How to Choose an AI Consultant: 10 Questions to Ask Before You Hire

A Houston-area manufacturer spent $85,000 on two AI consultants before they got any real value.

The first consultant was a large firm that delivered a 90-page “AI Transformation Roadmap.” Beautiful document. Executive summary, technology analysis, competitive benchmarking, risk assessment. It sat in a drawer. Nobody at the company knew what to do with it. The roadmap talked about “enterprise AI maturity frameworks” and “data mesh architectures” for a company with 120 employees and one IT person.

The second consultant was a development shop that built a prototype. A custom machine learning model for quality inspection that worked in their demo environment. But it couldn’t connect to the manufacturer’s actual camera systems, ran on hardware they’d need to buy separately, and required a data scientist to maintain. Phase 2 never happened because Phase 1 already burned through the budget.

The third consultant started with a different question: “What’s the most expensive problem in your operation right now?” The answer was unplanned equipment downtime. They installed sensors on three critical machines, connected them to an off-the-shelf predictive maintenance platform, and had it running in six weeks. Total cost: $22,000. Savings in the first year: $140,000 in reduced emergency repairs and avoided production losses.

Same company. Same problem space. Three very different outcomes. The difference was the consultant’s approach, not the technology.

This is not an unusual story. We hear versions of it regularly from businesses in Houston and Denver who came to us after a prior engagement left them with either a document they couldn’t act on or a system they couldn’t maintain. The goal of this post is to give you the questions that would have changed those outcomes earlier.

Red Flags That Should Stop You

These patterns show up often enough that they’re worth naming.

They lead with technology, not problems. If the first meeting is about their AI platform, their proprietary algorithm, or their technology stack, be cautious. Good agencies start by understanding your operations, your costs, your bottlenecks. The technology is a means, not the point. If they’re selling a hammer, every problem looks like a nail.

They can’t show relevant case studies. “We’ve done AI projects for Fortune 500 companies” doesn’t help if you’re a 50-person service business. Ask for examples with companies your size, in your industry or an adjacent one. If they don’t have them, you’re paying for their learning curve.

They quote a fixed price before understanding the problem. AI projects have unknowns. Data quality issues, integration complexity, user adoption challenges. An agency that quotes $50,000 for an “AI solution” before spending time understanding your data, your systems, and your team is either padding heavily for risk or underestimating the work. Either way, the number is fiction.

They promise specific ROI before discovery. “Our AI solution delivers 10x ROI” is a marketing claim, not an engineering estimate. Responsible agencies estimate ROI after they understand your specific situation: your data quality, your current costs, your team’s capacity to adopt new tools. Anyone guaranteeing outcomes before they’ve looked under the hood is selling hope, not service.

They want to build custom when off-the-shelf works. If your need is common (document processing, customer service chatbot, lead scoring, inventory forecasting), proven tools already exist. An agency that proposes custom development for a solved problem is either unfamiliar with the market or optimizing for a bigger project fee.

They don’t talk about change management. Technology implementation is half the work. Getting your team to actually use it is the other half. If the agency’s proposal doesn’t include training, onboarding, and adoption planning, they’re delivering software, not a solution. You’ll end up with a tool nobody uses.

What to Look For Instead

Problem-first thinking. The agency should spend more time in the first meeting asking about your business than talking about their capabilities. What does your day look like? Where do you lose time? What errors keep happening? What information do you wish you had? These questions lead to solutions that actually fit.

Right-sized solutions. For a 20-person company, the right AI solution is usually a commercial tool configured for your workflow, possibly connected to your existing systems with some automation. Not a custom-built machine learning model. Not an enterprise platform designed for 10,000 users. Right-sized means appropriate complexity, cost, and maintenance burden for your team.

Implementation experience, not just strategy. Strategy documents have value, but only if someone executes on them. Ask: who builds it? Who configures it? Who trains the team? Who supports it after launch? If the answer is “that’s a separate engagement,” you’re buying a plan, not a result. The best agencies handle implementation through to adoption.

Clear scope with realistic timelines. Good agencies break projects into phases with deliverables at each stage. Phase 1 might be discovery and data assessment (2-3 weeks). Phase 2 is implementation of the first use case (4-6 weeks). Phase 3 is optimization and expansion. Each phase has a defined cost, clear deliverables, and decision points where you can continue, pivot, or stop.

References you can actually call. Not just logos on a website. Actual people at actual businesses who will tell you what it was like to work with the agency. What went well. What was difficult. Whether they’d hire them again. Two reference calls tell you more than any sales presentation.

10 Questions to Ask an AI Consultant Before You Hire

These questions separate consultants who have done this work from consultants who have sold it. Use them in your first meeting.

1. “Walk me through a project that failed or went sideways.” Every consultant who has done real work has one. How they talk about it reveals their transparency and their problem-solving instincts. A claim that every project succeeded is either very new experience or not being straight with you.

2. “What will our team need to do during implementation?” If the answer is “nothing, we handle everything,” stop there. Successful AI implementations require your team’s input on processes, data, edge cases, and user acceptance. A detailed answer — “we’ll need two hours a week from your ops manager, access to your accounting system, and someone to test output for the first two weeks” — is a good sign.

3. “What happens after you leave?” Who maintains the system? Who handles updates when your processes change? Is there a support contract? Can another consultant take over if the relationship ends? Vendor lock-in, where you’re permanently dependent on one firm for a system you own, is a real risk worth probing.

4. “How do you measure success, and when do you define that?” Good consultants define metrics before they start building. Specific outcomes tied to your business. “Reduce invoice processing time from 8 minutes to 2 minutes.” Not “improve operational efficiency.” If they can’t define how they’ll measure their own impact before starting, they have no way to prove their value after.

5. “What data do you need from us, and what happens if it’s not clean?” Every business thinks their data is worse than it is. Some businesses are right. How a consultant handles that question reveals their experience. The naive answer is “we’ll figure it out.” The experienced answer includes data assessment, cleaning scope, and what that means for the timeline.

6. “Can you show me the approach working with similar data before we commit?” Ask for a proof of concept with a subset of your actual data. This removes the theoretical and shows you what real output looks like. Consultants who are confident in their work welcome this. Consultants selling something they haven’t built yet tend to avoid it.

7. “Who specifically will be working on our project?” Sales team and delivery team are often different people. The person who impresses you in the meeting may not be the person writing your code or training your model. Get names. Ask about their experience directly. For smaller engagements, this is often the most important question.

8. “Can you give me two references I can call — not companies, but people?” Logos on a website tell you nothing. A real conversation with a former client tells you what the work was actually like: what went well, what didn’t, whether they’d hire the firm again. Two reference calls give you more signal than any presentation.

9. “What’s your discovery process before you propose a solution?” A consultant who quotes a price before spending real time understanding your operations is either padding heavily for unknown risk or underestimating the work. Both cost you. Good discovery takes time. Ask what it looks like.

10. “Have you worked with companies at our scale, in our type of business?” A consultant whose case studies are all Fortune 500 manufacturing companies is not the right fit for a 30-person Denver service firm. Relevant experience at the right scale matters more than total years in AI. Ask for specific comparable examples.

What Pricing Transparency Actually Looks Like

AI consultants charge in several models, and the right one depends on the work.

Fixed-price projects work for well-defined scope. A chatbot configured and deployed, an invoice processing system set up, an automation workflow built — these have clear deliverables. Expect $5,000-$30,000 for small business implementations.

Time-and-materials works for exploratory or complex projects where scope can’t be fully defined upfront. Custom integrations, data projects with unknown complexity, phased implementations. Rates range from $150-$300/hour for qualified AI consultants. Set monthly caps and clear review checkpoints before you agree to any T&M engagement.

Retainer models cover ongoing support, optimization, and new projects once systems are running. Monthly retainers for small businesses typically run $2,000-$8,000 depending on scope.

Value-based pricing ties fees to documented outcomes — a percentage of measured savings, or a success fee on top of a base rate. This aligns incentives but requires both parties to agree on metrics upfront and measure them honestly.

What to watch for: any pricing that’s vague about what you’re getting. A “custom AI model” might cost $50,000 and be worth every dollar. Or it might be a pre-trained model with $500 in configuration wrapped in custom-project billing. Enough understanding to evaluate what you’re paying for is not optional — it’s the only protection you have against the second outcome.

What Good Discovery Looks Like

Before a consultant proposes anything, they should spend meaningful time understanding your business. Not an hour. Several hours, possibly spread across multiple conversations or a structured discovery session.

Good discovery covers: what your current processes look like, where time and money are being lost, what data you have and how clean it is, what systems you already use, what your team can realistically support after the engagement ends, and what success means to your specific situation.

The output of good discovery is a proposal that mentions your actual systems by name, quantifies the problem in your terms, and explains why this particular solution fits your situation — not a generic AI solution.

If a consultant proposes a solution without going through this, they’re selling you something they designed before they understood your problem. That’s how you end up with a 90-page roadmap that sits in a drawer.

Starting small protects you. The manufacturer’s third consultant worked because they started with one problem and a short timeline. They didn’t take $85,000 upfront. They took $22,000 to prove one thing worked.

That approach reveals several things early: how the consultant communicates, whether they deliver on time, whether the technology actually performs on your real data (not demo data), and whether your team can actually adopt what gets built. You learn all of that on a $22,000 pilot, not an $85,000 full engagement.

When You Don’t Need a Consultant

Not every AI project requires outside help.

If your need is a standard tool implementation — adding an AI chatbot, setting up document processing, configuring email automation — your team may handle it with vendor documentation and support. Many modern AI tools are designed for business users, not engineers.

If you have a developer on staff, many AI integrations can be built in-house using APIs and workflow automation platforms. The development work is approachable for someone comfortable with code.

Consultants add the most value when you don’t know what’s possible, when implementation involves connecting multiple systems, when the project requires AI expertise your team doesn’t have, or when the cost of getting it wrong is high.

Being honest about where you are before engaging a consultant saves time on both sides. If you’re still at the exploring stage, AI training may be the right first step rather than a consulting engagement. If you know what you want but lack the hands to build it, you need implementation help. If you’re not sure what AI can do in your specific situation, you need a consultant who will tell you the truth — including when AI is not the answer.

If you have a problem you think AI might solve and want a straight answer about whether that’s true, call us at (346) 389-5215 or take our AI Readiness Compass to see where you actually stand.

AI Agents for Business Automation: Beyond Simple Chatbots — What AI agents actually do and where they fit.
Building Your First AI Agent: A Non-Technical Guide — What the process looks like from problem to working system.