AI and the Potato Industry

Situation

Jordan is operations manager at Riverbend Storage Co-op, a potato storage facility in South Alberta. The co-op operates six storage sheds with a combined capacity of about 24,000 tonnes. Temperature, CO₂, O₂, and humidity are critical: a bad run can mean sugar ends, pressure bruising, or spoilage, and the board has made it clear that “no more surprises” is the priority.

Two years ago, a temperature spike in Shed 3 was caught too late; the co-op had to move a portion of the crop early and take a hit on quality. Since then, Jordan has relied on manual spot-checks and a few wall-mounted thermostats. The board is now asking whether the co-op should invest in proper monitoring: IoT sensors (temperature probes, gas sensors, humidity) plus software to turn that data into clear “storage health” readings and alerts.

Jordan has strong agronomy and storage experience but no software team. A vendor quoted $85,000 for a turnkey IoT and dashboard package for all six sheds. The board is hesitant to spend that without seeing how different setpoints and ventilation choices would play out first. Jordan has heard that AI can help generate code and simulations—including for temperature and gas behaviour in storage—and is wondering whether to try that before committing to the vendor.

Decision 1: How to explore storage behaviour before spending on IoT

Jordan wants to simulate how temperature and gases (CO₂, O₂) would behave in the sheds under different cooling and ventilation scenarios—before buying sensors or signing the vendor contract. The co-op has no in-house developers.

• Option A: Hire a consultant to build a custom simulation (estimated $15,000–$25,000). Pro: expert-built, defensible. Con: cost, delay, and the model may still need changes later.
• Option B: Use an AI coding tool (e.g., ChatGPT, Copilot) to generate simulation code based on Jordan’s descriptions of shed size, airflow, and storage rules. Pro: fast, cheap, easy to tweak. Con: Jordan must validate the logic and outputs; wrong assumptions could mislead the board.
• Option C: Do nothing with simulations; rely on the vendor’s generic demo and buy the IoT package. Pro: no internal effort. Con: no co-op-specific “what-if” analysis; board may still be uneasy.

What should Jordan recommend to the board?

Decision 2: Validating and using the AI-generated simulation

Jordan chose to try Option B. After several prompts, the AI produced a Python script that estimated temperature and CO₂ over time given shed dimensions, initial conditions, and ventilation settings. The outputs look plausible for the first 48 hours, but Jordan is not a programmer and cannot vouch for the equations inside the code.

• Option A: Run the simulation as-is for the board and use it to support the case for IoT. Pro: quick, shows initiative. Con: if the model is wrong, credibility is at risk.
• Option B: Pay an external expert (e.g., ag engineer or programmer) a one-time fee to review the AI-generated code and results before any presentation. Pro: more confidence. Con: extra cost and time.
• Option C: Use the simulation only internally to test ideas; do not present it formally to the board. Pro: no reputational risk. Con: the board does not see the benefit of the AI approach and may still favour the vendor-only path.

How should Jordan handle the AI-generated simulation before the next board meeting?

Decision 3: Building the logic that turns sensor data into actions

The board approved a smaller first step: IoT sensors in one shed (Shed 1) with a basic dashboard from the vendor. The vendor’s dashboard shows raw readings (temperature, CO₂, O₂, humidity) but does not implement the co-op’s own rules—e.g., “If core temp > 4°C and CO₂ > 5%, recommend ventilation for at least 2 hours.” Implementing that logic in the vendor’s system would require a custom project and another quote.

Jordan could use AI again to generate a small script that reads the sensor data (from an API or export), applies the co-op’s rules, and outputs a simple “ventilation recommended” or “conditions OK” message—or even a minimal dashboard. The alternative is to ask the vendor to add this logic (more cost, longer timeline) or to keep using the raw dashboard and apply the rules manually.

• Option A: Use AI to generate the “storage function” code and a simple pipeline that pulls sensor data and outputs recommendations. Run it in parallel with the vendor dashboard and compare. Pro: low cost, full control over rules. Con: Jordan must maintain and validate the code; if it’s wrong, operators might follow a bad recommendation.
• Option B: Request a formal quote from the vendor to implement the co-op’s rules in their system. Pro: single system, vendor support. Con: cost and delay.
• Option C: Keep using only the vendor’s raw dashboard; Jordan or staff apply the ventilation rules manually from the numbers. Pro: no new code, no new risk. Con: human error and no automated alerts.

How should Jordan decide between AI-generated logic, vendor-built logic, and manual application of rules?

Decision 4: Scaling to all sheds

After three months, the AI-generated “storage function” in Shed 1 has matched staff judgment on when to ventilate, and there have been no incidents. The board is now asking whether to roll out sensor-based monitoring and the same logic to the remaining five sheds. Options include: (1) extending the AI-generated pipeline and rules to all sheds with the current vendor sensors; (2) buying the vendor’s full dashboard and asking them to implement the co-op rules for all sheds; or (3) a hybrid—e.g., vendor sensors everywhere but AI-generated logic for the “ventilation recommended” alerts, with the vendor dashboard as backup.

What should Jordan recommend for scaling: double down on the AI-built approach, switch fully to the vendor, or choose a hybrid? What criteria should drive the decision?

Discussion: The Overarching Dilemma

Riverbend has limited budget and no in-house IT. AI has already helped Jordan produce simulation code and storage logic that the co-op could not easily afford otherwise. At the same time, the co-op’s reputation and member trust depend on storage quality and avoiding another “surprise” event.

How should Jordan and the board balance the benefits of using AI to build and extend potato storage tools (temperature, gases, IoT) against the risks of relying on code that no one in-house can fully verify? When is it appropriate to rely on AI-generated logic for critical storage decisions, and when should the co-op pay for vendor or expert-built systems?

Teaching note

This case is designed for discussion. There are no single “correct” answers. Encourage participants to argue for different options at each decision point, to consider stakeholder (board, members, staff) perspectives, and to weigh trade-offs between cost, speed, control, and risk in the context of potato storage and AI.