Facing down some key AI challenges

Getting a handle on the most common barriers involved in implementing AI can help you decide whether—and how—to implement it.

8-minute read

Artificial intelligence (AI) has advanced significantly over the past few years and now offers an explosive range of benefits for businesses. Some are within easy reach for many SMEs, such as automatic note-taking and transcribing, synthesizing complex information, or drafting policies. Others, such as the ability to improve or automate estimates, forecast demand, or optimize scheduling, will require a greater upfront investment of time and money.

Either way, there are limitations and caveats that it pays to know about. Understanding these can help you manage your expectations and avoid incurring unnecessary risks or costs.

First, it’s essential to distinguish between generative AI and AI-powered solutions. Generative AI is meant to produce content—for example, you ask ChatGPT or Gemini AI a question, and it generates an answer. Traditional AI, on the other hand, is generally based on machine learning (where you train your AI using data), and can be put to work analyzing, predicting, automating or solving complex problems.

When it comes to AI solutions, data is key.

What are some downsides of AI for small businesses—and how can you overcome them?

Some of the “downsides” are really more like caveats or obstacles on the path to adoption. For example, many traditional AI solutions need to be trained on vast amounts of data before they can be useful—a process that can be expensive and time-consuming. The highly skilled resources, software and hardware that you may need can also be very costly. Here is a breakdown of some of the issues and suggestions on how to address them.

Challenge	Issues	Response
Cost	When it comes to AI solutions, data is key. But collecting and cleaning the huge quantities of structured digital data that AI solutions need often requires well-paid specialists you may not have on staff—such as data engineers or even a machine learning operations team. You may also need servers and computing power. All this can add up.	Carefully set objectives for your AI and make sure you know where and how you want to apply it in your business. Then evaluate the feasibility, complexity, risks and benefits of the project. Weigh the benefits against the expected cost to make sure the payoff will be adequate.
Implementation	If your business is paper-based, your first step will be digitizing all the relevant information. Then you’ll need to spend a fair bit of time using the data to train your AI. Behind every successful AI solution is a significant amount of data prep, training and oversight—particularly if you’re interested in applying AI to a core process.	If you don’t have historical data, an alternative is to buy synthetic data. Synthetic data can replicate the characteristics and patterns of real data without relying on actual consumer or corporate data. The upsides are privacy, availability, and the possibility of enriching the data set. However, if the data is not completely relevant to your business, it won’t be as effective. Many paths to AI adoption require a digital transformation first. Ensure that you have modern systems to digitize your information and look for systems that are already powered by AI.
Time to benefit	It can take a long time for a new AI solution to begin producing quality results. The “first run” isn’t always accurate, and you may need to retrain. It may take longer than you hoped to begin realizing the expected operational benefits or see a return on your investment (ROI).	Figure out how long it may take to compile and digitize your data (if necessary) and to train and retrain your AI solution. Consider this information together with the projected cost before you make the decision to start. You also need to consider the level of risk in the event that the AI is inaccurate in its job.
Ethics	Ethical concerns range from bias in the data used to train your AI to legal, copyright and transparency issues—and sorting through these takes time. For example: Have you used personally identifiable information? (Often, you shouldn’t.) Is there bias in your data? Is it diverse enough? Has the data been approved for use in training? Even if you’re just using an off-the-shelf generative AI tool, do you know what its sources are, or if it’s relying on copyrighted information? Is there intellectual property in your training data or data sources that you don’t want syndicated out to your workforce or customers?	Clean the data you feed the AI and verify that it’s sound—that is, unbiased, free of sensitive, personally identifiable information, and approved for use in training. Use diverse data sets with good representation and get permission from the owner of the data, if applicable. All this will help ensure you can trust the results you get. In good data governance, the data enrichment stage is often forgotten—enhance your data with more accurate and diverse information to better train your AI.
Human involvement	It’s been said that humans should stand on the shoulders of AI—in other words, you can’t remove humans from the equation. You need a real person on the training end to gather and clean your data, and to feed it to the AI. You also need a real person on the receiving end to validate the output and ensure the AI did not hallucinate or make up information. You could also think of this as a reliability and trust issue—you can’t just leave your AI “unleashed.”	You need to adopt some amount of data integrity in your business, even if you’re simply using generative AI to synthesize notes. If your notes are not detailed or organized, the results may not be as useful. Establishing quality data pipelines and data warehouses that can effectively train AI models can be out of reach. Buying AI-enabled or -powered commercial solutions is a more efficient path to adoption: modern systems are already driving productivity by providing suggested “best next actions” for employees to consider (versus having to analyze large amounts of information themselves).
Maintenance	Historical data is just that: historical. It doesn’t always consider brand new information (depending on the approach or algorithm) or know what the future holds.	Plan to continually feed your AI with data. Your results will improve over time as you provide it with the structured information it needs.

Considering all this, for most SMEs, the most cost-effective path to AI adoption is to invest in commercially available technology that is powered by or embedded with AI—unless your business has the interest, capacity and funds to staff a team of data engineers and data scientists to build out pipelines and workflows, data warehouses and more. That said, the downsides of off-the-shelf AI could include limited customization, reliance on the vendor for updates, and limitations in flexibility, integration and long-term adaptability.

If you’re buying a commercial solution, make sure you trust it.

Is AI dangerous for SMEs?

Despite its many benefits, there are a few hazards associated with AI—but it’s not hard to avoid them if you know what they are. For example:

It’s dangerous to trust that AI is always accurate. A human must validate the results.
It’s dangerous to be unaware of what data sources were used to train the AI. Understand what data is being used, where the sources are, what is in the data, and where it is hosted.
It’s dangerous not to use quality, complete, comprehensive data to train your AI. Carefully assess the data you use for completeness, bias and risks.
It’s dangerous to feed your business intelligence to a commercial AI solution without knowing where that data may end up or how it will be used commercially. Make sure you’re aware and comfortable with it.

If you’re buying a commercial solution, it should offer a closed environment that is secure and not monetize your data—in other words, it should protect and safeguard your data. Make sure you trust it.

What if you’re not ready to adopt AI yet?

A mature business is a data-driven one, and it really does start with building a digital business. Many SMEs, especially those that still rely on paper, will need to undergo a digital transformation before they can begin to incorporate AI.

You can take some first steps by:

Implementing solutions and systems that maintain and store your data
Experimenting with simple AI solutions, like chatbots
Collaborating with AI consultants for advice on high-value use cases
Evaluating your tech stack against your business, ensuring that each core process or function is well-supported by technology and systems

Will these challenges be resolved soon?

AI is evolving by leaps and bounds, but its fundamental reliance on quality data is unlikely to change any time soon. That reliance will continue to be many SMEs’ biggest hurdle.

In addition, risks we don’t even know about yet may enter the picture—threats from bad actors, deep fakes and more. Against this backdrop, protecting your data through strong cybersecurity measures is critical, particularly if the data contains intellectual property.

Ultimately, if you’re contemplating adding AI to your business, the time to start is now. Just make sure you know what you’re getting into so you can choose the tool or path that best suits your needs, goals and budget.

Next step

Discover how to prepare your business for AI and plan its implementation by downloading BDC’s free guide Getting your business AI-ready.