The Red Belt does not apply AI to everything. It applies AI where leverage is maximum and fragility is minimum. The error of the 95% that fail is exactly the opposite: applying AI where it is prestigious (visible to the CEO, a brilliant demonstration) instead of where it produces real value.
The second section of the Red Belt is the technical arsenal: the three operational areas the book identifies as ideal for automated intelligence. Not because they are easy, but because they combine three conditions:
- They have high repetitiveness (similar decisions thousands of times)
- They have structured data available (measurable signal)
- They have material impact (moving the needle of the business, not only demonstrating technology)
We go through the three areas in the planes of business, automation and AI.
1. Area 1 (security): continuous detection of anomalous patterns
Why security is ideal for AI
Business security (digital and operational) meets the three conditions perfectly:
- Repetitiveness: millions of events per day (logins, transactions, accesses, queries)
- Structured data: every event has a timestamp, user, IP, action, result, an ideal format
- Impact: a breach can destroy 30-50% of the value of an SMB in weeks
A person cannot monitor millions of events. A well configured AI can, and it detects anomalous patterns a human would never see.
Concrete applications
Three applications of immediate value:
-
Transactional fraud detection: models that learn the normal pattern of each customer and alert when a transaction does not fit. They reduce loss from fraud by 60-80%.
-
Intrusion detection: systems that learn the normal network and alert when a behavior deviates. They reduce MTTR (mean time to respond) from days to minutes.
-
Compliance detection: models that review transactions against regulatory rules in real time. They reduce the risk of a fine and reputational risk.
The Red Belt rule in security
The human never leaves the loop completely in security. The system detects, prioritizes, recommends. The human confirms high impact decisions. The reason: the consequences of a false negative (missing a real threat) are catastrophic. The consequences of a false positive (alerting unnecessarily) are acceptable.
Intelligent automation reduces 95% of the human work (filtering signal from noise), but the remaining 5% continues to be a trained human decision.
2. Area 2 (finance): structured processing at speed
Why finance is ideal
Finance meets the three conditions as well:
- Repetitiveness: invoicing, reconciliation, reporting, variance analysis, all of them repetitive
- Structured data: accounting by definition is structured data
- Impact: financial decisions badly taken or applied late destroy value predictably
Concrete applications
Four applications of high leverage:
-
Automated reconciliation: models that reconcile bank transactions with accounting entries in seconds. What took 20 hours a week of an accountant now takes 30 minutes.
-
Predictive forecasting: models that project cash flow, sales, costs, based on structured history. They improve precision by 20-40% over human forecasts.
-
Variance analysis: models that detect when a KPI deviates from the normal pattern before it becomes an obvious problem.
-
Credit decisions: models that evaluate customer risk with hundreds of variables, deciding approval in seconds.
The Red Belt rule in finance
Final decisions with material impact continue to be human. AI proposes, recommends, prioritizes. The CFO (or equivalent) decides. The reason: errors in finance have regulatory, fiscal and credibility consequences that a system without context cannot evaluate.
Typical pattern: AI does 90% of the work (processing, calculations, alerts), the human does 10% (final decisions, strategic context, communication with stakeholders).
3. Area 3 (customer conversations): personalization at scale
Why conversations is ideal, with asterisks
This area is the most promising and the most dangerous of the arsenal. It meets the three conditions:
- Repetitiveness: thousands of similar inquiries per month (FAQs, basic support, lead qualification)
- Structured data: conversations are transcribed, tagged and measured
- Impact: customer experience determines retention, repurchase, NPS, all of it critical
But here the Red Belt rule becomes critical: NOT every conversation is delegated to AI. The distinction is brutal.
Conversations that ARE automated
- FAQs and level 1 support (frequent questions with known answers)
- Initial qualification of leads (is this lead serious? what stage is it in?)
- Transactional confirmations (order status, payment reminders)
- Initial onboarding (tutorials, guided configuration)
- Assisted search in the customer's knowledge base
These conversations have clear patterns, reasonable customer expectations, and a low cost of error.
Conversations that are NOT automated
- Conflict resolution (upset customer, serious problem)
- Negotiations (prices, terms, contracts)
- Conversations of high strategic value (top customers, partnerships)
- Emotional support (customer in crisis, sensitive situations)
- Communication of bad news (cancellation, serious error of the company)
Here AI produces more damage than value. The customer who receives an automated answer in a sensitive moment feels belittled. The permanent friction it generates surpasses any efficiency achieved.
The operational rule of the Red Belt in conversations: automate conversations where the customer does not lose dignity if he discovers it is an AI, and do not automate conversations where the customer does. A customer asking "how do I activate my account?" does not lose dignity. A customer canceling because he is angry does. The border is not technical: it is ethical and experiential.
The hybrid human + AI framework
The mature practice is not choosing between human and AI. It is designing the flow where AI and human play the correct roles:
- Triage: AI classifies every incoming inquiry: what level? what urgency? what tone from the customer?
- Autonomous resolution: if the inquiry meets the criteria (level 1, neutral tone, known FAQ), AI resolves it
- Assistance to the agent: if the inquiry moves up a level, AI prepares context, proposes an answer, suggests actions
- Human decision: the human agent decides and executes, with AI as a copilot, not as an autopilot
- Learning: every interaction feeds back into the model to improve the triage and the assistance
Automation is not the opposite of humanity: it is what protects humanity for the conversations that really matter. When AI does the routine well, the human can be fully present in what is important.
The test of maximum leverage
Before investing in any area for advanced automation, do this test:
- Is it repetitive? (minimum 1,000 instances per month)
- Does it have structured data? (measurable signal, not inferred from subjective free text)
- Does it have material impact? (it moves revenue, margin, risk, or NPS in a measurable way)
- Does the human stay in the loop for final decisions of high impact? (not total autonomy: autonomy with a perimeter)
- Is the consequence of an error recoverable? (an error of the AI does not destroy the customer relationship, does not generate an irreversible fine)
Five yeses = candidate for Red Belt. Fewer than five = go back to Green or to Blue before advancing.
Frequently asked questions
Three areas according to the Red Belt: security (fraud detection, intrusion, compliance), finance (reconciliation, forecasting, variance analysis, credit decisions), customer conversations (FAQs, level 1 support, lead qualification, transactional confirmations). The three meet three conditions: high repetitiveness (thousands of instances per month), structured data available, material impact on the business. Areas that are NOT ideal: strategic decisions, sensitive conversations with top customers, negotiations, senior hires, conflict resolution. The difference is where AI amplifies human capability versus where it substitutes it with high hidden costs.
With the principle "AI proposes, the human decides in material final decisions." AI does the bank reconciliation (what took 20 hours a week now takes 30 minutes), predictive forecasting (it improves precision by 20-40% over human forecasts), variance analysis (it detects deviations before they are an obvious problem), and credit processing (it evaluates hundreds of variables in seconds). But the CFO keeps deciding: the monthly accounting close, investment decisions, communication with auditors, handling of material exceptions. Errors in finance have regulatory and fiscal consequences that a system without human context cannot evaluate. Mature pattern: AI does 90% of the processing, the human does 10% of the final decisions.
No, they complement it in the correct flow. Well designed chatbots and AI agents handle FAQs, level 1 support, initial qualification of leads, transactional confirmations, initial onboarding, conversations with clear patterns and a low cost of error. BUT they do not replace the human in: conflict resolution (upset customer), negotiations (prices, terms, contracts), communication of bad news, emotional support, conversations with top customers of high strategic value. The mature company designs the hybrid flow: AI does the triage (it classifies urgency and level), resolves the routine, assists the human agent with context and proposals for the complex part. The customer is never left without a human option when he needs it.
Five areas where AI produces more damage than value: 1) Final strategic decisions (vision, direction, M&A). 2) Conversations of high value with top customers or partners. 3) Conflict resolution where the customer would lose dignity discovering he spoke with an AI. 4) Senior hires and performance evaluations with deep impact on people. 5) Communication of bad news or crisis. The operational rule: automate conversations where the customer does not lose dignity if he discovers it is AI; do not automate the ones where he does. The permanent friction generated by the inappropriate use of AI in these areas surpasses any efficiency achieved, and damages the brand in the long term.
The third section
Philosophy plus arsenal, both complete. The third section is missing: the ritual of integration. How to apply this arsenal with purpose, not as a technological demonstration. The difference between installing AI and living it as a continuous practice of the Red Belt.
To identify the right consultant to help you deploy this arsenal (12 filters that separate signal from noise), read How to Choose an AI Consultant. For the framework that separates AI strategy from AI implementation, read AI Strategy vs AI Implementation.
Want the full method? Read AI Black Belt: Fundamentals Before the Prompt. For executive AI consulting or keynotes and workshops.
Go deeper
Want to bring your team to the next belt?
Book a discovery call or explore the full book.