Decision Infrastructure

CSP's CIO named the gap automotive manufacturers already feel: an ERP that logs every transaction perfectly and still leaves the next decision waiting for someone to notice, route, and act on it.

Spot rates are up 25% YoY with no relief in sight. Uber Freight's Bob Daymon and Coty's Graeme Carter both say the operators winning 2026 stopped waiting on forecast confidence. What a decision-speed operating model requires instead.

Payables Agent, Copilot, Agent Designer, Power Automate, and OpsGrid all touch Business Central now. A field guide to what each one actually governs, its real maturity status as of August 26, 2026, and the accountability gap between them.

A 2026 FDA warning letter cited a pharmaceutical and cosmetics manufacturer for cGMP violations after AI agents generated specifications, procedures, and compliance records with no human review. The output wasn't shown to be wrong. The missing approval was the violation.

Business Central's excise tax engine reached general availability May 8, 2026. Microsoft's own release notes say it directly: no compliance or statutory reports ship with it. What that means for tax managers at alcohol, tobacco, and fuel manufacturers.

Business Central's Sales Order Agent converts a customer's email reply into a posted order with nobody touching a screen. Microsoft built a real write-scope control for that. It isn't the same thing as an audit trail.

Lexful raised an oversubscribed $7M seed round in June 2026 to solve documentation capture for MSPs. MachineMetrics and Tulip are building similar reasoning layers for manufacturing. What OpsGrid's Audit stage already logs on every exception approval.

Gartner puts 56% of non-human identities outside structured governance. Most of that growth is AI agents nobody assigned an owner. Why agent risk is an identity problem, not a model problem, and how it extends the lifecycle automation MSPs already run for humans.

Three benchmarks, Carnegie Mellon's TheAgentCompany, WebArena, and MIT Project NANDA, put agent failure rates between 70 and 95%. What actually breaks a deployment, and why verification has to be a loop, not a launch-day test.

Sage's 2026 State of Supply Chain Report found 95% of supply chain leaders call AI vital, and 10% have it live, an 85-point gap. The readiness signals, ERP access, named ownership, sponsor authority, pilot scope, that separate the two before a vendor is ever picked.

A 2026 forecast puts a number on autonomous disruption resolution: 60% by 2031. The exception-category rollout, inventory risk first, then supplier exceptions, then production scheduling, that has to start now to get there.

A purchase order sits in approval limbo for six days because the Power Automate flow that routed it was built by someone who left in March. What BC's native approvals actually cover, what the workaround costs, and what a governed alternative looks like.

A May 2026 legal series flags novel liability, regulatory, and contractual risk from ungoverned AI in manufacturing. What a General Counsel needs from the audit trail that a CFO never asks for: a named approver, a timestamp, and an exportable record.

EU AI Act Article 14 names three human oversight capabilities, and an approval gate delivers one. The Bank of England has gone further, calling human sign-off on every agent action unrealistic and floating kill switches instead. What Business Central ships natively, and what has to be built.

Half of AI agent deployment failures by 2030 are forecast to trace to governance runtime enforcement, not model capability. What that reorders on a multi-year Business Central AI roadmap: write inventory, autonomy and scope tiering, then the model last.

A July 2026 ecosystem analysis and a verified Siemens-NVIDIA-Humanoid deployment at Erlangen both point the same direction: AI vendor selection is being decided on ERP embeddedness, not model benchmark scores. The 5-question scorecard that predicts a real deployment instead of a stalled pilot.

A July 2026 roundup names a shift already underway at Hitachi: engineers embedded on-site for physical AI deployments. Why remote delivery fails on a shop floor, and what a 2-week embedded deployment across Business Central, Epicor, and Infor actually requires.

Cross-training transfers skills. It doesn't transfer the record of every decision that shaped how your operation runs today. The institutional knowledge gap in manufacturing is a record problem, addressable without any documentation effort from your team.

IoT Analytics surveyed 120 process manufacturing leaders expecting 12% lower operating costs from AI. The ambition is real. This is what separates the plants that get there: it's a decision infrastructure problem, not a modeling problem.

Decision intelligence surfaces signals. Decision infrastructure governs what happens next: who owns the decision, who approves it, and what gets audited. Your supply chain probably has the first layer. Here is why it needs the second.

Every operations team has more AI alerts than they can act on. The problem is not the AI: it is the absence of an execution layer that turns signals into governed decisions with named owners, approval workflows, and audit trails.

Business Central has the data. The gap is the decision layer between the signal and an approved response. Here is what OpsGrid adds on top of what you already have.

Manufacturing pilots succeed technically and fail operationally. The 5 root causes behind the gap, and what the organizations that close it actually do differently.

83% of organisations are running supply chain AI pilots. Most won't scale. The gap is an operating model problem, not a technology problem, and three structural gaps determine which side of that line you're on.

Your on-time-in-full rate is not a supply chain metric. It is a decision timing metric. The alerts that break OTIF most often fire outside business hours, and the gap between signal and approved response is where the ship window closes.

When component supply falls short of combined plant demand, who decides which site gets priority? Without a pre-defined governance model, the answer is whoever escalates to the COO first. That is not a decision. It is a political outcome.

No tariff assumption is stable right now. The answer is not better forecasting: it is a decision governance layer that makes procurement decisions documented, auditable, and executable regardless of what policy does next.

When a shipping route closes, most operations teams discover their gaps in the wrong order. One framework for sequencing the decisions that matter, with named owners, numeric thresholds, and escalation paths. Draws on MIT resilience research and ASCM practice.

When a P1 incident ends and the compliance audit begins, can you reconstruct every decision made during the response? Who approved overtime? Who authorised vendor escalation? If the answer is "check the Slack channel," the MSP has a governance problem.

The financial case for vertical integration is compelling. The execution gap (SOPs, decision ownership, knowledge transfer, cross-plant coordination) is real and under-discussed. This is what operational readiness actually looks like.

The four-variable framework and interactive calculator we use to build a business case before any deployment starts. Most teams find the number higher than they expected.

Expedited shipping rarely appears as a line item anyone owns. It accumulates in the hours between a system flagging a shortage and a human authorising the response. What sits inside that budget, and how to measure the delay before committing to a fix.

Most mid-market manufacturers run 4–7 disconnected systems. Decisions stall because nobody has the full picture. A practitioner guide to unifying ERP, CRM, and operations data, built around decision choke-points, a unified data pipeline, embedded BI, and selective automation.

4 people. 20+ hours daily. Duplicate POs. No trail when fabric was damaged on the floor. Fitting orders over-committing stock. Here's the integrated supplier portal, and what the cutover actually took.