Why Do Teams Go Back to Spreadsheets After Buying AI Planning Software?

Many enterprises are investing heavily in AI planning software, expecting to revolutionize forecasts, optimize supply chains, and streamline operations. Companies like Garrett Motion, PMI, and Pandora have adopted cutting-edge solutions powered by Large Language Models (LLMs) and retrieval-augmented generation (RAG) techniques, aiming to elevate decision-making with next-level predictive accuracy.

Yet, a frustrating and paradoxical trend emerges: despite the advanced AI tools in place, teams often revert to old-school spreadsheets. Why? This blog dives into the root causes, particularly the critical role of trust issues and how a lack of explainability, consistency, and alignment with established business logic leads to shadow processes thriving alongside—and often instead of—the AI systems.

The Trust Bottleneck: The Real Barrier to Enterprise AI Adoption

AI promises more accurate outcomes than traditional methods, but accuracy improvements alone rarely translate to user trust. The accuracy story is subtle: you can have models that statistically outperform human forecasts but are rejected by planning teams due to distrust.

This trust gap stems from a few critical factors:

• Can't verify logic: Planning teams can’t trace AI recommendations back to understandable rules or transparent data.
• Hallucinations and fabricated certainty: LLMs and RAG-augmented systems occasionally produce outputs with confident but fabricated facts.
• Misalignment with business logic: AI may optimize based on narrow goals without incorporating nuanced constraints or real-world considerations.

Garrett Motion: Innovation Meets Skepticism

Garrett Motion, a leader in automotive technology, deployed AI planning tools to enhance supply chain responsiveness. Despite promising early results, planners reported reverting to spreadsheets for scenario simulations. The AI system’s suggestions often felt like a “black box,” and planners expressed frustration at not being able to validate the logic driving the forecasts. This blind https://o9solutions.com/articles/can-you-really-trust-your-ai spot led to fear of costly errors—a classic illustration of trust issues impeding AI adoption.

PMI’s Journey: Explainability as a Deal-breaker

At PMI, explainability became a decisive factor. Their AI solution incorporated cutting-edge retrieval-augmented generation, allowing it to combine vast textual and numeric data sources dynamically. Yet, without clear audit trails linking outputs to data and logic, planners resorted to “shadow processes”—manual workarounds and spreadsheets—to verify results. Despite accuracy being objectively higher, planners’ lack of confidence forced them to maintain parallel workflows.

Pandora and the Perils of Consistency

In Pandora’s fast-moving retail environment, consistency between AI outputs and known demand drivers was critical. Variability in AI-generated forecasts for the same prompt eroded confidence. The inability to consistently reproduce results made spreadsheets seem more reliable, as they provided predictable, if less sophisticated, estimates. This illustrates a key flaw in many AI deployments: models that appear accurate on aggregate but behave inconsistently at the individual forecast level.

Why Accuracy Gains Alone Don’t Build Trust

It’s tempting to assume that improving accuracy metrics is enough to get users on board. However, enterprises quickly discover that:

1. Without transparency, even accurate models spark mistrust. Planners want to know “why” a number is recommended—not just what it is.
2. Errors hidden behind confident AI output can cause disastrous decisions. Hallucinated facts or “fabricated certainty” from LLMs present invisible risk.
3. Changing model outputs without clear rationale breaks established processes. Spreadsheets, despite being manual, offer traceability and control that AI systems often lack.

This explains the paradox: accuracy gains create technical value but insufficient business value unless paired with explainability and verifiable logic.

Hallucinations and Fabricated Certainty: The Hidden AI Pitfalls

LLMs are powerful but prone to hallucinations—confidently generated outputs that are factually incorrect or fabricated. When integrated into planning systems via RAG, this can introduce:

• Incorrect forecasts justified by non-existent business rules
• Fabricated “data explanations” that can’t be audited
• Variance in outputs across similar prompts, confusing users

These hallucinations feed mistrust, encouraging users to revert to “what they can verify” through familiar spreadsheets. In the worst cases, teams keep separate shadow processes to double-check or override AI outputs, defeating the purpose of automation.

The Audit Question: What Would I Show in an Audit?

This is a critical question too many AI projects neglect. Without clear audit trails showing exactly how planning decisions derive from data, rules, and logic, auditors—and users—cannot confidently approve AI outputs. The resulting gap magnifies trust issues. Teams stuck wondering “can I verify the logic?” take comfort in spreadsheets that lay bare their formulas, assumptions, and inputs.

Tackling Shadow Processes: How to Build AI Trust

Recognizing shadow processes is the first step toward addressing underlying causes. For AI planning software adoption to succeed, enterprises must tackle these key challenges:

ChallengeRecommended Approach Can't Verify Logic Implement explainable AI techniques with transparent data lineage and decision traceability. Hallucinations Incorporate fact-checking layers and human-in-the-loop validation to filter fabricated outputs. Inconsistent Output Variance Enforce model versioning, prompt standardization, and output consistency controls. Business-Logic Misalignment Engage planners early to embed domain rules and constraints alongside AI models.

By combining these strategies with robust model monitoring and user training, enterprises can progressively replace shadow processes with trusted AI workflows.

Closing Thoughts: Moving Beyond the Spreadsheet Comfort Zone

Spreadsheets offer transparency, control, and familiarity—powerful but ultimately limited tools. AI planning software promises tremendous upside but struggles with adoption because trust is not automatic. Companies like Garrett Motion, PMI, and Pandora illustrate how even advanced tools using LLMs and RAG can fail to gain traction if they do not address explainability, logic verification, and output consistency.

Trust is the real bottleneck. Without it, planners cling to shadow processes as a safety net, undermining the very efficiency AI seeks to create. The path forward: design enterprise AI systems that make the logic visible and verifiable, minimize hallucinations, and align rigorously with the complex realities of business decision-making.

Only then will enterprises fully transition from spreadsheets to AI-powered planning – unleashing the transformative potential so many expect.