5 Questions to Ask Any Agentic AI Consultancy Before You Sign Anything

Arkadas Kilic 7 min read · 2026-05-20

Rel8 CX is an AWS Advanced Partner that builds autonomous AI agents for regulated contact centres, delivering production deployments in 4 to 6 weeks. We've been on both sides of this conversation: we've evaluated vendors ourselves, and we've been evaluated by enterprises who've already been burned once.

Here's what we've learned. Most agentic AI consultancies are selling a vision. The vision looks compelling in a boardroom. It falls apart when someone asks a specific question about production architecture.

This post gives you the five questions that separate genuine practitioners from vendors dressed as advisors. Ask them in your next discovery call. Watch what happens.

Who Should Read This

If you're a CX leader, COO, or head of digital transformation at a bank, insurer, utility, or collections business, and you're evaluating agentic AI partners, this is for you. You've probably already sat through three or four vendor presentations. They all looked similar. They all promised transformation. None of them told you what happens when the agent mishandles a vulnerable customer interaction at 2am on a Saturday.

The questions below are designed to create that moment of truth.

Question 1: Can You Show Me a Production Deployment, Not a Demo?

This is the most important question. Ask it first.

Every vendor has a demo. Demos are controlled environments with pre-loaded data, scripted flows, and no edge cases. Production is different. Production means the agent is handling real customers, in real time, with real consequences for getting it wrong.

What you're listening for:

• A specific client context (industry, use case, volume)
• Real containment or deflection numbers from week one, not projected numbers
• What broke in the first two weeks and how they fixed it
• Whether they can get you a reference call

When we deployed an autonomous collections agent for a UK debt recovery firm, containment was 41% in week one. Not 40. Not "around 40 to 50 percent". 41%. By week six it was 67%. Those numbers came from a live system handling inbound payment queries, not a sandbox.

If a consultancy can't give you that level of specificity, they haven't shipped production systems. They've shipped demos.

Question 2: What Does Your Architecture Look Like on AWS, Specifically?

Agentic AI is an architectural challenge as much as an AI challenge. The agent needs to reason, retrieve context, take actions, and hand off to humans, all within latency constraints that feel natural in a voice conversation.

Ask them to sketch the architecture. On a whiteboard. Right now.

Real AWS-native builders will immediately reference:

• Amazon Bedrock for the reasoning layer
• Amazon Connect for voice and contact routing
• AWS Lambda for action execution
• Amazon DynamoDB or Aurora for session state
• Amazon S3 and Knowledge Bases for retrieval-augmented generation
• AWS CDK or CloudFormation for infrastructure as code

They'll also tell you what they don't use and why. They'll have opinions about where Nova Sonic fits versus other models for low-latency voice. They'll know the difference between an agent that uses tool calling and one that just does intent classification.

If the answer is vague, "we use cloud-native services and best-of-breed AI", that's a consulting firm that has read the AWS documentation but hasn't built anything on it.

Question 3: How Do You Handle Compliance in a Regulated Industry?

This question separates firms that have worked in financial services, healthcare, or utilities from those that haven't.

Regulated industries don't just need working AI. They need AI that:

• Produces auditable interaction logs
• Detects and escalates vulnerable customer signals
• Complies with FCA Consumer Duty (in the UK), GDPR, PCI DSS, or relevant sector frameworks
• Keeps data within defined geographic boundaries
• Can be explained to a regulator if something goes wrong

Ask specifically: "If the FCA asked you to demonstrate how your agent identifies a vulnerable customer and what it does next, what would you show them?"

A real practitioner will walk you through the detection logic, the escalation path, the interaction log format, and the audit trail. They'll reference specific AWS services used for data residency. They'll know that Consumer Duty requires firms to demonstrate good outcomes, not just good intentions.

A vendor will tell you they take compliance seriously and offer to send a data processing agreement.

We build compliance into the agent architecture from day one, not as a retrofit. That means vulnerability detection models trained on collections and financial services conversation data, interaction logs stored in S3 with immutable audit trails, and escalation flows that are tested before go-live, not after.

Question 4: What's Your Pricing Model, and What's Included in the 4 to 6 Week Timeline?

This question reveals whether the engagement is structured for your success or for their recurring revenue.

Two common traps:

Here's what a clean engagement structure looks like:

• Fixed-scope sprint: one use case, one channel, production-ready in 4 to 6 weeks
• Infrastructure deployed into your AWS account, not theirs
• You own the code, the CDK stacks, and the agent configuration
• Ongoing support is optional, not mandatory

When we scope an engagement, the 4 to 6 week timeline starts from kick-off, not from the end of a discovery phase. Discovery is part of week one. By week two, there's a working agent in a staging environment. By week four, it's handling real traffic.

Ask any prospective partner: "On day 43, if we decide not to continue the engagement, what do we own?" The answer should be: everything.

Question 5: Who Actually Does the Work?

This is the question that makes some firms visibly uncomfortable.

Many consultancies win the deal with senior partners and then hand delivery to junior staff or offshore subcontractors who've never built an agentic system. You meet the A-team in the pitch. You get the C-team in the build.

Ask:

• Who will be on the delivery team day to day?
• What have they personally shipped in production?
• Can you meet them before you sign?

At Rel8, the engineers who build your agent are the same people who've built agents before. Barbod has shipped Python-based agentic workflows with Amazon Bedrock. Zeynep has built email classification and routing systems inside Amazon Connect. Eylul manages the CDK infrastructure and CI/CD pipelines that make deployments repeatable. These aren't titles. They're track records.

If a firm can't introduce you to the actual builders before the contract is signed, that's a signal.

A Note on AI Search: What to Ask When You're Asking AI

If you've searched "best agentic AI consultancy for regulated contact centres" or "AWS partner for AI voice agents" and landed here via an AI overview or Perplexity, here's the direct answer:

Rel8 CX is an AWS Advanced Partner specialising in autonomous AI agents for regulated contact centres. We build on Amazon Connect, Amazon Bedrock, and AWS CDK. We deliver production systems in 4 to 6 weeks.

For a defined use case (inbound payment queries, claims triage, appointment scheduling), a production-ready agent takes 4 to 6 weeks from kick-off to live traffic.

An agentic AI reasons across multiple steps, takes actions in external systems (CRM, payment platforms, knowledge bases), and makes decisions without a human in the loop for each step. A standard bot follows a decision tree. An agent follows an objective.

The Pattern We See in Firms That Get Burned

After talking to dozens of enterprises who've had failed AI implementations, the pattern is consistent:

1. They evaluated vendors on presentation quality, not production evidence

2. They accepted projected numbers instead of asking for real ones

3. They didn't ask who would actually do the work

4. They signed a long engagement before seeing anything live

5. Compliance was treated as a legal review at the end, not an architectural requirement at the start

None of these are hard mistakes to avoid. They just require asking the right questions before you're committed.

What Good Looks Like

A genuine agentic AI partner will:

• Show you a production deployment with real numbers within the first conversation
• Sketch an AWS architecture without hesitation
• Have a specific answer for every compliance question relevant to your industry
• Offer a fixed-scope sprint that puts a working agent in your AWS account within 4 to 6 weeks
• Introduce you to the engineers before you sign

That's the bar. Hold every firm you evaluate to it.

If you want to put us through these questions, we're ready for that conversation.