Liam Lawson

Most companies train everyone the same way.
Then wonder why AI adoption fails.

Here's what each level actually needs:

FOUNDATION LEVEL: Basic Prompt Engineering
→ Who needs it: Every employee in the company
→ Skills required:

- Writing clear, specific instructions
- Understanding context windows
- Basic formatting and structure
→ Time investment: 2-4 hours of training
→ ROI: 5-10 hours saved per week per person
→ Tools: ChatGPT, Claude, basic AI interfaces

This is your baseline.
No one should skip this level.

MIDDLE LEVEL: Workflow Design & Tool Selection
→ Who needs it: Managers, team leads, project owners
→ Skills required:

- Identifying automation opportunities
- Evaluating AI tools for specific use cases
- Designing multi-step AI workflows
- Integration with existing systems
→ Time investment: 20-40 hours of learning + practice
→ ROI: 15-25 hours saved per week per team
→ Tools: Zapier, Make, API connections, workflow builders

This is where productivity multiplies.
One good workflow saves hundreds of hours.

TOP LEVEL: Model Fine-Tuning & Custom Development
→ Who needs it: AI specialists, ML engineers, technical leads
→ Skills required:

- Model selection and evaluation
- Fine-tuning on custom datasets
- API development and optimization
- Performance monitoring and iteration
→ Time investment: 100+ hours of deep learning
→ ROI: Company-wide competitive advantage
→ Tools: OpenAI API, Claude API, custom infrastructure

This is strategic differentiation.
Most companies don't need this yet.

THE BIGGEST MISTAKE:

Companies send everyone to advanced training.
Specialists sit through basic prompt engineering.
Frontline employees get overwhelmed by technical depth.

THE SMART APPROACH:

Train 100% of employees at Foundation Level.
Train 20% of employees at Middle Level.
Train 5% of employees at Top Level.

Match training to actual job requirements.
Stop wasting time on skills people won't use.

P.S. Want to learn more about AI?

1. Scroll to the top
2. Click "Visit my website"
3. Sign-up for our free newsletter

The experts were wrong.

Embarrassingly wrong.

Here's what they said versus reality.

1 - "AI will never create art"

Predicted: AI lacks creativity for real art.
Reality: Midjourney generates museum-quality work, AI wins art competitions.

2 - "AI can't code"

Predicted: Programming requires human logic forever.
Reality: GitHub Copilot writes 40% of code at major companies, non-coders build apps.

3 - "Deepfakes will destroy democracy"

Predicted: Synthetic media makes truth impossible to verify.
Reality: Detection tools evolved faster, most deepfakes debunked within hours.

4 - "AI will replace all jobs by 2020"

Predicted: Mass unemployment and economic collapse.
Reality: More jobs created than eliminated, new AI careers emerged.

5 - "AGI is 50 years away"

Predicted: Artificial General Intelligence won't arrive until 2070.
Reality: Leading researchers now predict AGI by 2027-2030.

6 - "AI can't understand context"

Predicted: AI lacks true comprehension and reasoning.
Reality: GPT-4 passes bar exam, understands sarcasm and metaphor.

7 - "Regulation will stop AI progress"

Predicted: Government oversight will slow development dramatically.
Reality: AI capabilities advanced faster than regulators could respond.

8 - "Only big tech can build AI"

Predicted: AI requires billions in compute resources.
Reality: Open-source models compete with proprietary systems, startups train models under $1M.

Every single expert prediction failed.

Not just wrong.
Catastrophically wrong.

The pattern is clear: experts consistently underestimate AI progress.

Assume everything happens faster than predicted.
Assume capabilities exceed expert expectations.
Assume your industry disrupts sooner than forecasted.

Which failed prediction shocked you most?

P.S. Want to learn more about AI?

1. Scroll to the top
2. Click "Visit my website"
3. Sign-up for our free newsletter.

Most companies blame lazy employees for low AI adoption.
The real problem is understanding what's actually blocking them.

Here's what's really happening:

When fear drives resistance:

31% of employees are scared.
Job security anxiety keeps them from trying.
They're terrified of looking incompetent.
Making mistakes with new technology feels risky.

How to fix it:
Leadership needs to message "AI augments, doesn't replace."
Create private learning environments for safe practice.
Celebrate early adopters to normalize usage.

When confusion drives resistance:

28% of employees are overwhelmed.
Too many tools with unclear purposes.
No clear use cases for their specific role.
Technical jargon shuts them down completely.

How to fix it:
Build role-specific use case libraries.
Keep tutorials to 15 minutes maximum.
Ban technical jargon in all communications.

When inertia drives resistance:

24% of employees see no reason to change.
Current processes work "well enough" for them.
No time allocated for learning anything new.
Zero accountability for adoption.

How to fix it:
Make AI usage part of manager check-ins.
Block learning time directly in calendars.
Link AI usage to performance reviews.

When multiple forces combine:

11% experience both fear and confusion.
They're terrified they won't understand it and will fall behind.
Solution: Pair them with patient power users.

8% experience both fear and inertia.
Change feels dangerous, staying put feels safer.
Solution: Show small wins with real numbers.

6% experience both confusion and inertia.
They don't understand it and don't have time to figure it out.
Solution: Give them pre-built workflows to copy-paste.

5% experience all three at once.
These are your hardest employees to convert.
They need one-on-one coaching and extended timelines.

The biggest mistake is treating all resistance the same.
Fear needs reassurance.
Confusion needs clarity.
Inertia needs accountability.

Figure out what's blocking your team.
Deploy the right solution for that specific problem.

P.S. Want to learn more about AI?

1. Scroll to the top
2. Click "Visit my website"
3. Sign-up for our free newsletter

AI didn’t just automate service, it automated calm.

VIPdesk redefined customer experience by removing chaos, not people.

They achieved 70% automation and 92% CSAT, not by replacing empathy

but by designing AI that knows when to pause.

The future of automation is also about psychological design.

When customer service feels calm, trust grows.

When agents feel calm, loyalty follows.

What if “calm per customer” became the next KPI for business performance?

Would leaders start designing for peace instead of for speed?

You don’t need seniority to earn trust, just depth.

He wasn’t hired as an expert. He became one.

By going deep into the product, obsessing over the work, and becoming a resource others relied on, he earned influence most 20-somethings don’t get.

This is how you become indispensable, regardless of age.

Watch the full episode here:
https://lnkd.in/eVbTAdbM

#EarlyCareer #TrustedByExpertise #DepthOverTitle #YoungLeadership #ProductKnowledge

Most companies don't realize they're vulnerable until after the breach.
By then, it's too late and expensive.

Here's how to assess your risk level:

LEVEL 1: MINIMAL THREAT - Isolated AI Tool Usage
→ Employees using personal AI accounts for work
→ Cost of breach: $50K-$200K
→ Prevention: Ban personal accounts, provide enterprise tools, monthly audits

LEVEL 2: LOW THREAT - Enterprise Tools Without Governance
→ Company-wide AI access with no usage guidelines
→ Cost of breach: $200K-$1M
→ Prevention: Written AI policy, data classification training, disable model training

LEVEL 3: MODERATE THREAT - Custom AI Without Security Review
→ Custom models deployed without security audit
→ Cost of breach: $1M-$5M
→ Prevention: Security audit before deployment, input validation, rate limiting

LEVEL 4: HIGH THREAT - AI With Access to Sensitive Systems
→ AI agents with API access to databases and CRMs
→ Cost of breach: $5M-$25M
→ Prevention: Zero-trust architecture, read-only default, real-time monitoring

LEVEL 5: CRITICAL THREAT - AI Making Autonomous Decisions
→ AI systems executing without human approval
→ Cost of breach: $25M-$100M+
→ Prevention: Human-in-the-loop, multi-signature approval, tested kill switches

Most companies sit at Level 2 or 3 without knowing it.

The progression that kills companies:
Month 1: Level 1 (seems harmless)
Month 6: Level 3 (still manageable)
Month 12: Level 4 (breach happens)
Recovery: 18+ months, millions spent

Every level you drop reduces breach probability by 40%.
Every month you wait increases breach cost by $200K.

P.S. Want to learn more about AI?

1. Scroll to the top
2. Click "Visit my website"
3. Sign-up for our free newsletter

Most companies throw money at AI tools randomly.
Then wonder why ROI never materializes.

Here's the allocation that actually works:

For a $100K AI budget:

Tools: $35K (35%)
→ AI platform subscriptions and API costs
→ What this covers: ChatGPT Enterprise, Claude, automation platforms
→ Common mistake: Spending 60%+ on tools, nothing left for training

Training: $25K (25%)
→ Employee education and skill development
→ What this covers: Workshops, certifications, internal training programs
→ Common mistake: Buying expensive tools then skipping training entirely

Consulting: $15K (15%)
→ External expertise for implementation guidance
→ What this covers: Strategy sessions, workflow design, best practice audits
→ Common mistake: Hiring consultants before understanding internal needs

Infrastructure: $10K (10%)
→ Integration, security, and technical setup
→ What this covers: API connections, security audits, monitoring tools
→ Common mistake: Assuming tools work out-of-box with no setup

Testing: $10K (10%)
→ Pilot programs and proof-of-concept validation
→ What this covers: Small-scale rollouts, A/B testing, performance measurement
→ Common mistake: Enterprise-wide rollout without pilots

Contingency: $5K (5%)
→ Buffer for unexpected costs and pivots
→ What this covers: Additional licenses, emergency consulting, tool switches
→ Common mistake: Zero buffer means failure when things go wrong

For a $500K AI budget:

Tools: $175K (35%)
Training: $125K (25%)
Consulting: $75K (15%)
Infrastructure: $50K (10%)
Testing: $50K (10%)
Contingency: $25K (5%)

For a $1M AI budget:

Tools: $350K (35%)
Training: $250K (25%)
Consulting: $150K (15%)
Infrastructure: $100K (10%)
Testing: $100K (10%)
Contingency: $50K (5%)

The pattern that separates winners from losers:

Failed implementations: 70% tools, 10% training, 20% other
Successful implementations: 35% tools, 25% training, 40% other

Tools don't drive adoption.
Trained employees drive adoption.

Most companies flip this ratio and waste millions.

P.S. Want to learn more about AI?

1. Scroll to the top
2. Click "Visit my website"
3. Sign-up for our free newsletter

How Ulta Beauty turned 38M loyalty members into loyal fans with AI

Ulta Beauty faced every marketer’s nightmare:

too much data, not enough connection.

They unified their loyalty, credit, and campaign data through SAS Customer Intelligence 360, using AI to predict what each of their 38M members actually wanted.

⚙️ AI Role:

- Connected siloed data into one journey
- Powered real-time, personalized recommendations
- Optimized campaigns and lowered costs

💥 Result:

95% of sales now come from returning customers.

Business takeaway:

The next generation of loyalty isn’t built on rewards, it’s built on relevance.

Most companies discover waste only after the CFO demands answers.
This flowchart finds savings before that happens.

Start here and follow your path:

START: What's your current monthly AI spend?

PATH A: Under $5K/month
→ Decision: Are you using enterprise plans?

- YES: Downgrade to team or individual tiers
• Savings: 40-60% ($2K-$3K/month)
• Action: Audit actual seat usage, remove inactive users
- NO: Check if you're using premium models for simple tasks
• Savings: 30-50% ($1.5K-$2.5K/month)
• Action: Route basic queries to GPT-4o-mini

PATH B: $5K-$25K/month
→ Decision: Is context caching enabled?

- YES: Evaluate model selection by task complexity
• Savings: 25-40% ($6K-$10K/month)
• Action: Create routing logic for query complexity
- NO: Enable caching immediately
• Savings: 50-70% ($12K-$17K/month)
• Action: Cache system prompts, repeated documents, common contexts

PATH C: $25K-$100K/month
→ Decision: Do you have rate limits on internal tools?

- YES: Audit for unused features and redundant subscriptions
• Savings: 20-35% ($20K-$35K/month)
• Action: Consolidate tools, negotiate enterprise contracts
- NO: Implement rate limiting and spending alerts
• Savings: 40-60% ($40K-$60K/month)
• Action: Cap per-user limits, separate dev from production

PATH D: Over $100K/month
→ Decision: Are you using rules-based logic where possible?

- YES: Optimize prompt engineering across teams
• Savings: 15-30% ($30K-$60K/month)
• Action: Build prompt libraries, reduce regeneration attempts
- NO: Replace AI with deterministic processes
• Savings: 35-50% ($70K-$100K/month)
• Action: Use regex for classification, if-then for simple routing

SECONDARY OPTIMIZATION: Token Management

After primary path, check document handling:
→ Are you truncating long documents?

- Switch to models with larger context windows (Claude 200K)
- Savings: 20-30% on document processing costs

→ Are prompts unclear causing multiple attempts?

- Implement prompt testing before deployment
- Savings: 40-60% reduction in wasted API calls

FINAL CHECK: Fallback Logic

Do simple queries hit your most expensive model?
→ NO: You're optimized
→ YES: Implement complexity routing

- Simple queries → GPT-4o-mini ($0.15/1M tokens)
- Medium queries → GPT-4o ($2.50/1M tokens)
- Complex queries → GPT-4 ($30/1M tokens)
- Savings: 60-80% on mixed workloads

YOUR OPTIMIZED OUTCOME:

Under $5K: Reduce to $2K-$3K
$5K-$25K: Reduce to $8K-$15K
$25K-$100K: Reduce to $15K-$60K
Over $100K: Reduce to $50K-$75K

One finance team followed this flowchart.
Found $847K in annual waste.
Fixed it in 14 days.

P.S. Want to learn more about AI?

1. Scroll to the top
2. Click "Visit my website"
3. Sign-up for our free newsletter