Working With Technical People

As a business leader, you probably don't build AI yourself. You work with people who do — internal teams, vendors, consultants, and contractors.

Getting value from AI means getting value from these relationships. This chapter covers how.

Understanding AI Roles

Not all technical people do the same thing. Knowing the roles helps you hire, staff, and communicate.

Data Scientists / ML Engineers

What they do: Build and train models. The "AI" part of AI.

What they need from you: Clear problem definition, access to data, patience for experimentation.

What to watch for: Tendency to optimize for interesting problems, not business value. May underestimate production requirements.

Data Engineers

What they do: Build pipelines that move and transform data. The plumbing.

What they need from you: Understanding that their work enables everything else. Time to do it right.

What to watch for: Often underestimated and under-resourced. Data engineering is frequently the bottleneck.

ML/MLOps Engineers

What they do: Deploy models to production and keep them running. The bridge from experiment to operation.

What they need from you: Clear production requirements. Infrastructure investment. Operational support.

What to watch for: This role is critical but often missing or understaffed. Production is harder than prototyping.

Product Managers (AI/ML)

What they do: Connect business needs to technical work. Define requirements. Prioritize.

What they need from you: Business context, stakeholder access, decision authority.

What to watch for: Good AI product managers are rare. Generic PMs may struggle with ML's iteration patterns and uncertainty.

AI/ML Architects

What they do: Design overall system approach. Make technology choices. Ensure scalability and maintainability.

What they need from you: Long-term vision. Authority to make technical decisions.

What to watch for: May over-engineer or under-deliver. Balance vision with pragmatism.

Team Structures

How should AI capabilities be organized?

Centralized Team

All AI people in one group, serving the whole organization.

Pros:

• Concentrated expertise
• Consistent approaches
• Easier to hire and develop talent
• Shared resources

Cons:

• Can feel distant from business units
• May prioritize interesting over important
• Potential bottleneck
• Business units lack ownership

Best for: Early-stage AI adoption, smaller organizations, when building capability.

Embedded Teams

AI people sit within business units.

Pros:

• Close to business problems
• Strong ownership
• Fast iteration
• Deep domain knowledge

Cons:

• Scattered expertise
• Inconsistent approaches
• Harder to share learning
• May lack critical mass

Best for: Organizations with clear AI-heavy business units, mature AI adoption.

Hub and Spoke

Central team for standards, tools, and specialists. Embedded people for execution.

Pros:

• Best of both approaches
• Shared standards with local execution
• Career paths for AI talent
• Scalable model

Cons:

• Coordination overhead
• Potential conflicts
• More complex to manage

Best for: Larger organizations with multiple AI use cases.

Hiring AI Talent

The Market Reality

AI talent is expensive and competitive. You're competing with tech giants, well-funded startups, and other enterprises. Accept that you may not get the "best" — focus on getting "good enough."

What to Hire

Start with:

• Strong data engineer (or team) — you need data infrastructure first
• ML engineer with production experience — not just modeling, but deployment
• Ideally, someone with your domain knowledge

Add when needed:

• Specialized data scientists for complex modeling
• MLOps for scaling
• AI product management for prioritization

Consider outsourcing:

• One-time projects
• Specialized skills you don't need full-time
• When learning before committing to hires

Hiring Tips

Don't over-credential: PhD not required for most work. Experience and pragmatism often matter more.

Test on real problems: Give candidates problems similar to what they'll actually work on.

Check for communication: Can they explain complex concepts to non-technical people? This matters.

Assess interest in business outcomes: Some AI people only care about technology. You need people who care about results.

Reference check carefully: AI talent is hyped. Verify claims.

Retention

AI talent is mobile. Retention matters as much as hiring.

What AI people want:

• Interesting problems
• Modern tools and infrastructure
• Access to data
• Supportive management
• Competitive compensation
• Learning and growth
• Visibility for their work

What drives them away:

• Bureaucracy blocking their work
• Outdated infrastructure
• Problems that aren't important
• Being ignored or undervalued
• Below-market compensation
• No growth path

Working With AI Vendors

Most companies will work with AI vendors. Here's how to get value.

Setting Up the Relationship

Clear scope: Define exactly what you expect. Ambiguity becomes argument.

Success metrics: Agree upfront how success will be measured.

Communication cadence: Regular check-ins. Don't wait for problems to surface.

Escalation path: Who resolves issues? How?

Change process: How are changes requested, evaluated, and implemented?

Getting Value

Be a good customer:

• Provide data and access promptly
• Make decisions when needed
• Give honest feedback
• Pay on time
• Treat vendors as partners, not adversaries

Demand accountability:

• Track against agreed metrics
• Address misses directly
• Require explanations for problems
• Hold to timelines (while being reasonable)

Maintain leverage:

• Don't let critical capabilities depend on one vendor
• Plan for vendor failure or exit
• Keep contracts reasonable in length
• Own your data

Common Vendor Problems

Overselling then underdelivering: Sales promises what delivery can't provide.

Fix: Pin down commitments in writing. Pilot before commitment. Talk to references.

Disappearing after sale: Attention drops once contract signed.

Fix: Build check-ins into contract. Require named contacts. Review regularly.

Scope creep from vendor: "You also need to buy this..."

Fix: Stick to original scope. Evaluate add-ons separately.

Your scope creep: "Can you also do this..."

Fix: Formal change requests. Expect cost and timeline impacts.

Technical drift: Solution evolves away from your needs.

Fix: Stay engaged in product roadmap. Provide feedback. Have alternatives.

Working With Consultants

Consultants can accelerate AI capability. Or waste your money.

When Consultants Make Sense

• You need specialized skills for a limited time
• You need to move fast and don't have internal capacity
• You want to learn while building
• You need external credibility for internal initiatives
• You're evaluating vendors and need objective assessment

When Consultants Don't Make Sense

• You're not ready to implement (they'll just produce decks)
• You don't have someone internal to receive the knowledge
• You expect them to make strategic decisions for you
• You want to outsource ownership

Getting Value from Consultants

Clear deliverables: What exactly will they produce?

Knowledge transfer: Insist on documentation and training. If learning stays with them, value leaves with them.

Internal ownership: Someone internal must own the outcome. Consultants support; they don't replace ownership.

Verification: Check their work. They can be wrong.

Phase gates: Don't commit to the whole engagement upfront. Gate payments on deliverables.

Managing Remote and Offshore Teams

AI work is often done by distributed teams.

Making It Work

Clear communication: Over-communicate. Document everything. Assume less shared context.

Overlap hours: Find some hours for real-time collaboration. All-async is hard for complex work.

Tools: Invest in collaboration tools. Video, chat, shared documentation.

Cultural awareness: Different cultures have different communication styles. Learn them.

Relationships: Build personal connections. Video on. Face time when possible.

Common Problems

Quality issues: Distance makes quality harder to verify. Build in reviews.

Communication gaps: Misunderstandings compound. Check understanding.

Timezone pain: Someone always has inconvenient hours. Share the burden fairly.

Knowledge silos: Remote teams can become isolated. Build bridges.

Leadership Behaviors

As the business leader, your behavior shapes AI success.

Helpful Behaviors

• Ask questions (show interest without micromanaging)
• Remove blockers (clear organizational obstacles)
• Make decisions (don't let things stall for lack of decision)
• Provide context (help technical teams understand business priorities)
• Shield from distractions (protect focus)
• Celebrate successes (recognize good work)
• Accept uncertainty (AI involves experimentation)

Harmful Behaviors

• Demanding certainty (AI is probabilistic; timelines are estimates)
• Changing priorities constantly (thrash destroys productivity)
• Bypassing process (direct requests to technical staff create chaos)
• Ignoring problems (issues raised need to be addressed)
• Taking credit, assigning blame (destroys trust)
• Treating AI as magic (unrealistic expectations set everyone up for failure)

Building Trust

Trust enables everything:

• Do what you say you'll do
• Give honest feedback
• Protect people who raise problems
• Admit when you're wrong
• Share credit broadly
• Back your team publicly
• Address conflicts directly

Trust takes time to build and moments to destroy. Invest carefully.