How to Choose an AI Consultant: What Good Engagement Looks Like
The AI consulting market has exploded. Every IT consultancy, management firm, and freelance developer now offers “AI strategy” and “AI implementation.” The quality range is enormous — from genuinely transformative engagements to six-figure PowerPoint exercises that leave companies worse off than before.
As someone who has been in this field for over ten years, I have seen both sides. I have seen consultants deliver extraordinary value, and I have cleaned up after engagements that destroyed budgets and internal trust in AI. This guide is an honest look at how to evaluate AI consulting, what good looks like, what bad looks like, and how to structure an engagement for success.
Yes, Intellineers is an AI consultancy. I am writing this knowing it will help you evaluate us alongside everyone else. That is the point. If a consultant cannot withstand scrutiny, that tells you something.
Why AI Consulting Is Different from Other Consulting
Traditional management consulting operates on established frameworks. The deliverables are well-understood. AI consulting is fundamentally different in several ways.
Outcomes are uncertain. A management consultant can reasonably guarantee a cost analysis or a market assessment. An AI consultant cannot guarantee that a model will reach a specific accuracy threshold on your data. Anyone who guarantees specific model performance before seeing your data is either lying or does not understand the field.
The gap between advice and execution is enormous. Many consulting firms can produce an AI strategy deck. Far fewer can actually build, deploy, and operate the systems described in that deck. The strategy-execution gap is where most AI consulting engagements fail to deliver value.
Technical depth matters. In traditional consulting, the partner sells the work and junior analysts execute. In AI consulting, the people doing the work need deep technical competence. A junior consultant cannot architect a production ML system any more than a medical student can perform surgery.
The field changes rapidly. Best practices from 18 months ago may be outdated. A consultant whose last hands-on project was in 2023 is working with an outdated mental model of what is possible and what things cost.
What Good AI Consulting Looks Like
Good engagements share consistent characteristics, regardless of the firm.
They Start with the Business Problem, Not the Technology
The first question a good consultant asks is not “what data do you have?” or “have you considered GPT-4?” It is “what business problem are you trying to solve, and what would it be worth to solve it?”
This matters because the answer sometimes reveals that AI is not the right tool. A good consultant will tell you that. A bad one will find a way to justify AI regardless, because that is what they are selling.
Concrete indicator: In the first meeting, a good consultant spends 70% of the time asking about your business and 30% on technology. A bad one inverts this ratio.
They Are Honest About What They Do Not Know
AI is broad. Nobody is an expert in computer vision, NLP, time series forecasting, reinforcement learning, and MLOps simultaneously. A good consultant is clear about their strengths and will either bring in specialists or refer you elsewhere for work outside their competence.
Concrete indicator: Ask “what types of AI projects would you turn down?” If the answer is “none” or “we can handle anything,” that is a red flag.
They Build Your Capability, Not Your Dependency
The best consulting engagements leave you more capable than before. That means knowledge transfer, documentation, training of your team, and architecture that your people can maintain and extend.
Concrete indicator: The proposal includes explicit knowledge transfer milestones — not as an afterthought in the final week, but woven throughout the engagement. At Intellineers, for instance, we structure engagements so that by the end, your team can operate and evolve what we built together. For companies looking to build long-term internal capability, our guide on building internal AI platforms outlines what that architecture looks like.
They Deliver Working Systems, Not Just Recommendations
A strategy document has value, but only if it leads to execution. The most impactful consulting engagements pair strategic thinking with hands-on implementation. A good consultant can go from whiteboard to production code.
Concrete indicator: Ask for examples of systems they have built that are currently running in production. Not prototypes, not demos — production systems that real users depend on daily.
They Scope Realistically and Communicate Proactively
Good consultants underpromise and overdeliver. They flag risks early, communicate setbacks immediately, and adjust scope when reality diverges from the plan — which it always does in AI projects.
Concrete indicator: The proposal includes explicit risk factors and contingency plans. Weekly status updates are specific, not vague. “The model accuracy is at 78% against our 85% target, and here is what we are doing about it” is good communication. “Everything is on track” for 12 weeks followed by “we need three more months” is not.
Seven Red Flags
These patterns consistently predict poor outcomes. If you see more than two, walk away.
1. Guaranteed Results Before Seeing Your Data
“We guarantee 30% cost reduction through AI” — without having seen a single data point. AI performance depends entirely on data quality, volume, and relevance. Any guarantee made before a thorough data assessment is marketing, not consulting.
2. The Team Is a Black Box
You meet impressive partners during the sales process, then discover the actual work is done by junior staff or subcontractors you never vetted. Ask explicitly: who will do the hands-on work? What is their background? Can you meet them before signing?
3. Proprietary Lock-In
The consultant builds on a proprietary platform that only they can maintain and extend. When the engagement ends, you are either stuck paying them indefinitely or facing a costly rebuild. Insist on open standards, documented code, and full IP ownership.
4. No Discovery Phase
A consultant who jumps straight to implementation without a discovery phase is either overconfident or padding the engagement. A 2–4 week discovery phase (data assessment, use case validation, feasibility analysis) is standard for any project above €50,000. It protects both sides.
5. Vague Deliverables
“We will develop an AI strategy” is not a deliverable. “We will deliver a prioritized roadmap of 3–5 use cases, each scored on feasibility, expected ROI, data readiness, and implementation complexity, with detailed implementation plans for the top two” is a deliverable. If the proposal reads like a brochure, ask for specifics.
6. No Mention of Data Quality or Change Management
A proposal that focuses entirely on models and algorithms while ignoring data preparation and organizational adoption is incomplete at best. Data quality and change management are the two factors that most frequently determine whether an AI project succeeds or fails.
7. Unwillingness to Provide References
Every competent consultant has at least a few clients willing to speak to their experience. If a firm cannot or will not provide references, treat that as information. NDA restrictions are legitimate for specific details, but a general reference conversation should always be possible.
Five Questions to Ask Before Signing
These questions cut through marketing and reveal what you are actually getting.
1. “Walk me through a project that failed. What happened and what did you learn?”
Every experienced consultant has failed projects. The answer reveals self-awareness, learning orientation, and honesty. If they claim a perfect track record, they are either inexperienced or dishonest.
2. “What does our data need to look like for this to work?”
This tests whether the consultant has thought about feasibility rather than just desirability. A good answer is specific: “We need at least 12 months of historical transaction data, with customer IDs consistently linked across your CRM and ERP, and fewer than 10% missing values in the key fields.” A bad answer is: “We will figure out the data situation during the project.”
3. “What happens after you leave?”
This reveals whether the engagement is designed for lasting impact or ongoing dependency. Look for: documentation, knowledge transfer sessions, training for your team, and architecture that your internal staff (or a different vendor) can maintain.
4. “How do you handle it when the approach is not working?”
AI projects often require pivots. The first model architecture might underperform. The data might not support the intended use case. A good consultant has a structured approach to handling this: clear decision points, alternative approaches prepared, and honest escalation to stakeholders.
5. “Can you show me a working demo of something similar?”
Not a slide deck. Not a video. A working system that you can interact with. This immediately separates firms that build real things from firms that produce presentations about building things.
Realistic Costs and Timelines
Understanding market rates helps you evaluate proposals and spot both lowball bids (likely to result in poor delivery) and inflated quotes.
Daily Rates (European Market, 2025–2026)
- Freelance AI/ML engineer: €800–€1,500/day
- Specialized AI consultancy (boutique): €1,200–€2,200/day
- Big Four / large management consultancy: €2,000–€3,500/day
- Academic/research partnerships: €600–€1,000/day (slower, more theoretical)
The spread is real. What you get for the money varies enormously. A boutique firm at €1,500/day with senior practitioners doing the hands-on work often delivers more value than a large firm at €3,000/day where junior staff execute while partners are on the next pitch.
Typical Engagement Structures
Discovery / Assessment (2–4 weeks): €15,000–€40,000. Deliverable: prioritized use case roadmap, data assessment, feasibility analysis, cost estimates for implementation. This is the single highest-ROI investment in the entire process. A solid AI readiness assessment at this stage prevents six-figure mistakes later.
Proof of Value / Pilot (6–12 weeks): €50,000–€150,000. Deliverable: working prototype on real data, validated against defined KPIs, with a production deployment plan. This should be structured as a fixed-scope engagement with clear success criteria.
Production Implementation (8–16 weeks): €60,000–€200,000. Deliverable: production-ready system, integrated with existing infrastructure, with monitoring, documentation, and knowledge transfer. Often a mix of time-and-materials for complex integration work and fixed deliverables for specific milestones.
Ongoing Support / Retainer: €3,000–€10,000/month. Monitoring, model maintenance, periodic retraining, ad-hoc support. Should decrease over time as your internal team builds capability.
Timeline Reality Check
Anyone promising a production AI system in 4 weeks is either working on a trivial problem, cutting corners on production readiness, or lying. Realistic timelines:
- From idea to validated strategy: 4–8 weeks
- From strategy to working pilot: 6–12 weeks
- From pilot to production: 6–12 weeks
- From first production system to scaled platform: 6–18 months
Total time from “we want to explore AI” to “we have a system in production generating measurable ROI”: 6–12 months. This aligns with what we describe in our AI strategy guide.
How to Structure the Engagement for Success
Start with a Paid Discovery Phase
Never commit to a large implementation based on a sales pitch. Pay for a proper discovery phase first. This gives you a deliverable (the assessment and roadmap) that has standalone value regardless of whether you continue with the same firm. It also tests the consultant’s working style, communication, and competence at low risk.
Define Success Metrics Before Starting
“Successful AI implementation” is not a measurable outcome. “Reduce average claim processing time from 45 minutes to 15 minutes for 80% of standard claims within 6 months” is. Define metrics that both sides agree on before work begins.
Insist on Regular Demos
Biweekly demos of working software — not slide updates. This is the single best mechanism for keeping projects on track. If the consultant cannot show progress in working software every two weeks, something is wrong.
Plan for Knowledge Transfer from Day One
Do not treat knowledge transfer as a final-phase activity. Your team should be involved throughout: pair programming, architecture reviews, joint testing, and shared documentation. The engagement should feel like a collaboration, not a handoff.
Retain IP Ownership
All code, models, documentation, and training data created during the engagement should belong to you. This is non-negotiable. Any consultant who insists on retaining IP ownership of work you paid for is building a dependency, not a solution.
A Note on Company Size and Approach
The right consulting approach depends heavily on your company size and existing capabilities.
Small companies (under 100 employees): You likely do not need a full-scale AI consultancy. Start with a focused assessment and a single use case. A senior freelancer or small boutique firm is often the best fit. Budget: €30,000–€80,000 for a first initiative.
Mid-sized companies (100–1,000 employees): The sweet spot for specialized AI consultancies. Large enough to have meaningful data and impactful use cases, but without massive in-house AI teams. Budget: €100,000–€300,000 for a first year including strategy and first production system.
Large enterprises (1,000+ employees): May need both strategic consulting (for roadmap and governance) and specialized implementation partners (for building). Consider building an internal AI team alongside consulting engagements. Budget: €300,000–€1,000,000+ for a comprehensive AI program.
Conclusion
Choosing the right AI consultant is one of the highest-leverage decisions in your AI journey. A good engagement accelerates your capabilities, delivers measurable value, and leaves your organization stronger. A bad one burns budget, destroys internal trust in AI, and sets you back by years.
Do your diligence. Ask hard questions. Demand specifics. Start small with a discovery phase. Judge consultants by what they build, not what they present. And remember: the best consulting engagement is the one that makes itself unnecessary.
Intellineers is a Hamburg-based AI consultancy specializing in strategy and implementation for mid-sized companies. We are happy to answer the hard questions — about our work and about AI in general. Reach out for a conversation.