The sales team is faster. Why is operations paying for it?
Imagine a distributor introducing an assistant to prepare quotes. Sales can respond sooner and handle more requests. The team reports a successful rollout.
Field notes
A closer look before you commit. Start with the question you need answered.
Four starting points
- 01
Understand the operational problem
SOP vs reality: why your documented processes lost to the informal onesAll notes on this topic → - 02
Choose and adopt the first workflow
The first workflow is worth more than twenty licensesAll notes on this topic → - 03
Keep knowledge and systems under control
Shadow AI: bring useful work under company controlAll notes on this topic → - 04
Buy with clear economics and responsibility
You are trying to buy AI the same way you bought softwareAll notes on this topic →
Complete archive
Understand the operational problem
Closed by the chatbot. Still open for the customer.
The conversation ended without a request for a human. Your dashboard counts it as contained.
Before we build the agent, who can cancel the report?
An engineer can automate an export. Removing the requirement takes a different decision. Make it before paying to preserve the old routine.
The CEO as middleware: how approval chains become the company bottleneck
Routine approvals can consume the attention needed for executive decisions. Follow a stalled decision and check its authority, information and consequence.
A manufacturing CEO describes the work of holding a growing company together
Jim Belosic described the coordination burden at 600 employees. The industrial lesson is to make operating knowledge transferable while developing people who can decide.
AI can amplify a workflow problem. Check the technical cause too.
A weak process, poor source data or a technical defect can each explain a stalled AI result. Test the layers separately before prescribing organizational change.
SOP vs reality: why your documented processes lost to the informal ones
Compare the documented process with the work people actually perform. Keep necessary controls and test whether the workarounds are useful, avoidable or unsafe.
The hidden payroll: what 'finding out' costs a mid-market company every week
Information search consumes capacity. Use a bounded calculation to separate necessary work, potentially released time and a benefit the company can actually use.
Key-person risk: the list nobody wrote down
Identify the knowledge whose absence would interrupt critical work, then test whether a second operator can actually take over.
Choose and adopt the first workflow
The machine raised an alert. Can anyone actually do the job?
Imagine a production line whose vibration monitoring flags a deteriorating bearing. The maintenance team can see the trend. Production knows there is a problem. The proposed intervention is sitting in the maintenance system.
Before adding another job to the shutdown, ask what it displaces
A planned shutdown creates an attractive opportunity: the equipment will already be stopped, so why not complete the extra work at the same time?
What happens to the person who reports an AI mistake?
Imagine a customer-service employee reviewing a proposed reply. The assistant has confused a contractual deadline with an internal target. The employee corrects the response before sending it, then considers whether to report the mistake.
If AI prepares the first draft, where will the next expert learn?
An experienced account manager can spot a weak proposal quickly. They notice the missing constraint, the condition that should not have been promised and the question the customer has not answered.
The storm has passed. Your claims backlog is still growing.
Imagine an insurer receiving a sudden wave of property claims after severe weather. Customers submit photographs through the app, call for confirmation and send additional documents by email. A broker forwards the same event under a different reference.
The agent escalated the case. Can anyone decide from what it sent?
Imagine a distributor receiving a request to split a delivery. The customer wants the available items tomorrow and the remainder next week.
The invoice is ready. Who can clear the review queue?
That can be a sensible boundary for an AI workflow. It still leaves an operating question unanswered: which person, with how much time, and what happens while the case waits?
The team has time now. The customer is still waiting.
The paperwork takes less time. Customer problems still sit untouched. What your team is allowed to do next can decide whether the investment matters.
Your AI adoption number is lying to you
Imagine two companies reporting the same adoption rate but handling work differently. Track what people use alongside the work and outcomes that changed.
Your company runs on rules nobody ever wrote down
Tacit rules need to become explicit before software can support the work. An AI model can interpret a request; enforced permissions and validation govern its actions.
Two AI rollout patterns that stall, and how to keep the work running
Mandates and broad license purchases can stall for different reasons. Look for a useful workflow, operator involvement and a named owner through adoption.
The first workflow is worth more than twenty licenses
A useful first workflow can create evidence and reusable capability. Select a clear problem, test it with operators and package what the next workflow can inherit.
Find your internal AI champions and make their methods reusable
Motivated operators can help colleagues learn. Their useful methods need company-owned context, controls and time to transfer.
Why AI use and operational results need separate evidence
A deployed tool can leave the workflow unchanged. Check the specification, adoption and outcome with different evidence.
You adopted AI. That is not the same as implementing it.
A technically delivered system still needs people to use it in real work. Treat implementation and adoption as complementary parts of the same accepted outcome.
Four operating modes for AI, with a control boundary at each step
Four operating modes help describe how work is delegated. Each needs defined authority, evaluation and human judgment where the consequence requires it.
When models improve, a tested workflow gives you a useful starting point
Model announcements do not prove production readiness. Build a reusable test set around one workflow and evaluate changes against it.
Keep knowledge and systems under control
The new rate is approved. Which quotes still use the old one?
A pricing committee approves a change for a defined segment. The internal rating service is updated. The project status turns green.
Each underwriting exception makes sense. What do they add up to?
An underwriter accepts a concession for a longstanding customer. Another team approves a similar adjustment for a broker relationship. A delegated partner writes business within its individual authority.
One payment service, several insurance businesses. What should stay different?
An insurance group wants to reuse an AI-enabled payment workflow across business lines. The technical components look familiar: identify the recipient, prepare an instruction, obtain approval, send it and reconcile the response.
The order changed after you approved it. What will the agent send?
Imagine a packaging manufacturer preparing a shipment. The cartons have been printed and packed; dispatch is waiting for release.
Which manual applies to this machine?
A service assistant finds a clear answer in a genuine company manual. The paragraph is quoted accurately. The link opens.
You replayed yesterday's failed request. Did you test yesterday's situation?
An AI assistant mishandles a refund request. The team changes its instructions and runs the customer's message again. This time the answer is correct.
Shadow AI: bring useful work under company control
Start with the tools people use, the data they share and the permissions a useful company environment needs.
Every vendor can rent the same model. Your moat is somewhere else.
Reusable business context, tested interfaces and control of the operating assets give your team a starting point that survives a change of model or supplier.
Your best people are funding AI out of their own pockets. Your AI budget is still under review.
Personal purchases can signal useful demand. A company still needs evidence of workflow value, permissions and operating cost before investing.
Glean and self-hosted enterprise search: compare control and running cost
Compare permissions, connector upkeep and total cost before choosing enterprise search. Self-hosting changes the operating responsibility.
Buy with clear economics and responsibility
This spare part has not moved in years. Should you really remove it?
The inventory report has found an expensive component with no recorded issue for several years. An assistant puts it on the disposal list. The warehouse would recover space and finance would reduce obsolete stock.
Your AI quoted the change in seconds. Who pays for the extra delivery?
A faster quotation process can commit the business to work it has not priced.
The agent is cheap. Who finishes the work?
A low monthly bill can hide a very busy employee. Before you approve an AI project, count what it takes to get one case properly finished.
Before the next hire, examine the work behind the capacity request
Industrial companies do not need a software-company identity. They need a sound choice between staffing, systems and work redesign.
Where a small-team operating model can help an established company
Keep your industrial assets and expertise. Examine one coordination-heavy function for unnecessary overhead and reusable capacity.
You are trying to buy AI the same way you bought software
A useful AI purchase needs an operating model as well as a price. Define usage, acceptance, cost controls and who will maintain the workflow.
AI agents vs hiring: what a 200-person company should automate first
Compare the next capacity need at workflow level. Some work needs another person; some needs better information flow, with full cost and decision risk made explicit.
What does an AI readiness assessment actually cost in 2026?
Compare the diagnostic scope, evidence and next decision. A programme price and an implementation quote cover different work.
AI consulting for mid-market companies: what it costs and what you actually get
Compare the work included in a quote: diagnostic evidence, implementation, adoption, ongoing operation and the conditions for another team to take over.
How to choose an AI consultant: start with one real problem
Use a real workflow to judge the approach, agree a limited first engagement and see what the work establishes before committing further.