← AI Operations collection
AI Operations Manufacturing Supply Chain Decision Infrastructure

Six Problems Got Worse for Manufacturers and Distributors in Fall 2026

From AI cost to input prices: the six problems that gained evidence fastest between 22 August and 28 October 2026, and one Monday action for each. Data through 28 October 2026.

Christopher Wakare
Updated
8 min read
AI & Operations

In August, operations leaders asked whether AI would pay. By October they asked who controls it and what it costs.

We read analyst reports, Fed surveys, trade press and practitioner podcasts every working day from 22 August to 28 October 2026. Six problems gained evidence fast enough that we call them rising. Two of them, AI cost and input cost, land on the CFO's desk.

Rising problem, defined

A rising problem, in our tracker, is one whose share of new evidence grew by a third or more over the last two weeks, compared with the two weeks before. Evidence means sources, each counted once per problem on the day it first raised the problem, weighted by type: analyst and institutional 3, named practitioner 2, trade and vendor content 1.

If you run operations or finance at a manufacturer or distributor on ERP/Business Central, this is what your peers wrestled with this fall, and what you can do about each one before year end.

Data through 28 October 2026. This is a snapshot, not a series.

From the work

Written from ten weeks of daily research, 22 August to 28 October 2026.

How we counted rising problems

Several pages from one publication count once: the Fed's October Beige Book has six district pages in our tracker, and we count it as one source. Analyst and institutional sources, such as Gartner, Forrester and the Federal Reserve, carry the most weight.

Ten weeks is short. Read these as signals for this quarter. The evidence is strongest for manufacturing and distribution, so that is who this is for.

The six rising problems for manufacturers and distributors

1. AI cost became a meter you can't read

Evidence: 3 sources in September, 7 in October.

What buyers say. AI used to be a licence line. Now it bills by use, and what a month costs depends on how much people use it. Gartner's Alicia Mullery, quoted by Digital Today: "costs for existing technologies are generally predictable, but AI costs are determined probabilistically when it is actually used."

Why it bites now. In Gartner's survey of 2,002 CIOs and technology executives (May to July 2026), 51% expected AI to increase total cost of ownership for technology (Digital Today). On 1 December, Microsoft turns on usage-based billing by default for new Microsoft 365 Copilot Business licences bought through partners (CSP), with a default limit of 4,000 Copilot Credits per user per month that admins can change (Microsoft Learn). The first rollout leaves out 11 markets, among them France, Germany, India and Spain (Neoteo). For the Business Central side of the bill, read why a default limit is not a budget and why rework, not tokens, is the agent cost.

What to do Monday.

  • Ask your Microsoft partner whether you will buy any new Copilot Business seats after 1 December. If yes, set a spending policy in the Microsoft 365 admin center before the seats go live.
  • Give every AI use case a named owner and one outcome it must move. Spend with no owner is the spend nobody can defend at budget time. The AI ROI Calculator nets the saving against credit spend and rework.

2. Input costs rise faster than you can reprice

Evidence: 1 source in September, 7 in October, with the Fed's Beige Book counted once. The largest jump of the six.

What buyers say. Tariffs, fuel and freight push costs up every month. Some manufacturers and wholesalers pass them on. Others have little room to.

Why it bites now. The Fed's October Beige Book (14 October) is the main source. In the New York district, input prices "continued to rise strongly," and tariffs were a widely cited cost pressure for manufacturers, wholesalers and retailers, some of whom had little room to pass costs on while others raised prices (New York Fed). Services contacts in the Minneapolis district reported that higher input costs were compressing profit margins (Minneapolis Fed). In the Cleveland district, higher fuel costs remained the dominant driver, and several contacts said they were spilling over into freight, packaging and petroleum-based materials (Cleveland Fed). Average diesel hit $6.30 a gallon in mid-October, according to AAA as reported by Fortune. On the decision side, see who decides when tariff policy moves and emergency freight as a tax on decision latency.

What to do Monday.

  • Pull every open quote valid past 30 days. Re-cost the ones built on summer prices before a customer accepts them.
  • In Business Central, compare each top-selling item's last direct cost with its sales price. Items where cost moved and price did not are your quiet margin leaks.

3. Leaders expect volatility and admit they aren't ready

Evidence: 1 source in September, 3 in October. The smallest of the six.

What buyers say. Disruption arrives weekly now: tariff changes, port delays, supplier surprises.

Why it bites now. In Gartner's CEO survey released 20 October (203 CEOs and senior executives, April and May 2026), 93% expected economic volatility to have a high or medium impact on strategic business objectives through 2027. High readiness for the disruptions it lists runs from 11% to 24% (Gartner). New York Fed manufacturing contacts reported that tariffs were raising input costs and disrupting supply chains, and a shipping contact said importers were paying a premium for air freight to land stock ahead of the holidays (New York Fed). The five-decision contingency protocol puts a named owner and a number on each disruption type.

What to do Monday.

  • List every item in Business Central with only one vendor. Rank them by revenue. That list is your real exposure.
  • Compare the lead time on each item card with actual receipt dates over the last 90 days. Planning on stale lead times turns a delay into a stockout.

4. Planning data stopped telling the truth

Evidence: 5 sources in September, 6 in October.

What buyers say. Lead times, min/max levels and supplier records were set years ago and never revisited. Planners work around them. An AI tool built on top inherits the same wrong numbers.

Why it bites now. Business Central's planning engine reads reorder point, safety stock and lead time as inputs, and the Microsoft pages we read describe no test of whether those values are still true. A planning run can finish clean, an alert can fire on time and an agent can recommend the right quantity for a lead time no supplier quotes anymore. We walked through the failure, and one reliability manager's account of it, in Your Min/Max Is Lying to You.

What to do Monday.

  • Pick your 20 highest-volume items. Check when someone last changed each reorder point and lead time in Business Central. Anything untouched for a year is a guess.
  • Before you buy any AI planning tool, fix those 20. A tool that learns from stale parameters repeats them faster.

5. Everyone sees the problem and nobody owns the fix

Evidence: 2 sources in August, 2 in September, 5 in October.

What buyers say. Teams paid for dashboards and alerts. The stockout warning fires, then sits in three inboxes while each person waits for someone else to decide.

Why it bites now. In Gartner's survey of 151 supply chain leaders (November to December 2025, companies with at least $250 million in revenue), 72% had to revisit final approvals for network decisions at least once, contributing to delays, and more than half said three or more times. Gartner's Vicky Forman blamed day-to-day instability, which Gartner calls turbulence, for steadily driving up costs, lowering service levels and forcing leaders to regret their decisions (MH&L). Those are larger companies than most readers here, but the pattern is the same one: a decision approved, then reopened.

What to do Monday.

  • List your five most common alerts: stockout risk, late purchase order, production delay. Write one name next to each. That person decides. A shared channel never will.
  • Time one alert end to end this week, from when it fired to when someone changed something in Business Central. That gap is your decision latency, and the Decision Latency Diagnostic scores where it sits.

Where OpsGrid fits. This is the problem we built OpsGrid for. It watches Business Central for stockouts, late purchase orders and supplier delays, and puts a ranked decision card in front of the named owner in Microsoft Teams. Nothing posts to Business Central until that person approves. The category is decision infrastructure: a named owner, an approval threshold and an audit trail on every decision.

6. Jobs are changing and nobody rewrote them

Evidence: first raised 30 September: 2 sources that week on skills and labour, 5 in October.

What buyers say. Plants cannot hire enough skilled people. Meanwhile AI takes over parts of planner and buyer work, and nobody has decided what those people now own.

Why it bites now. In Gartner's 8 October procurement research, 63% of senior procurement leaders expected AI to significantly improve procurement performance over three years, and 25% had redesigned jobs and roles as a direct result of AI implementation (surveys concluded April 2026, n=213 and n=1,303 including 134 procurement leaders) (Gartner). In its October CIO survey, only 4% of CIOs had plans to restructure roles so people and AI agents work together (Digital Today).

What to do Monday.

  • Pick one role your AI plans touch, say a buyer. Split their week into tasks the system now handles and judgments that still need them. Rewrite the role around the judgments.
  • For every AI output that reaches Business Central, name who checks it and what they check for. An approver with no criteria approves everything. The stop-criteria article shows who signs that call.

The biggest problem, AI accountability, held steady and changed shape

AI accountability had more sources than any other problem we track: more than 20 since August. It did not rise. It moved earlier in the workflow.

In September, buyers worried they could not prove what an AI agent had done or who signed off. By late October, the conversation had shifted to deciding what an agent may touch before it runs. Gartner's April guidance on agent sprawl lists six steps, and defining each agent's identity, permissions and life cycle is the third, after setting governance policies (Tech.co).

A new worry arrived with it: approval fatigue. Ask people to approve everything and they stop reading. Anthropic's own telemetry, vendor data on a developer tool, showed users approved roughly 93% of Claude Code permission prompts (Anthropic). An approval step only protects you if the person still reads what they approve. We cover the detection method, and the Approval Budget Calculator shows whether your queue is larger than anyone reads.

What to do Monday. Write down which Business Central records an AI tool may change, and which it may only suggest changes to. Keep the approval list short enough that the approver reads every item.

This is the second place OpsGrid fits. It can only write to the action types you enable, and every write needs a human approval in Teams (OpsGrid for CIOs). Ranking the cards keeps the approval list short. If the question is what to do before an agent goes live, the AI Agent Post-Mortem Template names the owner, the stop rule and the undo path.

What went quiet in our tracker, and what went quiet too early

Quiet: Business Central platform deadlines. In September this was the loudest topic we tracked, with about a dozen sources on SOAP retirement, the late Wave 2 release plan and the BC29 preview windows. We have logged no new source since 5 October, after BC29 shipped. Quiet in our tracker does not mean finished. Check your own upgrade plan for dated items before you take it off the risk register.

Falsely quiet: AI regulation. It dropped out of our tracker in October because our daily research stopped checking regulatory sources. We are adding them back. The rules did not pause. Under the EU's Digital Omnibus, generative AI systems already on the EU market before 2 August 2026 have until 2 December 2026 to mark their output as AI-generated in a machine-readable way (European Commission FAQ). The marking duty falls on providers of generative AI systems (Baker Botts). If you sell into the EU, ask your AI vendors in writing how they will meet it.

Dates to watch before year end 2026

Date What happens Problem it moves
25 November Fed releases its next Beige Book at 2:00 p.m. (Federal Reserve calendar) Input costs, volatility
1 December Usage-based billing on by default for new Copilot Business licences bought through CSP AI cost
2 December EU AI Act grace period for machine-readable marking ends AI regulation

Frequently asked questions

What are the biggest problems for manufacturers and distributors in late 2026?

From 22 August to 28 October 2026, six problems gained evidence fastest in our tracker: AI cost, input costs, volatility readiness, stale planning data, alerts nobody owns, and jobs nobody rewrote. AI accountability had the most sources overall and held steady.

How did you count the evidence?

Each source counts once per problem, on the day it first raised it, and several pages from one publication count once. Analyst and institutional sources weigh 3, named practitioners 2 and trade or vendor content 1. A problem is rising when its share of new evidence grew by a third or more over two weeks against the two before. Ten weeks is short, so read the result as a signal for this quarter.

Do these problems apply if you are not on Business Central?

The evidence comes from sources that are not specific to Business Central: Gartner, the Federal Reserve and trade press. The Monday actions name Business Central screens, because that is the system we work in. The same checks apply to any ERP with item cards, lead times and approval workflows.

Which problem should finance look at first?

AI cost and input costs. Both reach the budget directly. For AI cost, set a spending policy before new Copilot seats go live and give each use case an owner. For input costs, re-cost open quotes built on summer prices. The AI ROI Calculator and the AI ROI board report spec cover what finance will ask of an AI saving.

Sources

Evidence counts come from IntelliConnectQ's daily research, 22 August to 28 October 2026. Pages cited above:

If problem 5 sounded like your Monday, book a Decision Audit

A free 30-minute Decision Audit. We map the three to five decisions costing you the most before we quote anything.

Book a free Decision Audit →