How to pick an AI vendor without getting taken.
8-criteria framework + 5 red flags + 6 sales-meeting questions, from an agency that competes for these contracts. We win some, we lose some, here's what separates the legitimate vendors from the rest.
By BKND Development · Updated April 28, 2026 · ~10 min read
8 criteria for evaluating AI vendors.
A vendor that passes all 8 is rarely a bad bet. A vendor that fails 3+ is rarely a good one.
1. Shipped production AI systems (not slide decks)
Ask for 3 specific AI systems they've architected + shipped to clients. Real systems with real users. Not 'we did an AI workshop for [Fortune 500].' Real shipped systems.
2. Vendor-agnostic recommendations
Do they recommend the same model + platform regardless of use case? Or do they actually evaluate fit per workflow? Vendor lock-in risk is real. Ask: 'When would you NOT recommend Claude/GPT/your favorite tool?'
3. Fixed-fee pricing on at least one engagement type
Reputable practitioners offer fixed-fee on assessments, pilots, or working days. Hourly-only billing creates incentive misalignment. Worth asking about.
4. Written deliverables included in scope
Pure-verbal-advice consulting is rarely worth $5K+. Look for written PDF deliverables (roadmap, scope doc, comparison matrix) included in the engagement scope. Tangible outputs you can hand to anyone.
5. Clear post-launch tuning period included
AI workflows need 4-8 weeks of post-launch tuning to hit production reliability. Vendors that ship at day 30 and walk away leave you holding broken systems by day 90. Make sure tuning is in scope.
6. Written SLAs on response time + bug fixes
Production AI breaks. When it does, how fast does the vendor respond? Get this in writing. Real practitioners offer 24-48 hr response on critical issues. 'We'll get to it' is a red flag.
7. References from clients in your size range
Big-firm consultants will reference Fortune 500 work. SMB practitioners should reference SMB clients. Ask for 3 references from clients within 50% of your revenue size. Call them.
8. Code ownership transparency
Custom AI builds should produce code that lives in YOUR repository. Some vendors keep code in their own infrastructure (lock-in). Ask explicitly: 'Where does the code live? Who owns it? What happens if we end the engagement?'
5 red flags that mean "walk away."
Any one of these alone might not disqualify a vendor. Two or more together usually does.
Same recommendations regardless of business
If they pitch the same 5 AI workflows (chatbot, content, sales enablement, etc.) to every client, they're not consulting — they're selling a template. Real AI consulting is bespoke.
Vendor commission relationships hidden
Some 'consultants' get paid by AI vendors for referrals. Ask directly: 'Do you receive any compensation from AI vendors you recommend?' If they hesitate, walk away.
Pure strategy with no implementation
$50K+ for slide decks without anyone shipping a working system is rarely worth it. Real practitioners offer implementation OR partner you with implementers OR explicitly tell you they don't implement.
Fortune 500 case studies as the only credibility
If their only references are 'we worked with [big company name]' but no actual shipped-system case studies — they're consultants, not builders. SMB AI deployment requires builders who ship.
No post-launch tuning in scope
AI ships at 60% reliability and tunes to 95% over 4-8 weeks. Vendors that ship at 60% and walk away leave you with broken systems. Reputable shops bake the tuning weeks into engagement scope.
6 questions to ask in any sales meeting.
Bring this list to your discovery calls. The answers separate vendors faster than any pitch deck.
1. "What's the actual deliverable from this engagement?"
Why ask: Forces vendor to articulate what you'll receive. Vague answers = vague work.
2. "Can I see a sample deliverable from a similar past engagement (redacted)?"
Why ask: Real shops have these ready in 24-48 hrs. Slow response = unprepared.
3. "What's the post-launch tuning period and is it included?"
Why ask: Reasonable answer: 4-8 weeks, included. 'We'll see' = walk away.
4. "Where does the code live and who owns it?"
Why ask: Custom AI code should live in YOUR repo. Vendor-hosted = lock-in risk.
5. "Walk me through 1 AI system you've shipped to a client similar to me"
Why ask: Real practitioners describe specific architectures + outcomes. Vague 'we worked with...' = no actual shipped work.
6. "What's your fee structure if I want to add scope mid-engagement?"
Why ask: Reputable shops have written change-order processes. 'We'll figure it out' = scope-creep extraction risk.
Frequently asked questions
How do I know if an AI vendor is legitimate?+
Three signals: (1) Specific shipped-system case studies they can describe in detail. (2) Fixed-fee engagements available (not just hourly). (3) Code ownership transparency — they're explicit about where code lives and that you own it. Vendors that fail any of these are higher risk. Vendors that pass all three are usually safe to engage with.
What should I budget for AI consulting + implementation?+
Independent practitioner: $1,500-$5,000 assessment + $5K-$50K implementation. Boutique consultancy: $10K-$50K per engagement. Big-firm consultancy: $200K-$2M+. For most $1M-$50M revenue businesses, independent practitioner tier is the right fit. Big-firm consulting is usually overkill at SMB scale.
Should I use a generalist consultant or AI specialist?+
AI specialist for AI work. The technology moves fast (frontier model APIs change quarterly, voice infrastructure is evolving, agentic patterns are still maturing). Generalists fall behind. Look for AI-focused practitioners who've shipped AI systems in the last 6 months specifically — not consultants who 'also do AI' as one of 12 services.
What if the vendor wants me to sign a long-term contract?+
Walk away from anything longer than 6 months for a first engagement. Reputable AI vendors offer fixed-fee assessments + project-based pilots that let you evaluate fit before committing to retainers. 12-24 month contracts at the start of an engagement = extraction risk.
Should I get multiple bids?+
Yes — at least 3 for any project over $20K. The bid variance will tell you a lot. Bids ranging from $15K to $80K for the same scope = wildly different conceptions of what they're building. Walk away from the highest and lowest extremes; pick from the 2-3 middle bids based on track record + chemistry.
What's the difference between AI vendors and AI consultants?+
AI vendors sell you platforms (Zapier, Bardeen, Lindy, etc.). AI consultants build custom systems for you. Some shops do both. The right answer depends on your workflow needs — most operations end up using both vendors (for simple connections) and custom builds (for complex AI workflows). Read /ai/custom-ai-agents-vs-zapier for the framework.
Are big consulting firms (McKinsey, Deloitte) worth it for AI?+
Rarely for SMBs. Big-firm consulting works at Fortune 500 scale where: (1) board needs branded validation, (2) you have $1M+ AI strategy budget, (3) you have parallel workstreams across 5+ business units. For $1M-$50M revenue businesses, big-firm consulting is dramatically over-credentialed for the actual work + 10-20x more expensive than independent practitioners delivering equivalent or better implementation.
How do I evaluate AI vendor pitches in a sales meeting?+
Ask the 6 questions in this article. Listen for specifics, not buzzwords. Real practitioners give concrete answers ('we'd architect X for your CRM, Y for the voice piece, Z for the integration'). BS practitioners give vague answers ('we'd do an AI transformation focused on...'). Specificity = capability.
What's BKND's positioning vs other AI vendors?+
We're independent practitioners (3 people running a 25-client agency). Fixed-fee on every engagement. Written deliverables included. Code lives in your repo. We've shipped 5,000+ pages of client AI-generated code, 12+ voice agent systems, dozens of multi-workflow implementations. Pricing tiers transparent at /ai-consultant. We compete with boutique consultancies and lose to big-firm consultants when board validation matters more than working systems.
How do I get started?+
Book the AI Readiness Assessment ($1,500). Two-hour session + written 48-hour roadmap. The cheapest way to evaluate whether we're the right fit for your business — and to establish a benchmark against any other vendors you're evaluating.
Want to evaluate BKND against the criteria above?
Book a 30-min intro call. We'll walk through the 8 criteria together so you can compare us against any other vendors you're evaluating.