AI workflow redesign in manufacturing starts with the work: choose a bounded process, map its decisions and handoffs, and name the person accountable for the result. Spej recommends confirming data access and approval stages before building, then measuring the completed workflow. That gives operations and IT a shared basis for deciding whether to expand.

By Sean Blair, Co-Founder & Chief AI Officer, Spej | November 11, 2026

For a COO or plant leader, a faster task is useful when the surrounding process can use the saved time. If the output still waits for missing information or an unresolved approval, the benefit needs to be assessed across those steps. Spej's view is that the workflow deserves attention before another tool purchase.

What BCG means by workflow redesign

BCG's November 10 release distinguishes deploying AI in existing workflows from reshaping core functions and processes. It estimates that deploying AI can save individual workers 10% to 15% of their time on tasks, whereas reshaping core functions and processes can deliver a 30% to 50% improvement in efficiency and effectiveness. Those are different measures and scopes; the latter is not a task-time savings estimate. BCG

The same release says almost 70% of US industrial companies that adopt AI make no organizational, data, or process changes, and only 12% build new workflows. BCG attributes those figures to US Census Bureau data cited in its report. They describe reported adoption behavior, without establishing why a particular manufacturer's project succeeds or stalls. BCG

BCG's margin modeling covers large, established chemical, aerospace and defense, and automotive companies. The modeling is illustrative, not a forecast for individual companies, and that population is not Spej's mid-market buyer. A smaller manufacturer should build its own business case around an actual workflow, including the cost of changing how people work. BCG

The practical implication, in Spej's view, is to investigate where the process needs a different handoff or decision rule. A drafting assistant may be the appropriate scope. A workflow with repeated reconciliation may need changes to data ownership and review before an assistant can help.

What the supporting surveys can tell you

Rootstock's survey found that 73% prefer human-directed AI or AI that operates autonomously within defined limits and escalates when needed, versus 13% who prefer fully autonomous AI. The survey was commissioned by Rootstock and conducted by Researchscape among 588 manufacturing professionals at organizations with 100 or more employees in the United States, United Kingdom and Canada. Rootstock

In that survey, 86% were at least somewhat confident their ERP could keep pace with AI-driven manufacturing changes; only 43% were very or extremely confident. Confidence is a respondent assessment. It doesn't verify that a particular ERP can supply the records or permissions your proposed workflow needs. Rootstock

TeamViewer's separate research found uncertainty about AI's value or business impact was an adoption barrier for 32% of manufacturing IT decision-makers, compared with 16% in technology. Sapio Research conducted the study for TeamViewer in April and May 2026 among 4,200 respondents across nine markets: the US, UK, Australia, France, Germany, Italy, Japan, Singapore and Brazil. That is the full study sample, not a count of manufacturing respondents. TeamViewer

These surveys use different populations and questions. They don't combine into a forecast for your plant. They do give leaders useful questions to ask about authority, system readiness and proof of business value before funding implementation.

Pick one workflow and map the work

Spej recommends choosing a workflow with a clear trigger and final action. A description such as improving planning leaves too much unresolved. Specify what arrives, who handles it, which decision follows and what counts as completion. Include the work that happens after an AI output appears.

Consider a hypothetical schedule-change review. A proposed change arrives, operations checks constraints, someone resolves missing information and an authorized person decides whether to accept it. This is a design exercise, with no assumed result. Mapping it should reveal which inputs the reviewer needs and where the work waits today.

Ask the people doing the work to identify repeated entry and exceptions. Record which decisions follow an explicit rule and which require judgment that hasn't been documented. A process map should include rejected suggestions and requests for clarification so the build can handle ordinary uncertainty.

A pricing-update or document-review workflow using business-system data may also be worth assessing. Its suitability depends on access, data quality and the operational problem. Business process engineering helps define what should change before the technical scope is fixed.

Decide the data, owner and approval stage

BCG reports that only 12% of industrial companies have the data capabilities needed to be future-built for AI, compared with 18% in other sectors. This is a sector-level finding, not a readiness score for a particular manufacturer. BCG

For the selected workflow, Spej recommends naming the authoritative source for each required input. Check freshness, access permissions and how conflicting records are resolved. Include information that sits in someone's working notes if the decision depends on it. Decide whether it can be documented and maintained reliably enough for the proposed use.

Name a business owner who can approve changes to the workflow. Separately, assign responsibility for integrations and monitoring. The operations lead needs a route to stop use when the output is unreliable; the technical team needs someone who can resolve business-rule questions. Decision-rights work should make those responsibilities explicit.

BCG describes trust built in stages: agents advise first, then act with approval, and only later act on their own within defined limits. Spej recommends applying those stages to specific actions, with a person responsible for reviewing exceptions. BCG

If the workflow depends on shop-floor data, involve IT and OT in assessing access, uptime and change approval. Spej does not offer OT network engineering, wireless design or PLC/SCADA work. Relevant infrastructure and controls work belongs with the manufacturer's qualified teams and vendors; it needs its own scope and approval.

Measure the whole workflow before expanding

Establish a baseline before implementation. For the hypothetical schedule review, elapsed time should include waiting and clarification. Count human review effort and corrections, then compare the completed workflow after the change. A faster generated recommendation can still require more checking.

Spej recommends tracking operating costs and maintenance effort alongside the business measure. Check whether users follow the agreed workflow or keep a parallel process because they can't rely on it. Review exceptions with the people responsible for the outcome, and identify whether the problem comes from inputs, rules or the proposed implementation.

Agree conditions to continue, change or stop. Expansion should depend on evidence from the workflow and its operating conditions. An accountable owner needs authority to make that decision and funding for whatever maintenance the expanded scope requires.

Where AI falls short

AI cannot settle ownership disputes or supply a missing data standard. A system may produce an answer from incomplete information, and reliable source data can still be interpreted incorrectly. Human approval needs enough time and evidence to be useful. Spej recommends pausing consequential automation when inputs or ownership remain unresolved.

A shop-floor deployment can also introduce traffic and dependencies that require separate technical assessment. Research about larger industrial companies doesn't establish what a particular plant can support. Keep the business case specific to the chosen workflow and avoid treating survey confidence or modeled gains as an implementation guarantee.

How the AI Roadmap supports the first build

Spej's AI Roadmap, starting at $10K, provides a way to decide what should be built. Begin with the selected workflow, map how it works and identify the required data. Agree the owner, approval stage and measures, then use those decisions to define a build scope. Final pricing depends on scope.

AI implementation follows those decisions, with access and integration feasibility assessed for the systems involved. Continuing ownership and review can be planned through AI Office support where appropriate. The manufacturer retains responsibility for its business decisions and for involving the relevant plant and technical teams.

FAQ

What is AI workflow redesign in manufacturing?

AI workflow redesign means deciding how work should move through a process when AI participates. Spej recommends mapping the trigger, inputs, handoffs, decisions and exceptions, then assigning ownership and approval rights. The scope includes what people do with the output and how the business measures the completed workflow.

How do AI copilots differ from process redesign?

A copilot can help someone complete an existing task. Process redesign examines the surrounding work, including repeated entry, waiting, approval and exception handling. Both can be useful. Spej recommends choosing the scope from the operational problem and measuring the whole workflow, including any additional checking or maintenance it requires.

Should manufacturing AI agents act autonomously?

Spej recommends defining authority for each action and earning broader permissions through evaluation. Begin with advice, introduce action with human approval where appropriate, and consider autonomy only within explicit limits. Name the person who handles exceptions and decide how the workflow pauses when data or operating conditions change.

Where should a mid-market manufacturer start?

Pick a bounded workflow with a business owner who can change how the work happens. Map the current process, confirm the required data and permissions, and agree the human approval stage. Establish measures for elapsed time, errors, review effort and operating cost before deciding what to build.

Discuss the workflow you want to change

Bring the workflow, its current handoffs and the questions your operations and IT teams need to resolve to Spej's contact page. That gives the discussion a concrete starting point for deciding whether a Roadmap and a subsequent build fit the work.