10 Challenges of Implementing AI and How to Tackle Them

Artificial intelligence (AI) is set to be one of the big IT trends of 2021 and beyond. According to PwC, a quarter of companies in the US now report widespread adoption of AI, up from 18% last year. Meanwhile, another 54% are well on their way to this level.

AI can have applications across every level of your business, and it offers a wide range of benefits. For example, PwC's research found the key results seen by businesses that have fully implemented the technology include:

• Better customer experiences (86%)
• Improved decision-making (75%)
• Product and service innovation (75%)
• Cost savings (70%)
• Improved productivity/efficiency (64%)

But putting the right solutions in place to see these results can be tricky. There may be a range of barriers that can prevent AI deployments from reaching their potential, so it's vital firms are able to recognize these and put plans in place to overcome them.

These challenges tend to fall into one of three main categories. Firstly, there are the technical issues, which surround how the systems are implemented and managed in practice, often relating to the data you're using.

Then there are the business challenges that ensure your company is able to make the most of the technology. Finally, there are the cultural issues to consider in order to make certain your employees understand the solutions and are on board with any such initiatives.

Within these, here are ten specific problems you're likely to encounter during an AI implementation, and how you can address them.

1. Data

Any AI implementation is only as good as the data you feed into it, so ensuring this is of high quality is paramount. This starts with ensuring it's relevant and readable by the program. Ask where it's sourced from and in what form, and also make sure you're cleansing your data effectively to ensure accuracy.

To address this, more firms may turn away from traditional big data approaches towards smarter datasets. Gartner predicts that by 2025, 70% of organizations will move to "small and wide data" to provide more context for analytics and make their AI solutions less data demanding.

2. Security and storage

How your data is stored and protected is another major challenge, especially if you are working with very large data sets. You need to ensure you have the right storage tools so that data can be accessed quickly whenever it's needed. This may mean breaking down old silos in favor of solutions like data lakes.

At the same time, tools such as strong encryption that can protect data in motion and at rest, as well as protections such as access controls and data monitoring, are essential in ensuring these valuable assets are secure.

3. Infrastructure

As well as storage solutions, the physical infrastructure of your network may also need upgrading if you're to make the most of AI tools, both when it comes to data transfer and processing power. Most AI systems place high demands on compute resources that legacy systems may not be able to meet. Therefore, it's important to factor in the potential cost of upgrading these systems from the start, and consider options such as cloud services if these threaten to spiral out of control.

4. Building the right models

Constructing the most effective data model can prove highly challenging as this requires expertise and resources that many firms may not have. To be useful, algorithms such as machine learning tools need to be continuously trained. In particular, you need to be on the lookout for errors in your model and correct them as early as possible. Otherwise you may quickly find your AI solution keeps building on bad reasoning and its outputs become less and less accurate.

5. Combating bias

Many people may assume that handing processes over to machines will eliminate many of the human biases that often cloud our decision-making, but this is incorrect. In fact, many AI tools end up with the same inherent biases as the programmers who create them, or the data they receive.

There have been many examples of AI programs showing racial prejudice, often because the initial data they're trained on suffers from these problems. Therefore, ensuring you get information from unbiased sources and that algorithms can be clearly understood to spot any issues is vital.

6. Employee resistance

No technology can be effective unless people are willing to use it, and there may be a number of reasons why employees don't buy into AI solutions. From a lack of understanding about what the technology can do to worries about it taking their own jobs, you're likely to encounter some resistance. To tackle this, it's vital to educate people about the technology. Focus on showcasing what it can do and how it can make their own lives easier to get people on board.

7. Lack of talent

Skills shortages are a problem across the IT sector, and AI is no exception. In fact, research by Deloitte shows almost a quarter of advanced AI users (23%) report a shortage of talent, while overall, 39% of firms say a lack of technical expertise is a barrier to their AI adoption.

Attracting the best talent to your firm means more than just offering higher salaries - it's also about the culture you foster. However, smaller firms that find it harder to compete may have to look at outsourcing their AI projects, such as by licensing capabilities from other tech providers.

8. Presenting a business case

Getting skeptical board members to approve projects may be a challenge, especially for relatively new technologies where the hype often outweighs the reality. Managing expectations is a key part of this. AI can be quite a vague and wide-reaching term, so make sure you spell out specifically what you're looking to achieve and identify a few measurable key performance indicators to keep track of. This way, you can show senior executives in real-world terms what the impact of the system is and prove return on investment.

9. Integration into existing tools

Connecting AI systems to existing applications and business systems can often be more complex than many firms realize. It requires much more than a simple browser plugin or API to streamline data sharing. You'll need to work with your vendors to build a solution that works across the business, so considering their expertise in this area is a key factor when choosing suppliers.

Learn more: How Business Integration is Taking Over the IT Department

10. Legal issues

Firms must take extra care to meet legal and regulatory requirements when using AI, as the speed of technology always outpaces the law and it can be easy to inadvertently drift into breaches. For instance, if your tools are processing personal user data, it could be falling foul of legislation such as GDPR which has very strict rules on how customers' personal data can be used and when you must ask for explicit consent. It's therefore vital you have your data controller engaged with the project from the start to assess any potential impact on privacy and take steps to avoid such circumstances.