Machine Learning: 4 Adoption Challenges and How to Overcome Them
In the first quarter of 2022, global funding for artificial intelligence (AI) startups reached $15.1 billion, according to CB Insights’ State of AI report. However, machine learning (ML) algorithms can lead to counterproductive results when deployed unnecessarily.
Here are four common challenges companies implementing ML-based systems may encounter, along with some expert advice for maximizing the impact of algorithms while avoiding missteps.
1. Find an ML use case
For some companies, the first AI and ML adoption issues come before they get started. Machine learning is a broad, multi-faceted discipline that permeates most aspects of artificial intelligence. It paves the way for many potential applications, from intelligent process automation (IPA) and natural language processing (NLP) to computer vision and advanced data analytics.
Choosing a use case worth investing in is easier said than done. In this regard, O’Reilly’s 2020 AI adoption in the business survey ranked identifying use cases as the second most relevant challenge (mentioned by 20% of respondents) .
[ Also read The AI revolution: 4 tips to stay competitive. ]
Beyond the usual recommendations on defining your business goals – i.e. what you want machine learning to do for your business (improving operational efficiency, improving your products or services, mitigating risks) – a rule of thumb for choosing an appropriate ML use case is “Follow the money.”
Target the most strategic business functions and generate maximum profit for your organization, based on its size and industry. Examples may include computer vision-guided assembly for manufacturers or data-driven marketing for retailers.
Another selection criteria focuses on addressing weaknesses in your business, such as process bottlenecks. You can identify them through proper BPM surveys and KPI assessments.
2. Select the right data
Data is the fuel for machine learning. ML systems have to deal with huge datasets to be properly trained. The reliability of the results depends on the quality of the datasets and the training process itself. Here are some recommendations to consider:
- Rely on qualified data scientists to select the appropriate data sources, whether external or collected from corporate systems. Implement effective data management and governance strategies to ensure data is collected and stored correctly.
- Select a core feature subset from your datasets so that the training phase can focus on the most relevant variables and ignore redundant metrics, making it easier to interpret the results.
- Train your ML system with multiple subsequent data samples (usually referred to as training, validation, and test sets) to monitor and improve its performance under different conditions while avoiding overfitting issues, i.e. when algorithms are “tuned” on specific datasets but perform poorly with others.
3. Complement ML with human talent
Machine learning algorithms can still behave unpredictably after training to prepare for data analysis.
This lack of clarity can be a problem when using AI in decision-making leads to unexpected results. As Harvard Business School reported in its 2021 report Hidden Workers: Untapped Talent, automated ML-based recruiting software has rejected many candidates due to overly rigid selection criteria.
ML-based analysis should always be complemented by continuous human supervision.
This is why ML-based analysis should always be complemented by continuous human supervision. Talented experts should monitor the operation of your ML system in the field and fine-tune its parameters with additional training datasets that cover emerging trends or scenarios. Decision making should be ML driven and not ML imposed. The system’s recommendation should be carefully evaluated and not accepted at face value.
[ Read also AI ethics: 5 key pillars ]
Unfortunately, combining algorithms and human expertise remains difficult due to the lack of ML professionals in the job market. The scale of the skills shortage worries policymakers around the world. Investments in staff training and partnerships with other organizations interested in adopting machine learning can help solve this problem.
4. Manage resistance to change
Business inertia, resistance to change, and lack of preparation could be the worst enemy of ML adoption. According to O’Reilly’s study, as mentioned above, corporate culture is the biggest barrier to implementing AI-related technologies. This usually implies that senior management is unwilling to take investment risks and employees fear labor disruptions. To ensure buy-in from stakeholders and staff, consider implementing the following best practices:
- Instead of betting on moonshots, start with small-scale ML use cases that require reasonable investment to get quick wins and attract leaders.
- Foster innovation and digital literacy through corporate training, workshops, benefits and other incentives.
- Establish centers of excellence to oversee the implementation of ML in your organization, including the business and technology changes needed to integrate these tools into your business workflow and software ecosystem.
Fly high without getting burned
Machine learning can take businesses to new heights through interactive NLP-based solutions, business intelligence software, and process automation tools. However, adopting this powerful technology within a robust management framework will save businesses from many future challenges.
[ Check out our primer on 10 key artificial intelligence terms for IT and business leaders: Cheat sheet: AI glossary. ]