Field Note by Dr. Robert Urban

Why Most Companies Integrate AI Backwards

The fastest way to waste money on artificial intelligence is to start with artificial intelligence. Start with the work, the friction and the decision. Then decide whether AI belongs there at all.

Artificial intelligence has become the business equivalent of a treadmill in January. Everybody wants one, most people are convinced it will change their life, and by March there is a decent chance it is sitting in the corner holding laundry.

AI does not fix confusion. It multiplies whatever already exists inside the business.
The First Mistake

The first AI question most companies get wrong

I keep hearing versions of the same question: “How do I integrate AI into my business?” It is a reasonable question, but it is usually one question too early.

Leaders see competitors talking about AI, software vendors promising efficiency, employees experimenting with new tools and headlines suggesting that anyone who waits six months has already lost. That creates urgency before it creates clarity. The conversation starts with platforms, licenses and features before anyone has defined the business problem.

What problem are you trying to solve?

That question changes the room. Sometimes the problem is obvious: proposal turnaround is slow, customer questions are repetitive, people cannot find approved information, CRM notes are inconsistent, reports take too long, or experienced employees keep answering the same internal questions. Those are real problems. They can be measured, improved and assigned a value.

Sometimes the answer is simply, “I do not want to fall behind.” I understand that fear, but fear is a lousy systems architect. A business should not automate a process merely because automation is available. It should improve a process because the process is costing time, money, quality, capacity or customer trust.

A Better Starting Point

Five questions I would answer before buying another AI tool

These questions are deliberately boring. That is a feature. They force the conversation back toward operating reality.

1What work is actually broken?Name the delay, rework, repetition, information gap or decision bottleneck.
2How often does it happen?A task repeated every day deserves different attention than an irritation that appears twice a year.
3What information does the work depend on?AI is only as useful as the knowledge, data, permissions and context available to it.
4Where is judgment required?Separate assistance and preparation from decisions that need accountable human review.
5What happens when it is wrong?The consequence of error should determine testing, approval, escalation and whether AI belongs there at all.
Usefulness Before Novelty

Where AI tends to help, and where enthusiasm gets ahead of judgment

Artificial intelligence is often excellent at compressing repetitive knowledge work. It is much less impressive when the organization asks it to replace clarity, accountability or expertise.

Business workAI may be useful forHuman responsibility still matters for
Customer and client intakeSummarizing responses, classifying needs, preparing records, routing routine requestsExceptions, sensitive situations, commitments, relationship judgment
Internal knowledgeFinding approved information, summarizing long documents, drafting answers from known sourcesSource quality, access rules, policy interpretation, final accountability
Sales supportMeeting summaries, follow-up drafts, CRM preparation, research supportQualification, negotiation, promises, pricing judgment, relationship strategy
Marketing workResearch assistance, repurposing, versioning, analysis, workflow supportPositioning, voice, claims, taste, original insight, final editorial judgment
Reporting and analysisOrganizing inputs, identifying patterns, summarizing recurring informationChoosing the right question, validating assumptions, deciding what the result means
High-consequence decisionsPreparing information and surfacing relevant context when appropriateApproval, professional judgment, legal or regulatory responsibility, consequences for people
A useful rule: the more expensive, irreversible, regulated or personal the consequence of an error becomes, the more explicit human review should become.
Process Before Automation

Do not automate a mess and call it transformation

Automating a broken process simply allows the broken process to happen faster. That is efficient in the same way driving seventy miles an hour in the wrong direction is efficient.

1. ObserveFollow the real workflow, including inboxes, spreadsheets, handoffs, workarounds and approvals.
2. SimplifyRemove steps that exist only because “that is how it has always been done.”
3. Assist or automateUse AI where it can reduce repetitive effort without hiding responsibility.
4. MeasureJudge the change by time, quality, capacity, risk and business outcomes, not by how much AI was used.

This is why I am skeptical of AI projects that begin with a product demo. A demo proves that software can do something. It does not prove that your organization needs that thing, that your data is ready for it, that employees will use it correctly, or that the result is worth the new complexity.

The Part That Matters Most

AI should create more room for human value, not less

The best integration work is not a contest to remove people. It is a discipline for deciding what computers should carry so people can spend more time on judgment, relationships, creativity and decisions that have consequences.

Let machines carry more of the repeatable load

  • Searching and organizing approved information
  • Summarizing routine material
  • Preparing first drafts and structured records
  • Classifying and routing predictable work
  • Reducing duplicate entry and recurring administrative effort

Keep people responsible for what requires judgment

  • Choosing priorities and making tradeoffs
  • Understanding context that is not in the data
  • Handling exceptions and sensitive conversations
  • Protecting trust, voice and professional standards
  • Owning the consequences of important decisions

That boundary is not identical for every organization. A family business, manufacturer, professional firm, nonprofit and national enterprise can face very different risk, data and approval requirements. Good AI strategy respects those differences instead of forcing every business into the same software-shaped answer.

Frequently Asked Questions

Practical AI integration questions

What should a business do before adopting AI?
Start by defining the business problem, mapping the current workflow, identifying the information the work depends on, deciding where human judgment is required and understanding the consequence of an error. Tool selection should come after those questions, not before them.
What business tasks are good candidates for AI?
Good candidates are often repetitive, information-heavy tasks with clear inputs and review standards, such as summarization, information retrieval, meeting notes, CRM preparation, routine classification, research support and recurring administrative work. The right candidate still depends on data quality, frequency, value and risk.
Should a company automate a process that is already inefficient?
Usually not without fixing the process first. Automation can make an inefficient workflow move faster while preserving the same unnecessary steps, bad handoffs and unclear responsibilities. Observe and simplify the process before deciding which parts deserve automation.
Where should humans stay involved in an AI workflow?
Human involvement should increase as decisions become more consequential, ambiguous, regulated, sensitive or relationship-dependent. People should remain accountable for important approvals, exceptions, professional judgment, commitments, claims and decisions that materially affect customers, employees or the organization.
How should a business measure whether AI integration is working?
Measure the business outcome rather than the amount of AI activity. Useful measures can include time saved, reduced rework, faster response, improved consistency, increased capacity, fewer handoff errors, better customer experience and lower operational risk.
Is using ChatGPT the same as integrating AI into a business?
No. An individual using ChatGPT can be useful, but business integration usually means making AI part of a repeatable workflow with approved information, clear permissions, defined review, system connections where appropriate and accountability for what happens next.
Does every business need an AI strategy?
Every business should understand how AI may affect its work, customers and competitive environment, but that does not mean every organization needs a large AI program. Sometimes the right decision is a narrow workflow improvement, a controlled experiment or simply better process discipline before adding new technology.
When the problem is clear

Then the right system becomes a much easier conversation.

If you are trying to decide where AI can create useful capacity inside your organization, I can help map the work, separate real opportunities from expensive distractions and build a practical path forward.

For a broader risk-management reference, see the NIST AI Risk Management Framework.
Scroll to Top