Supply chain concern #1: Unpredictability. The cure? AI-enabled supply chain planning

Given everything that’s unfolded in 2020, it’s not surprising that the top concern for supply chain planning teams is unpredictability. What might be surprising, however, is the lack of consensus on what to do about it.

That was our takeaway from ASUGForward, Americas' SAP Users' Group virtual event that replaced its traditional physical counterpart alongside SAPPHIRE 2020. The event helped us understand the current mindset of the SAP customer community and gave us the opportunity to reflect on some really interesting attendee survey data.

Polling questions reveal a gap in understanding about AI

Reflecting the point about uncertainty made earlier, nearly 50% of respondents said that “supply chain shortages and demand unpredictability” was their biggest concern. No other answer came close:

Operating in this 2020 pandemic, what business has not been whipsawed by the noise and variability in their demand and supply? Our conversations with consumer products companies confirms that this is a top concern. The disruption hits at all phases of the COVID-19 phenomenon – at the beginning, ‘sustained middle’ and during recovery, which will vary in intensity depending on the type of SKU being forecasted.

The surprise came from the second polling question, where attendees were asked about their intention to implement AI-enabled supply chain planning software:

The responses show us that our industry has not done a good job of articulating and demonstrating the most effective antidote to unpredictability – implementing predictive, probabilistic planning tools that use AI. 

We were astonished to see that despite the huge concerns about unpredictability, roughly only 25% respondents have implemented or plan to implement AI-enabled supply chain planning software in the next 12 months.

The Snake Oil problem

In 2018, Gartner made this AI prediction: "Through 2022, 85% of AI projects will deliver erroneous outcomes due to bias in data, algorithms or the teams responsible for managing them…" In the context of a thought leader such as Gartner saying that 85% of AI projects will fail, it’s understandable that supply chain leaders would be reluctant to take on an AI project.

But why such a high failure rate? We think it’s because a lot of ‘AI Snake Oil’ is being sold, leading to high-cost, low-impact projects. For example:

• The “Do-it-all” platform – just add… everything
• The “Toolkit.” Almost built. Just assemble at home
• The “We help you do it yourself.” @$500/hour.
• The “CPG is just like any other industry. It is easy.”
• The “Roadmap” – custom build these ten applications

To help identify real AI in a landscape of AI Snake Oil, we’ve created a list of key questions that any company can use as they evaluate AI providers.

1. Is this a project or a product?
2. Can you show me your data model and accelerators?
3. Do you have proven data science models?
4. How long until I see value?
5. What investments have you made in your people and your tech stack?

The two final questions raise issues that come up frequently. For the question of “How long until I see value?” there is an obvious caveat pertaining to the state of your data infrastructure. But in general, we urge supply chain leaders to push vendors for a clearer definition of value and much shorter implementation time frames on their AI initiatives – and be wary of extended time to go-live.

How long to realize value from AI/ML?

AI/ML has sparked everyone’s imagination around the promise of new untapped value. However, realizing value, timely or otherwise, often remains a challenge.

Firstly, value measurement in the ‘counter-factual’ world of AI/ML risk predictions is a complex subject that is often treated too simplistically during the sales and budgetary processes. We advise customers to ask all vendors to explain how they would measure and attribute their product’s own true impact on supply chain drivers such as Lost Sales Recovery, Inventory Reductions and Logistic Costs.

Furthermore, AI/ML application use cases tend to be experimentation-heavy and there is no shortage of AI/ML concepts that fail to scale to production (a.k.a. proof-of-concept purgatory).

Successful value realization requires a certain repeatability of prior R&D successes. How does one quickly ascertain whether underlying data meets standards for target variable predictions to be meaningful? How does one ensure that AI/ML results are actionable for planners and operators to adopt as part of their workflow? If any of these are not identified and remediated quickly, a customer’s path to value is delayed, often indefinitely.

So how long until one can expect to see hard value? Rule of thumb – anything longer than 90 to 120 days (3-4 months) signals that a vendor may be spending too much of their R&D time on your dime (often masked as co-development). Results range based on use case complexity.

In order to support customers through their value journey from pilot configuration to production support, successful vendors invest heavily in product development and deployment teams spanning full-stack data scientists, data engineers, software engineers and supply chain domain specialists. Considerable development effort should already have been spent to ensure product features are generalizable and repeatable.

Finding success with intelligent planning tools

The skepticism about AI-Snake-Oil and marketing hype by supply chain leaders is understandable. However, there are plenty of examples of organizations successfully handling the current uncertainty by using intelligent planning tools. In order to be successful, an AI/ML application vendor needs to be explicitly clear on their target functional use case, the value impact timeframe and how they will measure it and prove it.

To learn more about Noodle.ai’s approach to realizing value from supply chain AI, watch my ASUGForward presentation on Tackling Uncertainty in Supply Chains with Enterprise AI® or contact us direct. Happy to open up a dialogue.