Chris Burchett is Senior Vice President AI Transformation at Blue Yonder, a leading AI company for end-to-end supply chain transformation.
From top-secret defense projects to supply chains, I’ve been working with AI for over 20 years. During that time, I’ve seen many failures blamed on technology when they often stemmed from weak strategies.
Today, AI failures are the norm. According to RAND, 80% of AI projects fail; that’s double the rate of non-IT projects. PwC found that only 20% of companies capture 74% of the economic gains from AI. What’s the problem, and how can we fix it?
Uber is bucking the trend. Praveen Neppalli Naga, CTO at Uber, explained what they learned from experiments with agentic AI: “The biggest wins rarely come from automating one task. They come from rethinking an entire workflow.”
This is key, but it’s only one part of an effective strategy.
Defining An AI Strategy
Where do we begin? By not forgetting the lessons of the last digital transformation. This starts with leadership, and it means clearly defining:
1. The Project’s Alignment With The Business Strategy
This will vary by organization, but the objective could be related to carrier compliance, supplier collaboration or employee upskilling and retention.
2. The Degree Of Change The Organization Can Handle
It should be more than just automation in pursuit of productivity gains. Rather, try to reimagine entire workflows from start to finish. The value is in eliminating or reducing steps, gaps, handoffs and even roles. Due to its complexity, the supply chain is rich with opportunities to consolidate roles and redefine work. That could mean redeploying schedulers, whose jobs could now be done by agents, working as orchestrators guiding scheduling, load booking and carrier selection agents.
3. The Organization’s Ideal Speed
If the speed is too slow, you risk losing momentum and learning opportunities. However, if it’s too fast, you risk cutting corners and compromising the outcome.
Identifying Quick Wins
Once you’ve established these definitions, assemble a trusted team of internal and external experts and get started. You’ll learn more through execution than from analysis.
Identify the key pain points and start there. This is more likely to uncover areas where AI can deliver high-value outcomes and will make proving ROI easier.
Define the key measurements that the project should impact. What specific KPIs will change? What can’t we measure but could be impacted? (Think higher customer satisfaction, reuse of tribal knowledge, improved decision quality and upskilled employees.)
Set the bar high and strive for rapid results. Uber reported that they shipped AI agents in two weeks. That might be aggressive, but the timeline should make you uncomfortable. It will keep everyone focused on execution and iteration. Make sure everyone understands the goal: not perfection, but learning.
Prove value in the pilot before moving it to production. Don’t assume you’ll see it after scaling it in production, and be skeptical of vendors who tell you otherwise.
Technology Stack And Governance
As part of the process, evaluate technologies—their strengths, limitations and risks—with the intention of developing a reusable tech stack that has the capabilities you require.
This is not easy, especially in the supply chain with many siloed functions and relationships with trading partners. That’s why some organizations will work with strategic technology partners who have both AI and industry expertise.
A big question is: Do you go with frontier lab models or open source? That depends on the problems you are trying to solve, your organization and your strategy. In general, if the problems you are tackling are small-scale general reasoning problems, a frontier lab model can work well. But if the problems are complex, dynamic and require specialized domain knowledge, as in supply chain, then a specially trained open-source model may be better.
For example, our company tested both types of models in solving supply chain problems. In some key use cases, domain-specific models beat the generic LLMs by being both more effective and cheaper. The reason is that generic LLMs aren’t trained in the intricacies of things like running a warehouse or collaborating with suppliers. That data and information are scattered across enterprise systems and their frontline people.
Lastly, before deploying to production, rigorously test your model for errors, resistance to malicious inputs and compliance with organizational policies and regulations. Once in production, keep a close check on these and watch for model drift.
Monitoring The Key Metrics
It’s unlikely that you’ll see optimal value in the first iteration. This is where hill climbing comes in as you focus on minimizing a particular cost function or set of metrics. Once in production, it is important to continue monitoring. When you see KPIs degrade, even a little, address it immediately. Otherwise, you risk losing business value and trust.
Strategy Is A Start
A good strategy is not the only thing you need for success, but it is a precondition and a north star for everything else, amplifying the value of other contributing factors. So, make sure you have a clear vision and experts who understand it and can help execute it.
Remember the Uber lesson: It’s tempting to tinker with AI tools and automate tasks. Don’t. Imagine you’re an AI-native startup out to disrupt your business. Ask questions like, “Where are the big pain points, and what workflows could be eliminated or completely reengineered?” Then define KPIs for these high-impact areas and track them. (Unlike Uber, use business outcomes rather than token usage as the metric.) Use them as feedback to correct course as you iterate. This will help realize significant value quickly.
Finally, don’t neglect the ethical and responsible dimension of AI. With the EU AI Act now entering enforcement, you need comprehensive risk management and monitoring in place for this and other regulations that may apply to your enterprise. Doing so will help protect the value you create, reduce your legal and regulatory risk, and strengthen your customers’ trust in you.
This is just the start. Document and apply what you learn to your next project. Get it right, and you’ll be part of the elite group beating the odds and generating enduring business value.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
