Signal to Action: What CHAINge 2026 Said About AI

Supply chain leaders met in Long Beach on September 29 and 30 for ASCM’s annual CHAINge conference, and AI received extensive attention in almost every session. Two sessions framed the subject best:

• Closing the Signal-to-Action Gap: Reinventing Supply Chain Resilience with AI. Gerald Jackson, Vice President of Supply Chain Product Strategy and Chief Evangelist at Oracle, explained what he calls the signal-to-action gap and how AI can close it.
• Forecasting in the AI Era: Best Practices, Maturity Barriers, and a Practitioner’s Map for Real Results. Steve Hainey, Senior Operating Director of the IBP Practice at Beckway, and Jeff Tackes, Head of Forecasting and Data Science Principal at Kraft Heinz, showed a packed room what it takes to put AI to work in forecasting.

Both presentations argued that AI pays off when it shortens the time between learning something and acting on it, and that a better AI model can’t achieve that acceleration by itself.

Oracle’s Gerald Jackson and the 21-day problem

Jackson opened his presentation with disruption figures. In 2025, 94% of companies had a revenue-impacting disruption, up from 76% the year before and 65% the year before that. He put the average cost of each event at about $8 million.

He defines the signal-to-action gap as the time between the first sign of an issue and the completed fix. His example began with three signals arriving on the same day:

• A major customer places an order that absorbs 40% of Q3 demand
• A supplier flags a problem with a critical component
• A carrier runs into trouble

Each signal lands with a different team. The shortage surfaces at the biweekly S&OP meeting on day 7, a leader hears about it on day 18, and executives approve a response on day 21. Jackson said that’s typical for a good, collaborative supply chain today.

With AI, he said, the same response can happen within an hour. An agent catches the delivery slip and checks it against every open commitment. It links the three signals, decides which customers get product first based on revenue and contract terms, and rebooks the carrier. The leader gets the full picture and three options. Jackson expects market share to go to whoever closes that gap fastest.

Resilience as a learning system

Jackson discussed his own doctoral research, which framed a resilient supply chain as “a continuous adaptive system” built on five elements:

• Anchor: Agree on the outcomes that stay fixed when everything else changes. He said most teams have never written them down.
• Sense: Monitor demand, operations, supply, and transportation, including newer signals like cargo-theft risk on specific lanes.
• Connect: Close the information gaps between teams, which he called the number one barrier to resilience.
• Adapt: Shorten the delay between knowing what to do and doing it.
• Learn: Improve with every cycle. Maturity comes in stages, and nobody skips several levels overnight.

He added two pieces of advice. First, don’t start by choosing agents. Ask your Lean or Six Sigma lead which processes carry the most waste, then find the AI solution that fixes them. Second, governance doesn’t have to slow things down. Oracle embeds agents in the transaction layer, where they inherit the same roles, approvals, and audit trail a new employee would get.

Oracle’s agent roadmap

Jackson demonstrated agentic applications in Oracle Fusion, including a design-to-source demo that read a CAD file, created the missing BOM items, and launched RFQs in a few minutes. He also introduced Fusion Claw, which Oracle calls a governed agentic execution runtime. It pairs large language models with operations-research optimization to model, simulate, and re-plan on live transaction data.

Oracle expects autonomy to arrive in stages. Agents start with work assistance, move to deep research that ends in a recommendation for a person to review, and eventually run on full auto within delegated authority. People set the objectives and stay accountable throughout.

How AI compounds strong data and magnifies weak data

The other top presentation was by Steve Hainey and Jeff Tackes. Hainey ran planning at Ashley Furniture starting the week before the pandemic, and Tackes leads forecasting at Kraft Heinz. Their session made the case that AI compounds the value of a strong data foundation and amplifies the problems created by a weak one.

Tackes started with a common ambiguity: ask three people in your business how forecast accuracy is calculated and you’ll get three answers. Ask AI, and you’ll get a fourth. AI scales that ambiguity, so his recommendation is to make data AI-ready by giving agents proprietary context they can use safely. That context needs four elements:

1. Definitions the AI can read, so it isn’t guessing from column names
2. Business context that explains why each metric exists and who uses it
3. A written workflow with the steps in order, so two people asking the same question get the same answer
4. Data quality fixed for the decision at hand, without waiting on a two-year cleanup

He also laid out a five-level maturity ladder for any single decision. It runs from access (what’s the forecast?) to assumptions (what’s baked in?), insights (why did it change?), options (what if we move the promotion?), and action (what should we do?). He recommended taking one decision as high as it will go before starting the next.

Audience polls showed varied AI maturity

In live polls, about 40% of attendees reported that they came from companies with more than $1 billion in revenue. Excel was still the most common forecasting tool, and one in five attendees didn’t use AI at all. Smaller companies tended to be further along on AI in forecasting than larger ones, mostly because they move faster.

Cautions learned from hard experience

In a backtest, AI can “know too much” because it has already seen the actuals. It can also be “convincingly right” on incomplete data. In one case, a missing data load made a model look 10% biased and sent a business scrambling over a problem that didn’t exist. Both speakers urged teams to prove value in dollars through forecast value add. A positive bias on a four-month lead time can quietly add a month of working capital.

The common ground

Despite their different vantage points, all three speakers agreed on four practices for effective AI adoption:

• Start with the outcome. Jackson’s anchor and Tackes’s single-decision approach both require knowing what you want to change before you pick a technology.
• Put AI close to the work and the data. Jackson wants intelligence inside the transaction layer, and Tackes wants definitions and context handed to the AI deliberately.
• Keep people in charge. Jackson reserves full automation for routine, repeated decisions, and Tackes said autonomous planning isn’t real yet. Both expect planners to spend more time challenging assumptions and less time copying data.
• Bring process discipline. AI won’t supply structure, so teams have to build it first.

Building the last mile

Both sessions ended on building. Oracle’s AI Agent Studio lets Fusion customers create agentic apps on their own transaction data. Hainey described building custom applications for clients covering forecasting, ABC segmentation, and safety-stock planning, delivered in months with no new licenses.

At RapidCanvas, we hear the same thing from supply chain leaders. An ERP backbone handles transactions and packaged agents cover common work, but the decision that moves results usually belongs to one team or group alone. It depends on that company’s SKU mix, its promotion calendar, and the outside signals only its planners watch. Turning that decision into a working application is the last mile.

RapidCanvas helps supply chain teams cover that last mile by building custom AI applications one decision at a time. Our Hybrid Approach™ pairs our data scientists with a proven agentic AI platform, so each solution fits your goals and tech stack and reaches production in weeks. The applications run on a company’s existing systems of record and respect the definitions, context, and governance both sessions called essential.

MTE-THOMSON, a Brazilian maker of automotive parts with more than 3,000 products, followed this playbook. Working with us on how much of each product to stock, the company had a demand forecasting application in production and was delivering ROI within three months. Forecast error improved by 9%, and better stocking saved almost $500,000.

Five steps you can take now

Supply chain teams have a strong orientation toward action now. Here are five ways you can action the insights outlined at CHAINge 2026:

• Identify your anchor. Get leadership to agree in writing on the outcomes you’ll protect when a disruption hits.
• Pick one decision. Choose the one with the most waste or the highest dollar impact.
• Write a one-page brief for that decision covering definitions, business context, a step-by-step workflow, and the data.
• Measure your signal-to-action time. Count the days from first signal to resolved issue, then set a target in hours.
• Prove it in dollars. Track forecast value add, working capital, and expedite costs so the next investment funds itself.

If Jackson’s figure of roughly $8 million per disruption holds, every day taken out of the signal-to-action gap goes straight to the bottom line. A team that gets one decision right this year will have its definitions and process in place when it’s time to automate the next one.

To learn more about AI solutions for the supply chain, or RapidCanvas, visit our website, request a consultation, or read verified customer reviews on G2.