Agentic AI Pilot Contracts: The Commercial Terms UK Contact Centres Must Insist On

Arkadas Kilic 6 min read · 2026-09-06

Most contact centre leaders approaching agentic AI for the first time are thinking about the technology. Which tasks will the agent handle? How does it connect to the CRM? What happens when it cannot resolve a query?

Those are the right questions. But the conversation that actually determines whether a pilot succeeds or fails happens before any code is written. It happens in the contract.

I have reviewed dozens of AI pilot agreements on behalf of contact centres in financial services, insurance, utilities, and healthcare. The same patterns appear repeatedly. Vendors who are confident in their delivery write tight, measurable contracts. Vendors who are not write contracts full of vague language, uncapped scope, and exit clauses that only favour one side.

This post gives you the specific commercial terms to insist on before you sign anything.

Why Pilot Contracts Are Different From Standard Software Agreements

A standard SaaS agreement governs a product that already exists. A pilot contract governs a build that has not happened yet. That distinction matters enormously.

With agentic AI, you are not licensing software off a shelf. You are commissioning the design, build, testing, and deployment of an autonomous system that will make real decisions in your live contact centre environment. The contract must reflect that reality.

If it reads like a standard software licence with a few lines about AI appended at the end, walk away.

Term 1: A Fixed Scope With a Written Definition of Done

The single most common source of pilot failure is scope creep dressed up as progress.

Your contract must define, in writing, exactly what the pilot covers. Not "AI-assisted call handling" but something like: "Autonomous handling of FNOL calls for motor insurance, covering the first 7 minutes of interaction, with structured data capture to Salesforce and warm transfer to a human agent for claims above £5,000."

The definition of done must be equally specific. What does a successful pilot look like? Acceptable terms include:

• Containment rate above a defined threshold (for example, 65% of targeted call types handled without human escalation)
• Average handle time reduction of at least 20% on in-scope call types
• Zero compliance breaches across a defined test period
• Successful integration with named systems confirmed by your IT team

If the vendor cannot write a definition of done before the pilot starts, they do not know what they are building.

Term 2: A Timeline With Milestones, Not Just an End Date

A credible agentic AI pilot for a UK contact centre should reach production in 4 to 6 weeks. That is not a marketing claim. It is a reflection of what is achievable when the scope is defined, the AWS infrastructure is pre-built, and the team has done this before.

Your contract should include:

• Week 1 to 2: Discovery, data access, and architecture sign-off
• Week 2 to 3: Agent build and integration development
• Week 3 to 4: Internal testing and compliance review
• Week 4 to 5: UAT with your team
• Week 5 to 6: Production deployment and hypercare period

If a vendor is quoting 6 months for a pilot, ask them to justify every week. Pilots that stretch beyond 8 weeks almost always fail to reach production because internal stakeholders lose confidence and budgets get reallocated.

The contract should specify what happens if milestones are missed. A 10 to 15 day grace period is reasonable. Beyond that, you should have the right to pause payments and renegotiate scope without penalty.

Term 3: Data Ownership and Residency, Written Explicitly

For any UK contact centre operating under FCA, ICO, or NHS governance, data terms are not a negotiating point. They are a hard requirement.

Your contract must state:

• All customer data processed by the AI agent remains your property at all times
• Data is processed and stored within the UK or EU (specify the AWS region, for example eu-west-2 for London)
• The vendor has no right to use your data to train models for other clients
• Data deletion timelines on contract termination (30 days is standard; 60 days maximum)
• A clear description of which subprocessors handle data and under what terms

Any vendor unwilling to specify AWS region in the contract is either not using AWS or is not sure where your data lives. Neither is acceptable.

Term 4: Compliance Accountability, Not Just Compliance Claims

Every vendor will tell you their system is compliant. The question is who is accountable when it is not.

For regulated UK contact centres, your contract should specify:

• The vendor's responsibility for maintaining audit logs of every agent decision and action
• A defined process for producing those logs within 48 hours of an ICO or FCA request
• Clear liability allocation if an agent action results in a regulatory breach
• The vendor's obligation to notify you within 24 hours of any security incident affecting your data
• Confirmation that the system architecture supports your existing call recording and quality monitoring obligations under COBS, GDPR, or equivalent frameworks

Compliance built into the system from day one is not a feature. It is a baseline requirement. If the vendor is treating it as an add-on, that tells you something important about how they build.

Term 5: A Clean Exit Clause With IP Portability

This is the term most contact centres forget to negotiate and the one they most regret omitting.

At the end of a pilot, you may choose not to proceed to full deployment. That is a legitimate outcome. Your contract must ensure that decision does not leave you worse off than when you started.

Insist on:

• A 30-day written notice period to exit after the pilot, with no financial penalty beyond fees already incurred
• Full handover of all configuration files, integration code, prompt libraries, and workflow definitions that were built using your data and your processes
• No vendor lock-in clauses that prevent you from using the same AWS services or contact centre platform with a different partner
• Clarity on what the vendor retains (their proprietary frameworks, generic accelerators) versus what belongs to you (the specific implementation built for your environment)

If a vendor insists on retaining ownership of the implementation built for your use case, that is a red flag. You paid for it. You own it.

Term 6: Pricing Transparency and a Cap on Pilot Costs

Pilot pricing for agentic AI in UK contact centres typically falls in one of three models:

1. Fixed fee for the pilot build, with separate commercial terms for production deployment

2. Time and materials with a defined cap (acceptable if the cap is enforceable)

3. Outcome-based, where fees are tied to measurable results (rare but the most aligned model)

What is not acceptable is a pilot contract with uncapped infrastructure costs, vague references to "additional charges for integrations", or a clause that allows the vendor to invoice for work outside scope without prior written approval.

A credible pilot for a mid-sized contact centre (50 to 200 seats) should cost between £25,000 and £75,000 all-in, depending on the number of integrations, the complexity of the call types, and the compliance requirements. If a vendor cannot give you a fixed number, they have not scoped the work properly.

Term 7: Hypercare and Knowledge Transfer After Go-Live

Deployment is not delivery. The 2 to 4 weeks after a production go-live are when most issues surface and when the quality of the vendor relationship becomes apparent.

Your contract should include:

• A defined hypercare period of at least 2 weeks post go-live, with named contacts and response time SLAs (P1 issues resolved within 4 hours; P2 within 24 hours)
• At least one structured knowledge transfer session with your internal team covering how to monitor agent performance, interpret logs, and escalate issues
• Documentation of the agent architecture, integration points, and configuration in a format your team can maintain

A vendor who builds something your team cannot understand or maintain has created dependency, not value.

What a Good Contract Signals About the Vendor

The commercial terms a vendor is willing to sign tell you more about their confidence in their own delivery than any case study or demo.

Vendors who have built agentic AI systems in production, in regulated environments, with real contact centres, know exactly what a fair contract looks like. They are not afraid of measurable outcomes, clean exit clauses, or compliance accountability because they have delivered against those terms before.

Vendors who hedge, obscure, or resist specific terms are usually doing so because they are not certain they can deliver.

At Rel8 CX, we build enterprise-grade agentic AI systems for contact centres in financial services, insurance, and utilities. We reach production in 4 to 6 weeks. We sign the kind of contracts described in this post because we are confident in what we build.

If you are preparing to evaluate vendors or review an agreement you have already received, we are happy to walk through it with you.