Artificial intelligence (AI) technologies have the potential to exert a profound influence on supply chains. However, data shows that the technology is certainly not being used to its full potential — or even close to it.
A recent McKinsey & Company survey of 2,300 executives showed a 25% year-over-year increase in the use of AI in standard business processes. Nevertheless, drawbacks such as the high cost of creation and limited creative thinking can limit the impact of AI and lead to costly project failures. Furthermore, in a survey carried out by Boston Consulting Group and the Massachusetts Institute of Technology, 70% of executives polled said their AI initiatives produce only a small effect.
There are four main business challenges holding back organizations from true achievement. These include fear of missing out (FOMO), deployment difficulty, data readiness and a lack of talent. Following are some solutions:
1. Soothe your FOMO with measurable strategic guidance. One common error that firms make is rushing to act on the hype surrounding AI instead of starting projects with a clear strategy. Executives have read or been told that AI can help their businesses, and they worry that, if they don’t act now, they will miss out on the benefits their competitors are enjoying. However, having a well-thought-out vision is crucial for any project, especially one that includes new technology.
Start your strategy by building a portfolio of use cases that demonstrate the value of technology to the supply chain and include deliverables in a known timeframe. Consider options including route optimization, sales forecasting, product categorization, safety stock calculation, supplier management and warehouse management. Prioritize these use cases by potential return and time-to-value in order to guarantee that investments link to tangible outcomes. Quick wins show the potential of AI technologies and build excitement within the organization.
Pre-project planning should include current and future applications that could benefit from AI technologies. And the final business impact of any supply chain implementation must be measurable. The teams involved need to describe as clearly as possible the value proposition and determine the relevant key performance indicators (KPIs) to measure the future solutions and the financial impact, such as a cost reduction or revenue increase. Return on investment should then be estimated by evaluating project-related costs including development, data preparation and ownership, and talent.
2. Opt for packaged applications to ease deployment woes. After years of hype, the business community is starting to have doubts about whether AI can deliver. AI projects can be costly and difficult to deploy, and margins are tight. Therefore, organizations are becoming more reluctant to invest in these technologies. The solution to this is to choose your AI options wisely. Depending on the use case, there are several to consider:
- Focused solutions: These are packaged applications with pre-built AI models. They are easy to implement because only the input data is required. According to Gartner, by 2023, 85% of AI solutions by vendors will focus on concrete domains and industry verticals.
- Embedded AI solutions: Many common applications, such as advanced planning, transportation management and warehouse management systems, have embedded AI capabilities. These combine traditional techniques, such as algorithms and statistical calculations, with AI technologies.
- Custom solutions: Organizations can build custom AI models using open-source platforms, frameworks and application programming interfaces.
Adoption is a whole other challenge. Employees tend to resist new ways of working and changes to existing practices. For example, AI-based route optimization tools such as the UPS On-Road Integrated Optimization and Navigation change the way carriers drive their trucks. This particular solution predicts costs and time losses that drivers don’t perceive. Drivers, however, are sometimes resistant to trusting the AI models to be as good as or better than their traditional routes and schedules. Hence, change management is a must to show the domain experts the quick wins delivered by AI and the ability to scale the technology.
3. Work with your best data first. Data readiness is a major challenge for any AI project. Data may exist, but it’s often irrelevant or unusable. On average, only 3% of a company’s data meets basic quality standards. Companies should focus on data quality and relevance rather than volume. Contrary to common beliefs, AI doesn’t need a huge quantity of data. The technology can add value by using a reasonable amount of good-quality data at the start and then enrich the data pool gradually as new data is made available. Many AI-embedded supply chain solutions have optimized their AI capabilities to use smaller data sets to tackle the lack of data and make AI more accessible.
For instance, in machine-learning projects for product categorization and clustering, some companies add many data sources without being sure of their relevance to the categorization. Such data points add noise to the machine-learning model and create bias in the generated clusters. It’s better to begin with only some data that definitely is relevant. A lack of data quality regularly leads to project failure.
Exploring other data sources also is an option. External data providers offer quality datasets, including weather, macroeconomic and demographic data. For instance, for forecasting fresh food and beverage sales, adding weather data to the AI model enhances the quality of the forecast. Furthermore, simulations can be used to complete data sets that aren’t sufficiently diversified. The goal is to interpolate data to correct errors and simulate atypical situations that had not occurred in the real training data. This data, in addition to internal data, will form an enhanced input for AI models.
Organizations need to anticipate that data preparation will be time-consuming and take up a big part of AI projects. Data preparation includes cleaning and correctly integrating the data to be certain of its quality and completeness. In addition, it is key to check the sources of data and their frequency of updates to assess data sustainability. AI experts also need to have a direct collaboration with business experts to identify existing data sources, detect anomalies and update the business experts.
4. Train internal talent to fill AI roles. In a 2019 Gartner Research Circle survey, 56% of respondents cited a lack of staff skills as the first obstacle to AI deployment. To skirt this challenge, companies should work on retaining existing AI experts and identifying internal candidates who can be trained to fulfill new AI roles. Typically, existing application development professionals are a good talent pool. Curious application engineers also can become good data scientists.
If there is not enough knowledge to train these employees in-house, companies can pay for employees to participate in online training programs or graduate programs at local universities. This means that companies need to budget for AI training. While it is critical to educate new and existing AI recruits, it also is helpful to educate the business experts who know supply chain but don’t quite understand how AI can help.
More supply chain solutions have embedded AI, which minimizes the need for AI experts. Moreover, application developers can create and embed AI capabilities into existing applications or future application projects, thanks to the democratization of building AI systems. This also means that companies can hire independent contractors or information technology (IT) service providers to fill in talent gaps.
Of equal importance are organizational aspects. AI experts can be grouped in an AI team that supports the different supply chain business units. This transversal team should participate in IT supply chain projects to determine if AI techniques can solve the challenges at hand. Many complex problems are not viewed as problems that AI can fix. But once they are, any suggested projects should be added to the use cases portfolio.
But involvement should not stop there. The AI team must be involved on a continuous basis. Business experts should keep the AI experts updated about changes in business conditions. As an example, demand planners may notice a change in the market not identified by the external data, as used in the machine-learning-based sales forecast. AI teams can analyze these new trends and add other relevant external data to the existing machine-learning model, such as gross domestic product growth estimates, to solve the puzzle.
Use these solutions to help your company overcome the initial hurdles of AI projects and build a strong foundation for successful ongoing AI programs. Every successful use case, in turn, opens up additional opportunities. The biggest step forward is to move beyond individual use cases to ensure that more projects get to production.