By clicking “Accept”, you agree to the storing of cookies on your device to enhance site navigation, analyze site usage, and assist in our marketing efforts. View our Cookie Policy and Privacy Policy for more information.

Why do so many AI Programmes Fail?

Why do so many AI Programmes Fail?

Many companies have started a digital programme to transform their business and deliver improved results. But why do so many of them fail?

Most companies have started or planned to start a digital program. One they hope will transform their business, deliver improved results and move their company to higher levels of performance and shareholder value. So far, very few have seemed to achieve this.

In some cases, the early significant investments in Centres of Excellence (CoE) created 2-3 years ago, have been cut back. So why is this happening? These companies are full of smart data scientists with PhDs. Yet they seem to struggle to deliver the results they hoped for.

One recent interview with a senior executive of a manufacturing company was quoted as saying, “We have more pilots than Lufthansa.” This seems to be the case across many companies. Lots of great ideas and small pilot projects that have failed to roll out properly and deliver value.

Below I'll explain some common reasons for why this is happening.


First, companies traditionally think with a big program mentality. They form centers of excellence (CoE), appoint senior executives to run it, often from within the organization and with little experience of delivering a digital program.

The first step of these newly appointed executives is to run for the first safe haven, “We need lots of data. Let’s build a data lake. All the data vendors are advocating this.” Off they go, investing millions in creating a data lake, which often never gets completed the way they intended.

Data lakes are expensive, difficult to build and not always necessary.


The next step is often: “We need a culture change. Delivering AI programmes requires a different way of thinking across the company.” In comes their favorite consulting firm, armed with methodologies and mountains of advice on how to change the company. Before they know it, large numbers of expensive consultants are appearing all over their business, running workshops and hackathons to root out the best ideas to bring new levels of performance that the company is looking for.


The third and often the final nail in the coffin of the artificial intelligence (AI) program, is that companies want to apply all these new ideas using their near enormous and growing (daily) data lake. In come the armies of data scientists, analyzing data and kicking off pilots to prove their hypothesis. Each pilot soaks up the valuable time of the business unit’s staff, who still have their normal daily tasks to carry out, are pressed into this new program which will “change the company”, yet are only trying to do the best job they can.

A few months later, the company has many pilots underway, often showing encouraging results (according to the Chief Digital Officer). Yet the rollout fails to materialize because the focus of the pilot was on creating cool results, presented in endless powerpoint slides, without any thought going into how to scale the solution.

A year or so after all the above has taken place and millions have been invested, the company's business units become more dismissive of the AI program. In turn, budget holders in the business units are less interested in investing in the new ‘cool ideas’ generated by the CoE and the business experience limited change.



You don’t really need a data lake. AI programs rarely need all the data. They need data specific to the business problem, which is used to create models built around machine learning (ML) approaches. You need a comprehensive data integration strategy and the necessary internal processes aimed at providing fast access to good data for ML models, to deliver the results. This is easy to say and often there are stranded assets or processes where data is hard to access. New technologies such as smart edge agents today make it significantly easier to access this stranded data than has ever been the case.

Let go of the data lake!


Most companies realize that the consultants’ job is to be billable at a company and stay there. The cross-company workshops and hackathons keep them busy for months at a time. There is no question - they are very talented and bring good advice. However, if you ask them if they can use the new technology and methodologies available today that dramatically reduces consulting time, they will be less enthusiastic.

All companies have bright, motivated staff who understand how things can be improved within their area of responsibility. You need to give them the tools to test and implement these ideas quickly with low risk. This will help develop the momentum of change you’re looking for, with much less money spent on consulting hours.


It’s true, you can eventually tune an ML model to detect almost every type of anomalous behavior in a process or asset and dramatically reduce risk of failure or degradation. However, this could take a long time and deliver a return on investment which is questionable.

Other important points are that data scientists like to do cool things, develop new algorithms, deliver complex models no one has delivered before and solving previously unsolvable problems. This, of course, has value. However, it’s important to have these talented people focused on delivering value quickly. It’s not always a cool new algorithm that’s necessary to detect a failure or degradation in performance resulting in considerable savings. Often, a proven, off-the-shelf algorithm/model with well-structured and properly labeled data, will deliver the results needed in much less time.

At Arundo, we've seen the typical big company reactions across industries to the new AI era we’re moving into. Now it’s time to put the power in the hands of the talented staff each company have and let them create the difference and value sought by the senior executives.

Let ‘Speed and Value at Scale’ be the watchwords of your AI program, give people the tools to deliver this and the environment to try new ideas without fear of risk.

This working environment will bring the large scale company performance improvements sought after by the company executives who were sold on the promises of AI, which according to Forbes, the AI market will grow to $72Bn by 2021.