Starting With Problems, Not Technology

The best AI projects don't start with "We should use AI." They start with "This problem is killing us" — and then AI turns out to be the solution.

This chapter helps you find those problems and evaluate which ones AI can actually solve.

The AI Opportunity Filter

Not every business problem is an AI problem. Run opportunities through this filter:

Filter 1: Is There Data?

AI needs data. No data, no AI.

Questions to ask:

• Do we have historical data about this process?
• How much data? (Thousands of examples minimum for most AI)
• What format is it in? (Structured, unstructured, PDFs, images)
• Can we actually access it? (Systems, permissions, privacy)
• Is it labeled? (Do we know the outcomes?)

If data is missing: Stop. Either build data collection or choose another problem.

Filter 2: Is There a Pattern?

AI finds patterns. If outcomes are random or depend on factors you don't capture, AI won't help.

Questions to ask:

• Do experienced humans do this successfully? (If humans can learn it, AI might)
• Are there identifiable signals in the data?
• Is the outcome predictable or random?

If outcomes are random: AI won't help. Look elsewhere.

Filter 3: Is Automation Valuable?

Some tasks aren't worth automating. The juice isn't worth the squeeze.

Questions to ask:

• How often does this task occur? (Frequency)
• How long does it take? (Duration)
• How expensive are the people doing it? (Cost)
• What's the error rate currently? (Quality)
• What's the cost of errors? (Risk)

Quick math: Hours spent × hourly cost × potential improvement = rough value. If this doesn't justify investment, stop.

Filter 4: Can It Be Wrong Sometimes?

AI makes mistakes. If perfect accuracy is required, AI alone isn't sufficient.

Questions to ask:

• What's the current error rate?
• What error rate would still be valuable?
• What's the cost of an error?
• Can humans review AI decisions?

If errors are catastrophic: AI can assist humans but shouldn't decide alone.

Filter 5: Will People Use It?

Technology that isn't adopted is worthless. Consider the human element.

Questions to ask:

• Who would use this?
• What's in it for them?
• Will they trust it?
• Does it fit their workflow?
• What change management is needed?

If adoption is unlikely: Fix the adoption problem first or choose another opportunity.

High-Value AI Use Cases by Function

Here's where AI typically delivers value, organized by business function:

Customer Service

Opportunities:

• Chatbots for common questions
• Routing inquiries to right agents
• Suggested responses for agents
• Sentiment analysis to prioritize
• Automated ticket classification

Data requirements: Historical tickets, resolutions, customer satisfaction scores

Typical results: 30-50% of inquiries handled automatically, faster response times, more consistent quality

Sales

Opportunities:

• Lead scoring (which leads to prioritize)
• Churn prediction (which customers are at risk)
• Next-best-action recommendations
• Email and outreach drafting
• Call analysis and coaching

Data requirements: CRM data, historical win/loss, customer behavior

Typical results: Sales team focuses on better leads, earlier churn intervention, more personalized outreach

Marketing

Opportunities:

• Content generation and variation
• Personalization at scale
• Ad copy optimization
• Customer segmentation
• Campaign performance prediction

Data requirements: Campaign history, customer data, engagement metrics

Typical results: More content produced, better targeting, improved conversion rates

Operations

Opportunities:

• Demand forecasting
• Inventory optimization
• Route optimization
• Predictive maintenance
• Quality control

Data requirements: Historical operations data, sensor data, transaction records

Typical results: Lower inventory costs, fewer stockouts, reduced downtime, improved quality

Finance

Opportunities:

• Fraud detection
• Invoice processing
• Expense categorization
• Cash flow forecasting
• Anomaly detection

Data requirements: Transaction history, invoices, financial records

Typical results: Faster processing, fewer errors, reduced fraud losses

HR

Opportunities:

• Resume screening
• Employee churn prediction
• Skills matching
• Training recommendations
• Survey analysis

Data requirements: HR records, performance data, survey responses

Typical results: Faster hiring, better retention, more relevant training

Legal and Compliance

Opportunities:

• Contract review and extraction
• Regulatory monitoring
• Due diligence research
• Document classification
• Policy compliance checking

Data requirements: Contracts, regulations, historical decisions

Typical results: Faster review, lower risk, reduced legal costs

The Opportunity Assessment Framework

For each potential opportunity, assess across four dimensions:

Impact

How much business value if successful?

High: Revenue increase, major cost reduction, competitive advantage Medium: Efficiency improvement, quality increase, risk reduction Low: Minor improvement, convenience

Feasibility

How likely is technical success?

High: Data exists and is accessible, problem is well-defined, similar solutions exist Medium: Data needs work, some uncertainty, but achievable Low: Data is poor, problem is ill-defined, no similar solutions exist

Effort

How much investment is required?

Low: Off-the-shelf solution, minimal integration, quick pilot Medium: Some customization, moderate integration, few months High: Custom development, complex integration, year-plus

Risk

What could go wrong?

Low: Errors are minor, reversible, low stakes Medium: Errors have cost but are manageable High: Errors are expensive, embarrassing, or dangerous

The Priority Matrix

Plot opportunities on a 2x2:

High Impact + High Feasibility = Do First These are your best opportunities. Prioritize them.

High Impact + Low Feasibility = Investigate Worth pursuing if you can improve feasibility (better data, clearer problem).

Low Impact + High Feasibility = Quick Wins Good for building momentum and capability, but don't over-invest.

Low Impact + Low Feasibility = Don't Bother These waste resources. Cut them.

The Discovery Process

How to systematically find opportunities in your organization:

Step 1: Pain Point Interviews

Talk to people doing actual work. Ask:

• What tasks take the most time?
• What's frustrating or repetitive?
• Where do errors happen?
• What decisions are hard to make?
• What information do you wish you had?

Listen for volume, repetition, and frustration.

Step 2: Process Mapping

For promising areas, map the current process:

• What triggers the process?
• What steps are involved?
• What decisions are made?
• What data is used?
• What's the output?
• How long does it take?
• Where are the bottlenecks?

This reveals where AI could intervene.

Step 3: Data Audit

For each opportunity, assess data reality:

• What data exists?
• Where is it?
• What format?
• What quality?
• What access requirements?
• What privacy constraints?

Don't trust assumptions. Actually look.

Step 4: Quick Validation

Before committing, validate:

• Can we get a sample of real data?
• Can we define success metrics?
• Do stakeholders agree on the problem?
• Is someone willing to own it?

If any answer is no, address it before proceeding.

Common Mistakes in Opportunity Selection

Chasing Shiny Objects

Pursuing AI because it's interesting, not because it solves important problems.

Fix: Always tie AI to business metrics. If you can't connect to revenue, cost, or risk, reconsider.

Ignoring Data Reality

Assuming data exists and is usable without checking.

Fix: Data audit before commitment. Every time.

Overscoping

Trying to solve everything at once.

Fix: Start narrow. Prove value. Expand.

Underestimating Integration

Forgetting that AI needs to connect to existing systems and workflows.

Fix: Include integration in feasibility assessment. It's often harder than the AI.

Ignoring Humans

Forgetting that people need to use, trust, and adapt to AI.

Fix: Involve users early. Plan for change management.

AI Opportunity Template

Use this template to document each opportunity:

OPPORTUNITY: [Name]

PROBLEM
- What pain are we solving?
- Who feels this pain?
- How much does it cost (time, money, risk)?

DATA
- What data exists?
- Where is it?
- What quality?
- What access?

SOLUTION CONCEPT
- What would AI do?
- How would it fit the workflow?
- What human oversight needed?

SUCCESS METRICS
- How will we measure success?
- What's the target?
- What's the current baseline?

ASSESSMENT
- Impact: High / Medium / Low
- Feasibility: High / Medium / Low
- Effort: High / Medium / Low
- Risk: High / Medium / Low

RECOMMENDATION
- Priority: Do First / Investigate / Quick Win / Don't Bother
- Next step:

What's Next

You've identified promising opportunities. Now you need to decide: Build custom AI or buy existing solutions?

The next chapter covers the build vs. buy decision.