How to Evaluate an AI Agency (Before Wasting 5 Lakhs)

Five criteria that separate real AI agencies from pitch-deck factories

We are an AI studio. We know how the industry works from the inside. And honestly, most AI agencies are repackaging basic SaaS tools with a 3-5x markup and a deck full of buzzwords. Here is how to find the ones that are not.

Criterion 1: They can show you a live system, not a presentation.

Ask this in your first call: "Can you show me a system you built that is running right now?" Not a case study PDF. Not a recorded demo from 6 months ago. A live system. Log in, show me the dashboard, show me data flowing, show me what happens when a new lead comes in.

At Vikrama, we show our own systems. Our outreach pipeline running on n8n and Smartlead. Our client dashboards with real data. Our Dizios product with actual users. If an agency cannot show you something running live, they have not built anything worth paying for.

Criterion 2: Outcome-based pricing, not hours.

An agency that charges ₹X per hour for "AI consulting" is selling you time, not results. Good AI agencies tie pricing to outcomes: cost per qualified lead, revenue increase percentage, time saved on specific processes, or project-based fees for defined deliverables.

Hourly billing creates a perverse incentive. The longer the project takes, the more the agency earns. Outcome-based pricing aligns incentives: the agency earns more when you succeed. Not every engagement can be purely outcome-based, but the agency should be willing to tie at least part of their compensation to results.

Criterion 3: Vertical-specific proof.

"We do AI for everyone" means "we have not mastered AI for anyone." Ask for proof in your specific vertical. If you are D2C, show me a D2C client. If you are real estate, show me a real estate project. Industry context matters enormously in AI implementation.

We work with D2C brands, real estate developers, and manufacturing companies. That is it. We say no to industries we do not understand because AI implementation without domain knowledge produces generic systems that underperform. Read our AI systems guide to see how vertical expertise shapes implementation.

Criterion 4: You own everything. Data, code, systems.

This is non-negotiable. After the engagement, you should own: all source code, all data, all automations, all accounts, and all documentation. If the agency has built something only they can access or modify, you are not their client. You are their hostage.

Ask specifically: "If we end this engagement tomorrow, what do we keep?" The right answer is "everything." If they hesitate, if they talk about "proprietary platforms" or "managed environments" that only they can access, walk away.

Criterion 5: They explain AI in plain language.

If an agency cannot explain what their AI does in terms a non-technical founder understands, one of two things is true: they do not understand it themselves, or they are hiding simplicity behind complexity to justify their fee.

"We use Claude API to analyse incoming leads based on 5 criteria and score them 1-100. Leads above 70 get a WhatsApp message within 5 minutes." That is a clear explanation. "Our proprietary AI engine processes multi-dimensional data through advanced machine learning algorithms." That is a smoke screen.

Red flags vs green flags: the checklist

Red flag	Green flag
Guaranteed ROI ("we guarantee 300% return")	Honest ranges based on past results ("clients typically see 20-40% improvement in X")
NDA required before seeing any work	Public case studies and portfolio
"Proprietary AI" for common tasks	Names specific tools: Claude, n8n, HubSpot, Shopify
Only shows slide decks	Demos live systems on the call
Hourly billing for ongoing work	Project-based or outcome-tied pricing
Vague deliverables ("AI strategy")	Specific deliverables ("5 automated workflows, 1 dashboard, training docs")
"We handle everything, you just sit back"	"We build it, train your team, then you own it"
No client references	Offers to connect you with past clients
Refuses to explain technical approach	Walks you through architecture in plain language
Claims AI replaces your entire team	Explains what AI handles vs what humans still do

Count the flags. If you see 3+ red flags, disengage. If you see 5+ green flags, you are probably talking to a legitimate team.

7 questions to ask (and what good vs bad answers look like)

Question 1: "What specific AI models and tools do you use?"

Good answer: "We primarily use Claude for text analysis, n8n for automation, PostgreSQL with pgvector for data, and HubSpot for CRM. For ad optimisation, we rely on Meta Advantage+ and Google PMax."

Bad answer: "We use our proprietary AI platform."

Question 2: "Can you show me a system running right now for another client?"

Good answer: "Yes, let me share my screen. Here is the dashboard for [client]. You can see leads coming in, the automation scoring them, and the WhatsApp sequences going out."

Bad answer: "Due to NDAs, we cannot show client work. But here is our case study deck."

Question 3: "What happens to the systems if we stop working together?"

Good answer: "You keep everything. Code is in your GitHub, automations are in your n8n instance, accounts are in your name. We will do a handover session to make sure your team can maintain it."

Bad answer: "The systems are built on our platform, so they would need to be migrated." (Translation: you are locked in.)

Question 4: "What does the first month look like?"

Good answer: "Week 1: audit your current stack and define goals. Week 2-3: build the first 2-3 automated workflows. Week 4: testing, training your team, and measuring baseline metrics. You will see the first results within 30 days."

Bad answer: "Month 1 is discovery and strategy. We will deliver a roadmap by end of month." (Translation: you pay ₹1-2L for a document.)

Question 5: "What does NOT work well with AI right now?"

Good answer: "AI is bad at complex creative decisions, nuanced brand voice, and anything requiring deep empathy. It is also unreliable for tasks requiring 100% accuracy without human review. We build human checkpoints into every system."

Bad answer: "AI can handle pretty much everything. It just needs the right implementation." (Translation: they are overselling.)

Question 6: "How do you price this?"

Good answer: "For implementation, it is project-based: ₹X for defined deliverables. For ongoing management, it is ₹Y per month tied to specific outcomes. Here is how past clients' costs scaled."

Bad answer: "It depends. We will scope it after the discovery phase." (Translation: you will not know the cost until you have already spent money.)

Question 7: "Can I talk to a current or past client?"

Good answer: "Absolutely. I will connect you with [name] who runs a [similar business]. They can share their experience."

Bad answer: "Our clients prefer privacy." (Translation: they either do not have happy clients, or do not have clients.)

Pricing models: what to expect in 2026 India

Model	Typical range	Best for	Watch out for
Project-based	₹3-20L (one-time)	Defined builds (automation system, dashboard, AI feature)	Scope creep. Define deliverables precisely.
Monthly retainer	₹1-3L/month	Ongoing management and optimisation	Ensure deliverables are specific, not "AI support."
Outcome-based	₹50K base + % of improvement	When metrics are clear (leads, revenue, time saved)	Baseline measurement must be agreed upfront.
Hourly consulting	₹3-8K/hour	Specific advice, code review, architecture planning	Hours add up. Cap total hours in the contract.
Hybrid (project + retainer)	₹5-10L build + ₹50K-1L/month	Build then maintain	Most realistic model. Ensure retainer reduces over time.

Our preference at Vikrama is hybrid: project-based for the build, reducing retainer for maintenance. The retainer should decrease over time as your team learns the system. An agency that keeps the retainer constant forever is not training your team. They are maintaining dependence.

What most people get wrong

"The cheapest agency is the best deal." An agency charging ₹50K per month for "AI implementation" is either using your project as a learning exercise or reselling a SaaS tool with a thin layer on top. Real AI implementation requires senior engineering time. Senior engineers in India cost ₹1.5-3L per month in salary. An agency charging below their cost base is cutting corners somewhere.

"Big agencies are safer." Big agencies have beautiful decks and impressive client logos. They also have junior staff doing the actual work, layers of management taking a cut, and processes designed for their efficiency, not yours. A small studio where senior engineers do the work often delivers better results than a 50-person agency where your project is assigned to a junior team.

"I need an AI strategy before implementation." You need a clear problem before implementation. Not a strategy document. "Our sales team spends 15 hours per week manually qualifying leads" is a clear problem. The solution (AI lead scoring with automated routing) follows naturally. Do not pay for a strategy document. Pay for a solved problem.

"AI agencies need access to all our data." They need access to the data relevant to the specific problem they are solving. Not your entire database. Not your financial records. Not your customer PII beyond what the system needs. Define data access boundaries in the contract. A legitimate agency will not push for more access than necessary.

Why we are transparent about this (and what we get wrong)

Writing an article that helps people evaluate agencies, including us, might seem like bad business. It is not. We would rather lose a prospect to a better-qualified agency than win a client through opacity.

What we get right: we show live systems, we name our tools (n8n, Smartlead, Claude, HubSpot, Next.js, NestJS), we price on deliverables, and clients own everything we build. Check our about page for who we are and what we have built.

What we get wrong: we are a small team, so we cannot take on more than 3-4 clients at a time. We are opinionated about tech stacks (we will push for n8n over Zapier even if your team prefers Zapier). And we move fast, which sometimes means less documentation than a larger agency would produce.

If our approach fits what you are looking for, start with our audit. It is free. We will review your current systems, identify 3-5 specific improvements, and give you a prioritised plan. If we are not the right fit, we will tell you who might be.

For more on the systems we build, read about AI revenue systems and why agencies cannot build them. For how AI is replacing specific agency functions, see our guide on 7 AI systems that replace 7 agencies for D2C. For the full landscape of AI tools, start with our AI systems for business guide.