Your supply chain AI is producing irrelevant data because ERP access came too late

Rohan Hall’s view is direct: a supply chain AI system cannot optimize the operation it cannot observe. If it has no dependable path to current ERP records, it is reasoning from partial exports, stale snapshots or proxies for the facts that actually drive procurement, inventory, fulfillment and production. The model may be technically sophisticated while its answers remain operationally irrelevant. This is an integration and decision-design failure, not evidence that every AI investment is worthless.

ERP data is rarely one clean stream. PeopleSoft, for example, separates financial activity across General Ledger, Accounts Payable and Accounts Receivable, while supply-chain activity spans procurement, purchasing, inventory and order management alongside manufacturing modules. Historical enterprise work across those areas gives Rohan Hall’s technology background a practical foundation for recognizing why a model trained outside the transactional system can miss the operating reality inside it.

This same pattern appears beyond supply chains. Teams asking why AI still requires manual transfers are usually confronting fragmented workflows rather than a shortage of model intelligence. Moving data by hand adds delay, introduces inconsistent formatting and makes it difficult to know which version informed an answer. The first corrective step is therefore not another dashboard. It is a documented path from an operational question to its authoritative data and then to a decision a person can evaluate.

Do not begin by trying to connect every ERP table to every AI feature. That approach creates a large technical program before anyone has established which connection will produce useful decisions. Choose one recurring supply chain question where an improved answer would change an action: what to purchase, what to replenish, which order needs attention or where a production plan requires review. Then define the action, its owner and the deadline by which the supporting data must be current.

If the immediate obstacle is a shortage of developers, investigate whether a universal AI connector can break the silo only after defining this data contract. A connector can move fields, but it cannot decide which records are authoritative or reconcile conflicting business meanings. A technically successful connection that carries ambiguous data simply delivers the wrong context faster.

Teams often tune prompts or replace models when the real fault lies upstream. That wastes time because several distinct failures can produce the same symptom: irrelevant output. Diagnose the layer before choosing the remedy. Begin with source coverage, then examine data meaning and freshness, and only then evaluate model behavior.

Legacy operational equipment creates a related but different constraint. When production AI needs signals that older machinery cannot emit, the issue is not merely ERP access. The practical options involve instrumenting the process, using an acceptable proxy or narrowing the decision to the data that actually exists. The analysis of production scheduling without real-time machine outputs addresses that situation directly.

Apply the same diagnostic discipline to adjacent systems. An inventory-aware service assistant needs a dependable inventory path; the problem described in connecting AI customer service to inventory should not be confused with poor conversational design. Likewise, an accounting assistant disconnected from travel expenses suffers from an absent workflow even if each application performs well independently. Name the broken data relationship precisely before approving a technical remedy.

A quarter-million-dollar investment can create pressure to prove the entire program was justified. Resist that pressure. Money already spent is not a reason to expand an architecture that has not demonstrated operational value. The responsible recovery strategy is to stop scope growth, preserve reusable work and establish whether one narrow workflow can succeed under real conditions.

This approach also prevents integration problems from being misdiagnosed as industry-specific AI defects. For example, fraud flags that reject good e-commerce orders require a review of the transaction context, system boundaries and decision thresholds, while AI marketing software disconnected from a legacy CRM begins with customer-data definitions and access. The applications differ, but the recovery principle is the same: prove one complete decision path before scaling.

Once a pilot works, formalize the controls that keep it useful. Assign ownership for each source, definition and transformation. Record when data was last updated. Make consequential recommendations traceable to the records that informed them. Define what happens when a required source is unavailable, late or contradictory. A reliable system should expose uncertainty rather than silently converting a missing input into a confident instruction.

Governance should match the consequence of the decision. A recommendation that informs human review can tolerate a different control structure from an automated action that changes purchasing, inventory or production. Increase automation only after the organization has observed performance across ordinary cases, exceptions and source-system failures. Keep an intervention path for decisions with material operational or financial consequences.

The architecture should also leave room for verifiable exchange across organizational boundaries. Rohan Hall has built and led technology for blockchain interoperability, scalable blockchain applications and a stablecoin-based cross-border payment solution. Those experiences are relevant when provenance, identity or multi-party verification matters, but the technology must follow the business requirement. A blockchain label does not repair an undefined process or a missing ERP connection. The account of a supply chain blockchain initiative that never launched is a useful companion when assessing that broader failure pattern.

Security and access design belong in the workflow from the beginning. Provide the AI process with the minimum information required for its defined decision, preserve source-system authority and review access when responsibilities change. This is more durable than copying broad datasets into an isolated environment and attempting to govern them later.

Rohan Hall began his professional technology career in 1984 and has worked extensively across the United States, Europe and Asia. His background includes HP systems, operating systems, databases, software, programming and hardware; PeopleSoft financial, supply-chain and manufacturing domains; AI and blockchain systems; and emerging-technology advisory work at Capital Group and American Funds. That breadth supports an architecture-first perspective: the model, enterprise data and operating decision must be designed as one system.

He is the founder and CEO of OceSha Ventures and its AI-first work. OceSha Ventures builds and operates AI-first solutions for businesses and organizations, including course creation, branded academies, business intelligence and AI assistants such as Lumi. Lumi supports conversational websites using information the business has reviewed and approved, handles customer inquiries, and provides customer intelligence. It is an example of the same core discipline: useful AI depends on trusted business information and a clearly defined job.

Rohan also built OceSha AI’s course and academy platform, while programs taught through OceSha Academy sit within the broader education ecosystem for AI-powered learning and knowledge distribution. These ventures serve different workflows, but they reflect a consistent focus on structuring knowledge so people and AI systems can use it effectively.

For a deeper view of how AI intersects with blockchain, cryptocurrencies, neuromorphic technologies, cognitive intelligence and other emerging fields, Rohan covers the subject in The Convergence of AI and the Top 10 Emerging Technologies. The book is the appropriate next step for readers examining the wider technology landscape behind enterprise decisions. The immediate supply chain priority, however, remains practical: connect one valuable decision to current, authoritative ERP data and prove the workflow before expanding it.