Overview

Bad AI output looks professional. It's well-formatted, grammatically correct, and sounds confident. That's what makes it dangerous. You could hand it to your team, they could implement it, and only months later realize it was built on assumptions that don't apply to your operation. Or worse, compliance-critical steps are missing, and nobody notices until an audit.

Your job as an operations leader is to recognize the failure modes. AI is a tool. It can do useful things. But it will confidently generate content that sounds right but is fundamentally broken. You need to know what "broken" looks like.

The Five Red Flags of Bad Operations Output

Red Flag #1: Generic and Obvious

The output could apply to any company in any industry. You read it and think: "I could have written this myself, in fact I kind of did write this, five years ago."

Example (bad SOP for customer complaint handling):

Step 1: Listen to the customer complaint
Step 2: Document the complaint
Step 3: Apologize for the inconvenience
Step 4: Investigate the issue
Step 5: Offer a solution
Step 6: Follow up with the customer

This is objectively true but useless. It's like telling someone "to be successful, do good work." Not wrong. Not helpful. Missing all the specificity that would make it actually runnable.

What's really wrong: The output doesn't account for your specific failure mode. Maybe your real problem is that complaints about the same issue from multiple customers aren't being aggregated. Maybe you have different complaint types (billing vs. service vs. product quality) that need different response paths. Maybe your complaint volume is so high that "investigate" needs to be specified as a 30-minute investigation, not a 5-day deep dive.

Generic output wastes time because you have to convert it to something specific anyway.

How to catch it: Read the output and ask: "Could I hand this to someone outside our company and they'd execute it the same way?" If yes, it's too generic. Good operations output is idiosyncratic to your situation.

Red Flag #2: Fabricated Metrics and Numbers

The AI invents specificity that doesn't exist. It adds numbers, timelines, and percentages that have no basis in reality. They just sound authoritative.

Example (bad capacity plan):

"Based on industry standards, this process should be completable by one person in 8 hours per week. At your volume of 150 transactions per month, you'll need 0.5 FTE allocated to this process. Consider adding a second person at 50% utilization during peak months."

Where did "8 hours per week" come from? Where did "industry standards" come from? The AI doesn't know. It generated a number that sounds reasonable.

Here's what happens: You read this, think "okay, so we need a half person," build your headcount plan around it, then in month three of implementation you realize each transaction takes 45 minutes, not the implied 3.2 minutes, and you're massively under-resourced.

What's really wrong: The AI is confabulating. It read a lot of text that included numbers and timelines, and it's replicating that pattern without any actual basis. The real timeline depends on your specific operation: your tools, your customer types, your quality standards, your team's experience level.

How to catch it: Look for numbers that aren't sourced. Any time an AI says "industry standard," "typically," "based on benchmarks," or gives a specific number without context, that's a red flag. Ask: "Where did that number come from?" If the AI can't point to something you verified, assume the number is made up.

Red Flag #3: Missing Critical Steps

The output looks complete but it's missing something fundamental to your operation. Usually something compliance-related, security-related, or something specific to your industry.

Example (bad inventory process for a medical device company):

Step 1: Receive inventory
Step 2: Inspect for damage
Step 3: Enter into system
Step 4: Store in location
Step 5: Pick for orders
Step 6: Ship to customer

Looks reasonable. Except: there's no step for lot tracking (required by FDA). There's no step for expiration date monitoring (critical for medical devices with shelf life). There's no step for temperature control monitoring (required for certain device types). There's no step for quarantine of suspicious units pending quality review.

The AI generated a generic inventory process. It has no idea that medical device inventory is compliance-critical and has rules the generic process completely misses.

What's really wrong: The AI doesn't know your industry's specific rules. It doesn't know what your auditors are looking for. It doesn't know what happens if you get this wrong (regulatory fines, product recalls, liability).

How to catch it: After you read an AI-generated SOP, process, or workflow, ask yourself: "What could go wrong if someone followed this process but forgot step X?" If you can immediately name critical steps that aren't included, the output is incomplete.

The compliance question

For any AI-generated process involving compliance, legal, safety, or financial controls, ask: "Does this account for our audit requirements? Does it create an audit trail? Can we prove we did this?" If you're not sure, have your compliance person review it before using it. AI doesn't know what your auditors care about.

Red Flag #4: Assumes Resources You Don't Have

The output is well-designed for a much larger organization with different constraints. It assumes you have budget, staff, or tools you don't actually have.

Example (bad vendor management process):

Step 1: Create vendor scorecard
Step 2: Track vendor KPIs quarterly
Step 3: Schedule quarterly business reviews
Step 4: Conduct annual strategic planning sessions with top vendors
Step 5: Maintain vendor development program
Step 6: Use vendor relationship management software to track all interactions

This is a beautiful process. It's also impossible if you're a 40-person company with one operations person managing 15 vendors. Quarterly business reviews with all vendors? Vendor development program? That's a 2-3 person job.

What's really wrong: The AI generated the "ideal" process without constraints. It doesn't know that in reality, you're doing vendor management in your spare time, and the process needs to be "bare minimum sufficient" not "strategically optimized."

How to catch it: Read the output and ask: "Who would execute each step and when?" If the answer is "we'd need to hire someone" or "this would take 40 hours per week," you've got a resource mismatch. AI tends to generate processes for Fortune 500 companies even when you're a 50-person startup.

Red Flag #5: Contradicts How You Actually Work

The output assumes a way of working that fundamentally conflicts with your operation. You have dependencies it doesn't mention. Your customers or vendors or internal teams operate in ways that would break the process.

Example (bad contract approval process):

Step 1: Sales submits contract to Legal
Step 2: Legal reviews for 5 business days
Step 3: If issues found, Legal redlines and sends back to Sales
Step 4: Sales negotiates with customer
Step 5: Contract approved and signed

But your reality: Your customers demand 48-hour turnaround on contract approvals. You don't have an in-house legal team. You pay outside counsel $400/hour. A 5-day review is too slow and too expensive. The process as written is broken in your operation.

What's really wrong: The AI designed the process in a vacuum. It doesn't know your customer expectations, your cost structure, your staffing model, or your competitive reality.

How to catch it: After reading the output, ask your team: "Could we actually do this?" If anyone says "yeah, but we'd need to change X first" or "that won't work because of Y constraint," you've got a contradiction. The output doesn't match your operating model.

Red Flag Intensity Matrix

Not all red flags are equally serious. Some are fixable. Some are deal-breakers.

Low severity (fixable):

• Generic phrasing that you can make specific with 15 minutes of customization
  - Missing examples or context that you can add
  - Format that's close but needs tweaking

Medium severity (requires rework):

• Missing non-critical steps that your operation needs (like escalation paths or exception handling)
  - Assumes resources you need to replace with substitutes
  - Timelines that are realistic but don't match your cadence

High severity (reject and start over):

• Missing compliance or safety-critical steps
  - Fabricated metrics that could drive wrong decisions
  - Contradicts your core operating model or customer expectations
  - Process that would create audit or liability risks

Red flags in the high-severity category should trigger a complete rethink. Don't try to iterate and fix them. The foundation is wrong.

Real Example: The Bad Process That Looked Good

Let's walk through an example where I'll show you a bad AI output and break down why it's bad.

Scenario: You're a 60-person professional services firm. You asked AI to create a process for how consultants get allocated to projects.

AI Output (bad):

Consultant Allocation Process

Step 1: Project Manager submits resource request
- Input: Project name, required skills, start date, duration
- Output: Resource request logged in tracking system
- Owner: Project Manager
- Timeline: Same day

Step 2: Resource Manager reviews request
- Input: Resource request from Step 1
- Action: Review consultant availability
- Output: Identified available consultants matching required skills
- Timeline: 2 business days

Step 3: Assign consultant
- Input: List of available consultants
- Action: Match consultant skills to project needs
- Output: Consultant assigned to project
- Timeline: 1 business day

Step 4: Notify consultant
- Input: Assignment confirmation
- Action: Send email to consultant with project details
- Output: Consultant acknowledges assignment
- Timeline: Same day

Step 5: Begin project
- Input: Consultant acknowledgment
- Action: Consultant begins work on project
- Timeline: Project start date

Why this is bad (and where you'd catch it):

Red Flag #1, Generic: This applies to any services firm. It doesn't account for your specific issues. (Maybe you have overallocated consultants. Maybe consultants push back on assignments. Maybe the resource manager job is actually three people's job split across the company.)

Red Flag #2, Fabricated timelines: "2 business days" for resource manager review. Where did that come from? Your resource manager works part-time on allocation. Realistically, it takes 5 business days because they're working on other things too. Now you tell project managers "2 days" and they're angry by day 3.

Red Flag #3, Missing critical steps:

• No mention of bench time or utilization targets. How does the resource manager decide between two equally qualified available consultants? AI doesn't know.
  - No mention of consultant preferences or conflicts. Maybe Consultant A doesn't want to work with Client X, or is fully booked in two weeks but the AI doesn't know to ask.
  - No mention of what happens if there's no available consultant matching the skills. The process just stops?
  - No mention of who has authority to override the resource manager (for example, if a senior partner demands a specific consultant). Missing escalation path.
  - No mention of updating project financials with the consultant's rate. Your accounting department is waiting for this information, and it's not in the process.

Red Flag #4, Assumes resources: This assumes you have a "tracking system" where resource requests are logged. You don't. You use email. Consultants acknowledge assignments via Slack. The process doesn't match your tools.

Red Flag #5, Contradicts how you work: The process assumes project managers submit requests and wait. In reality, they call the resource manager directly and negotiate. They'll push back if their preferred consultant isn't available. The process doesn't account for negotiation or pushback.

What you'd do: Look at this output and say "No, this isn't it. The skeleton is there, but the critical pieces are missing." Then either ask AI to fix these specific issues, or build it yourself because you understand your operation better.

The Verification Checklist for Bad Outputs

Use this checklist when you get AI output. If you check "yes" on any red flag items, the output needs significant work.

Could this apply to any company? (Generic flag)

Are there numbers or metrics without sources? (Fabrication flag)

Are there critical steps I know are essential but don't appear? (Missing steps flag)

Does this assume people/budget/tools I don't have? (Resource flag)

Does this contradict how we actually work? (Operating model flag)

Would an auditor care about this? (Compliance flag)

Would my team push back on this? (Usability flag)

More than two "yes" answers? The output needs significant work. More than three? Build it yourself using AI as a brainstorming tool, not as the source of truth.

Try This Now: Spot the Red Flags

I'm going to give you a bad SOP and I want you to find the red flags. This is practice in pattern recognition.

Scenario: You're an Operations Manager at a 100-person manufacturing company. You asked AI: "Create an SOP for our purchase order approval process."

AI Output (contains red flags):

Purchase Order Approval Process

Step 1: Employee submits purchase request
- Employee identifies needed item and submits form with: item description, quantity, estimated cost
- Timeline: Same day

Step 2: Manager reviews request
- Employee's direct manager reviews request for accuracy and business need
- Timeline: 1 business day

Step 3: Purchasing reviews request
- Purchasing department verifies vendor and pricing are appropriate
- Timeline: 2 business days

Step 4: Finance approves request
- Finance approves if budget is available
- Timeline: 1 business day

Step 5: Purchase order issued
- Purchasing creates PO and sends to vendor
- Timeline: Same day

Step 6: Goods received
- Receiving team inspects goods upon delivery
- Timeline: Upon arrival

Step 7: Invoice match and payment
- Finance matches invoice to PO and authorizes payment
- Timeline: 3 business days from receipt

Red flags to find:

(1) Generic: Applies to any company. Doesn't account for manufacturing-specific issues like supplier qualification, material specifications, or lead times for long-lead items. No mention of how purchasing handles rush orders or how supply chain contingencies work.

(2) Missing critical step: No mention of approval limits. Who approves a $500 request? Who approves a $50,000 request? Manufacturing equipment orders are often large. There's no tier-based approval. This is critical for financial controls.

(3) Missing compliance: No mention of purchase order tracking for regulatory compliance or inventory management. For manufacturing, POs feed into inventory systems and compliance tracking. That's missing.

(4) Resource assumption: Assumes separate Purchasing and Finance teams. Many smaller manufacturers have one person doing both. The process breaks down if you don't have that headcount.

(5) Missing step: For manufacturing, suppliers often need to be pre-qualified (quality certifications, ISO standards, capacity, price agreements). The process skips vendor qualification. You can't just use any vendor for critical parts.

(6) Timeline mismatch: "2 business days" for purchasing review is fine for standard items. For specialized manufacturing parts, lead time negotiation can take weeks. The process doesn't account for different timelines for different item types.

(7) Missing exception handling: What if goods arrive damaged? What if the invoice doesn't match the PO? What if the supplier is late? The process doesn't cover any of these scenarios.

Your conclusion: This is 60% right and 40% dangerously incomplete. You could try to iterate and fix it, but honestly, the foundation is too generic. It's easier to build the real process yourself with AI as a brainstorming tool, not as the source of truth.

What to Do Monday Morning

• Get one piece of AI-generated content from your recent work. (A process, an SOP, a plan, anything operational.) Run it through the red flag checklist.
  - Mark the red flags you find. Don't grade yourself on how many you find. Just build your pattern recognition muscle.
  - For each red flag, note what information the AI was missing. Example: "Red flag #3: missing approval limits. The AI didn't know about our tiered approval structure for different spend levels."
  - Ask yourself: Is this fixable with iteration, or is the foundation wrong? This judgment is the core skill.
  - Document one pattern you notice. Example: "AI always misses our approval authority structure. Next time, I'll include that in the context upfront."

Key Takeaways

• Good output is specific to you. Generic output that applies to any company is a red flag. Real processes are idiosyncratic.
  - Watch for fabricated metrics and numbers. If the AI can't source a number, assume it made it up. Don't build plans around made-up numbers.
  - Missing steps are your job to catch. AI doesn't know your compliance requirements, your audit needs, or your industry-specific rules. You have to verify it didn't miss critical steps.
  - Compare the output to your constraints. Does it assume resources you don't have? Does it contradict how you actually work? If yes, it's not implementable.
  - Distinguish between fixable and unfixable red flags. Missing details you can add? Fixable. Missing entire approval structure? Unfixable, foundation is wrong.
  - Use the red flag checklist. Seven questions. More than two "yes" answers means significant rework needed.

FAQs

Can I use bad output and just fix it as we go?

Only if the red flags are low-severity (wording, examples, format). If the red flags are structural (missing steps, wrong assumptions about resources), don't implement and iterate. That's how you end up with a broken process that everyone resents because it never worked from day one.

What if my team uses an AI-generated process and it breaks down?

Capture what went wrong. That's feedback about which red flags you need to watch for in your operation. Example: "Team had to workaround the lack of exception handling." Next time you review AI output, you'll know to check for exception handling.

Should I show my team the red flags I find?

If you're asking them to implement something, yes. Flag the issues upfront. "This output is 70% done. It's missing approval limits and exception handling. We're going to add those before you use it." That prevents them from blindly following broken logic.

What if AI is right and I'm wrong?

That's possible. If you flag something as a red flag and a team member pushes back ("Actually, that's how we should do it"), listen. But "we should" and "we do" are different things. If it contradicts current practice, that's a change management issue, not a quality issue.

How do I know if I'm being too critical?

If the red flags are low-severity and fixable in 30 minutes, you're being too critical. If they're high-severity and would take days to rework or would break if implemented as-is, you're being appropriately critical. The time-to-fix is the dividing line.