How to know if you're getting ripped off on AI development

Hey,

A founder asked me last week: "Is $12K reasonable for an AI chatbot?"

I looked at the scope. It was a $4K project.

He was about to get ripped off.

Here's exactly what AI development should cost (and what you actually get at each price point).

THE PROBLEM:

AI development pricing is all over the place.

Same project, three quotes:

• Agency A: $15K, 6 weeks

• Agency B: $45K, 12 weeks

• Agency C: $8K, 4 weeks

How do you know what's fair?

Most founders pick the middle option thinking it's "safe."

Bad move.

Price doesn't indicate quality. It indicates what the agency thinks they can charge you.

Here's what you should actually pay:

PRICING BY PROJECT TYPE:

SIMPLE AI AGENT: $4K - $12K (4-6 weeks) What you get:

• Single function (email summarizer, document analyzer, etc.)

• Basic error handling

• Simple UI or API

• 14-day post-launch support

• Documentation

Examples:

• Email categorizer that sorts your inbox

• Meeting notes generator from recordings

• Invoice data extractor

• Basic customer support chatbot

Red flag: If someone quotes $30K+ for this, walk away.

MEDIUM AI AGENT: $15K - $28K (8-12 weeks) What you get:

• Multiple functions or complex workflow

• Production-grade error handling

• Custom UI + backend

• Integration with 2-3 systems

• 30-day support

• Training for your team

Examples:

• Document processing system (multiple formats)

• Customer support AI (with ticketing system)

• Content generation tool (with brand guidelines)

• Sales outreach automation

Red flag: If they can't explain the tech stack in simple terms, they're hiding something.

COMPLEX AI SYSTEM: $28K - $40K (12-18 weeks) What you get:

• Multi-agent system or complex logic

• Enterprise-grade everything

• Multiple integrations (5+)

• Advanced monitoring dashboard

• 60-day support

• Ongoing maintenance plan

Examples:

• Full customer service automation

• Multi-step document workflow

• AI-powered analytics platform

• Custom RAG system with knowledge base

Red flag: If timeline is under 8 weeks, they're cutting corners.

WHAT SHOULD BE INCLUDED (AT EVERY PRICE POINT):

These are NON-NEGOTIABLE:

✓ Error handling (when APIs fail, it recovers gracefully)

✓ Rate limiting (doesn't hit API limits)

✓ Basic monitoring (you can see what's happening)

✓ Testing (they actually tested edge cases)

✓ Documentation (so you're not locked in)

✓ Source code (you own what you paid for)

✓ Post-launch support (at least 14 days)

If these aren't included, the price should be 30-40% lower.

RED FLAGS IN PRICING:

🚩 "We charge $200/hour" Translation: Your project will have infinite scope creep. Final bill will be 2-3x estimate.

🚩 "50% upfront, 50% on delivery" Translation: No milestones = no accountability. If they disappear, you're screwed.

🚩 "We need to start before we can estimate" Translation: They have no idea what they're doing. You'll fund their learning.

🚩 "Basic package: $5K, Premium: $50K" Translation: Artificial anchoring. The "basic" is useless, forcing you to pay for "premium."

🚩 "Price depends on your budget" Translation: They'll charge whatever you say you have.

GREEN FLAGS IN PRICING:

✅ Fixed price with clear scope They know exactly what they're building and what it costs.

✅ Payment tied to milestones (30/40/30) 30% to start, 40% at halfway point, 30% at completion.

✅ Detailed breakdown of what's included Not just "we'll build an AI agent" but specific features, deliverables, timeline.

✅ Clear exclusions listed "This does NOT include X, Y, Z" shows they've thought it through.

✅ Refund policy for discovery phase "If we can't build what you need, we'll refund the discovery fee."

THE 7 QUESTIONS TO ASK:

Before accepting any quote, ask these:

1. "What's included in this price?"

   1. Good answer: Detailed list of features, deliverables, support

   2. Bad answer: "Everything you need" (vague)

2. "What's NOT included?"

   1. Good answer: Specific exclusions listed

   2. Bad answer: "We'll handle everything" (scope creep incoming)

3. "What happens when the API fails?"

   1. Good answer: "We have retry logic, fallbacks, and monitoring"

   2. Bad answer: "It won't fail" (yes it will)

4. "Can I see a similar project you've built?"

   1. Good answer: Shows you a real example, explains tech decisions

   2. Bad answer: "All our work is confidential" (they haven't built anything)

5. "What's the payment structure?"

   1. Good answer: "30/40/30 tied to milestones"

   2. Bad answer: "50% upfront" (they need your money before starting)

6. "Who owns the code?"

   1. Good answer: "You do, we'll transfer everything"

   2. Bad answer: "We'll license it to you" (vendor lock-in)

7. "What happens after launch?"

   1. Good answer: "30 days included support, then $X/month for maintenance"

   2. Bad answer: "We'll figure it out" (no plan)

REAL EXAMPLE BREAKDOWN:

Project: Invoice processing AI agent What it does: Extracts data from 200+ invoices daily

QUOTED: $45,000, 12 weeks ACTUAL FAIR PRICE: $18,000, 8 weeks

Why the difference?

The $45K quote included:

• "AI consulting" (useless)

• "Custom ML model training" (unnecessary, Claude API works)

• "Enterprise security audit" (overkill for internal tool)

• 40% agency markup

The $18K includes everything actually needed:

• Claude API integration ($150/month in API costs)

• Error handling & retry logic

• CSV export functionality

• Monitoring dashboard

• 30-day support

• Full documentation

Same result. 60% less cost.

HOW TO CALCULATE FAIR PRICE:

Use this formula:

(Developer weeks × $1,000) + (API costs × 12 months) + 20% buffer

Example:

• 6 weeks of development = $6,000

• API costs = $150/month × 12 = $1,800

• 20% buffer = $3,960

Round to $12,000 for a 6-week project.

If someone quotes you $40K for the same scope, they're either:

• Overcharging (most common)

• Extremely inefficient (avoid)

• Including things you don't need (question everything)

WHAT INCREASES PRICE (LEGITIMATELY):

These DO justify higher costs:

✓ Multiple integrations (each adds 1-2 weeks)

✓ Compliance requirements (HIPAA, SOC2, etc.)

✓ High accuracy needs (98%+ vs 85%)

✓ Custom UI/UX design (not template)

✓ Training and documentation (extensive)

✓ High volume/scale (1000+ users)

These DON'T:

✗ "Premium AI models" (you're paying for Claude/GPT anyway)

✗ "Proprietary technology" (usually just wrapper code)

✗ "Agency overhead" (not your problem)

✗ "Market rate" (meaningless phrase)

THE DECISION FRAMEWORK:

Quote seems high? Ask yourself:

□ Is the scope crystal clear? (If no → high risk)

□ Are deliverables specific? (If no → scope creep)

□ Is timeline realistic? (Under 4 weeks = cutting corners)

□ Are exclusions listed? (If no → surprises later)

□ Can they show similar work? (If no → inexperienced)

□ Do they explain tech choices? (If no → don't understand)

□ Is payment tied to milestones? (If no → no accountability)

If you check 6/7: Price is probably fair

If you check 4/7: Negotiate or walk

If you check <3/7: Definitely getting ripped off

THE BOTTOM LINE:

Fair AI development pricing:

• Simple: $4K - $12K

• Medium: $12K - $28K

• Complex: $28K - $50K

Anything significantly higher? Ask why.

Most founders overpay because they:

• Don't know what questions to ask

• Think higher price = better quality (wrong)

• Fear negotiating (agencies expect it)

• Can't evaluate technical proposals

Now you can.

See you next Saturday,

Immanuel

P.S. - Evaluating a proposal right now? Forward it to me. I'll tell you if it's fair (no charge, just want to help founders not get ripped off).