducing a new program. According to a 2018 McKinsey report, What is the Value of Better Models?, even a small increase in predictive power can have enormous economic value.

Inappropriate corporate structure for analytics

AI is not a Plug and Play technology that provides an immediate return on investment. It also requires a change in the mindset of the company as a whole and a change in internal institutions as a result. They are usually over-focused on talent, tools, and infrastructure, not paying too much attention to how the structure of an organization should change as to how it should change.

Some formal corporate structures supported vertically must achieve the critical mass, momentum, and cultural change needed to transform traditional non-analytics into data-driven ones. This requires not only a professional organization, but also new roles and responsibilities. Organizations of Professionals (COEs) will take different forms depending on the business environment.

In general, the bicameral model appears to be the most effective. Here, the core of AI responsibility is handled centrally, and the COE’s ‘satellite’ dispatched to each business unit is responsible for coordinating the delivery. This structure usually results in increased coordination and synchronization across business units and generally business units, leading to increased shared responsibility for AI transformation.

A COE, led by a Chief Analyst, can do well with responsibilities such as developing education and training programs, creating AI process libraries (data science methodology), cataloging data, building maturity models, and evaluating project performance. COEs are essentially dealing with tasks that benefit from economies of scale. This includes developing AI talent and negotiating with third-party providers, setting governance and technical standards, and creating an internal AI community.

COE representatives from various business units are ideally positioned to provide training, promote adoption, confirm AI-enhanced decision making, maintain implementation, and promote programs. The same goes for deciding when, where, and how to introduce AI initiatives into your business. The number of business unit representatives may be increased by projects run by the COE’s ‘SWAT Team’.

Don’t embed intelligence into business processes

One of the most common obstacles to extracting value from AI initiatives is integrating data insights into existing business processes. This ‘last mile’ problem is easiest to solve using a Business Rules Management System (BRMS). BRMS is an old technology that has been installed since the early 2000s and has been used as a means of deploying predictive models. BRMS makes the ideal decision point in automated business processes that are affordable and reliable. If your company isn’t using a Business Process Management (BPM) system to automate, streamline, or streamline your core business processes, stop investing in AI here. In this situation, basic things such as BPM and BRMS are needed first, not AI.

Most modern BRM systems include model management and cloud-based deployment options. In cloud scenarios, citizen data scientists can create models using tools such as Azure Machine Learning Studio and InRule BRMS, which are deployed directly to the business via REST endpoints. This cloud-based integration makes it easier to experiment with decision-making processes at a much more affordable cost than a full-featured AI program.

experiment failure

Now let’s look at the other side. How are users using AI to create new business models, innovate markets, develop new products, and boldly step into places no one has gone before? For startups receiving venture support, the failure rate is about 75%, making it difficult for them to adopt AI business models. If your AI-powered product or business initiative has a lower failure rate, you’re outperforming some of the best investors.

Even many tech experts fail. Former Google CEO Eric Schmidt revealed some of the ways Google was taking during his 2011 Senate testimony.

According to Google, in 2010, in order to help users understand the scale of the change, Google conducted 13,311 precision evaluations to determine whether the proposed algorithm change improved the quality of search results. We also conducted 8,157 experiments in which two sets of search results were provided to the experimental group and raters ranked the results, and 2,800 clicks in which a small number of actual Google users responded to changes. This process was ultimately determined to be useful for users based on the data. After going through this process, 516 changes were made. This change was applied to the Google algorithm as it was judged to be useful to users based on the data. Most of these changes are imperceptible to the user and have very negligible impact on the website, but each of these changes is only implemented when we are sure that it will provide a benefit to the user.

The failure rate of the proposed change is 96%.

The point here is that if there is an important lesson to be learned, failure is inevitable. What makes Google different is that Google’s data-driven culture allows other companies to learn from their mistakes. Please also pay attention to ‘experiment’, the keyword of Schmidt’s testimony. Experiments are how Google, Apple, Netflix, Amazon, and other technology leaders are leveraging the benefits of AI at scale.
A company’s ability to experiment is directly related to its ability to create and improve processes, products, customer experiences, and business models.

What’s next?

Just as the companies that failed to introduce machine manufacturing beyond handicrafts disappeared during the Industrial Revolution, companies that cannot adapt to the new environment amid the big changes in AI and ML will disappear. AI problems mainly arise from the technical part, and it is easy to blame them for failure because of technology, but in fact, most AI project failures stem from problems of strategy and execution.

This is good news for businesses in many ways. Companies are well aware of the old business problems behind the failure of AI projects. Although inevitable changes in culture, corporate structure and business processes are inevitable, you can be relieved that the direction has been set. The key is to steer the ship and avoid the rocks. Starting with small, simple experiments in applying AI to existing processes will help gain valuable experience before embarking on a longer AI journey. [email protected]

Source: ITWorld Korea by www.itworld.co.kr.

*The article has been translated based on the content of ITWorld Korea by www.itworld.co.kr. If there is any problem regarding the content, copyright, please leave a report below the article. We will try to process as quickly as possible to protect the rights of the author. Thank you very much!

*We just want readers to access information more quickly and easily with other multilingual content, instead of information only available in a certain language.

*We always respect the copyright of the content of the author and always include the original link of the source article.If the author disagrees, just leave the report below the article, the article will be edited or deleted at the request of the author. Thanks very much! Best regards!