The 3 Types of AI Consulting (And Which You Actually Need)
Not all AI consulting is the same. Before you hire anyone, understand the three distinct types and which one solves your problem.
Type 1: Strategy Consulting
Strategy consultants help you figure out what to build. They audit your product, identify AI opportunities, and create a roadmap. Deliverables: slide decks, roadmaps, competitive analyses, and recommended architectures.
When you need this: You know AI is important but don't know where to start. You need someone to evaluate which AI features would have the most impact on your product.
When you don't: You already know what to build. You need someone to build it.
Red flag: Strategy consultants who never touch code. If they can't validate their own recommendations with a working prototype, their strategy is theoretical.
Type 2: Implementation Consulting
Implementation consultants build the AI feature for you. They integrate AI into your existing codebase, handle the production engineering, and hand you a working feature. Deliverables: production code, documentation, monitoring, and knowledge transfer.
When you need this: You know what to build but don't have the in-house expertise to build it. Or you have the expertise but not the bandwidth.
When you don't: You have a strong AI/ML team that just needs more time.
This is what most SaaS companies actually need. The bottleneck isn't strategy — it's execution.
Type 3: Staff Augmentation
Staff augmentation provides AI engineers who work embedded in your team, often for 3-6 months. They look and work like employees, using your tools, attending your standups, and following your processes.
When you need this: You need long-term AI capacity but can't hire fast enough. Or you need to build internal AI expertise through knowledge transfer.
When you don't: You need a specific feature shipped in 30 days. Staff augmentation is slow to onboard and optimize for throughput, not outcomes.
What Good AI Consulting Delivers vs What Most Sell
What most sell:
- •A proof of concept that works in a demo but not in production
- •A fancy architecture diagram that doesn't account for real-world constraints
- •"Cutting-edge" solutions using the latest models without considering cost or reliability
- •Vague deliverables: "AI strategy," "model evaluation," "architecture recommendations"
What good consulting delivers:
- •Production code that's committed to your repository, not a separate repo
- •Working in your stack — not a custom framework you can't maintain
- •Measurable outcomes — "38% reduction in support tickets" not "AI-powered support enhancement"
- •Knowledge transfer — your team can maintain, iterate, and extend what was built
- •Monitoring from day one — cost dashboards, quality metrics, alerting
- •Documentation — not of the AI concepts, but of the specific implementation decisions and trade-offs
Red Flags: 7 Signs You're About to Waste Money
1. They can't explain their approach in your terms. If they speak in ML jargon without connecting it to your business metrics, they're optimizing for their expertise, not your outcomes.
2. The timeline is longer than 8 weeks for the first deliverable. Good AI consultants can ship a meaningful first feature in 4-6 weeks. If they need 3 months of "research and evaluation" before building anything, you're paying for their learning curve.
3. They want to build custom infrastructure. Unless your problem is genuinely unique (it probably isn't), off-the-shelf LLMs with good integration are the answer. Custom model training, custom vector databases, and custom frameworks are almost never needed.
4. They don't ask about your existing architecture. AI features don't exist in isolation. If they don't ask about your tech stack, deployment process, monitoring, and team structure in the first meeting, their solution won't fit your reality.
5. The proposal doesn't include production hardening. If the scope only covers "integration" without mentioning fallbacks, monitoring, cost controls, rate limiting, and security, you're buying a demo, not a production feature.
6. They charge by the hour with no scope cap. Time-and-materials billing with open-ended scope is how AI consulting bills balloon. Demand fixed scope with a price cap, or milestone-based billing.
7. No knowledge transfer plan. If the consulting engagement ends with "call us when you need changes," you're being set up for perpetual dependency. Good consultants leave your team capable of maintaining and extending the work.
How to Scope an AI Consulting Engagement
The Scoping Meeting
The first meeting should cover:
- •Your problem (not your solution). What business outcome do you need?
- •Your constraints — budget, timeline, team, tech stack
- •Your data — what data does the AI feature need access to?
- •Your success metrics — how will you measure if this worked?
After this meeting, the consultant should provide a fixed-scope proposal within a week.
The Good Proposal
A good proposal includes:
- •Clear deliverables — specific features, not vague capabilities
- •Fixed timeline — with milestones at 2-week intervals
- •Fixed price — or a price cap with clear scope boundaries
- •Success criteria — measurable outcomes that define "done"
- •Knowledge transfer plan — how your team takes ownership
- •What's NOT included — explicit scope boundaries
The Engagement Structure
We recommend a 4-6 week engagement with these milestones:
Week 1-2: Architecture + foundation. Deliverable: working prototype in your codebase.
Week 3-4: Feature completion + evaluation. Deliverable: feature-complete with automated testing.
Week 5-6: Production hardening + handoff. Deliverable: production-ready with monitoring, documentation, and team training.
Each milestone should be independently valuable. If you stop after week 2, you have a working prototype your team can continue building on.
Fixed Price vs Time-and-Materials
Fixed price (recommended for most engagements):
- •You know the total cost upfront
- •The consultant is incentivized to be efficient
- •Scope is defined clearly
- •Risk: scope changes require renegotiation
Time-and-materials (appropriate for exploration):
- •Flexible scope for undefined problems
- •You pay for actual work done
- •Risk: costs can spiral without active management
- •Only use with a hard budget cap
Our strong recommendation: fixed price with milestone-based payments. Pay 30% at kickoff, 40% at mid-point milestone, 30% at completion. This aligns incentives and protects both parties.
When You Don't Need a Consultant
Not every AI feature requires external help. Build internally if:
- •Your team has production AI experience. Not ML experience — production AI experience. Building models and shipping production features are different skills, but if you have both, you don't need outside help.
- •The feature is a standard integration. Adding a chatbot with a standard API? Adding AI-powered search with a managed service? Your team can follow the documentation.
- •You have the time. If your timeline is 6+ months and you have strong engineers available, internal development builds lasting capability.
- •The feature is experimental. If you're not sure the feature is valuable yet, build a rough version internally first. Hire a consultant to productionize it after you've validated the concept.
Questions to Ask Before Signing
- •"Can you show me a production AI feature you've built in a codebase similar to ours?"
- •"What happens when the LLM provider goes down? How does your implementation handle it?"
- •"What will my team need to know to maintain this after you leave?"
- •"What's your approach to cost control for AI features?"
- •"How do you measure the quality of AI outputs in production?"
- •"What's included in the price, and what would trigger additional costs?"
The answers tell you whether this consultant has shipped AI in production or just built demos.