Ask for evidence related to your workflow
A polished demonstration can show design quality, but it cannot establish every aspect of a live deployment. Ask the agency to distinguish concepts, working demonstrations, deployed features and integrations that still require activation. Find out what it actually designed, built or optimised.
For a property workflow, look for experience with communication, records and access between staff and customers. For a legal practice, ask how the team approaches permissions, operational requirements and change management. A recognisable client name is useful context, but the scope and evidence matter more than the logo alone.
Compare the discovery and delivery approach
- How will you map our current process and establish a baseline?
- Which parts need AI, and which can use ordinary automation?
- How will you test exceptions and permissions?
- What stays in review before a live action is allowed?
- Who is responsible for deployment, training and support?
A strong proposal explains what will be delivered and how both sides will decide it is ready. It also identifies dependencies on your staff, data and provider accounts. This makes timelines and pricing more meaningful.
Understand the quote and the recurring costs
Compare like-for-like scope: discovery, build, integrations, migration, testing, training and handover. Then list hosting, model usage, messaging charges, subscription tiers and ongoing maintenance. Ask how the price changes if task volume or document length increases.
There is no reliable universal price for “an AI agent”. A narrow internal drafting tool and a customer-facing workflow connected to several systems have different requirements. Request assumptions, exclusions and a change-control process in writing. An initial discovery phase can reduce uncertainty before a larger commitment.
Settle data, ownership and access
Ask which providers receive your data, who controls their accounts and how access is restricted. Establish the relevant contractual and data-protection requirements with your advisers. An agency’s UK address does not, by itself, determine where every connected provider processes information.
Clarify ownership and handover of custom code, configuration, prompts, evaluation cases and documentation. Check whether you can export your records and move the operation to another provider. Some components will remain subject to third-party terms and licences.
Use a small paid project to test the relationship
A bounded first project can reveal how a supplier handles scope, feedback and unexpected constraints. Agree a useful outcome, representative test cases and a review point. Evaluate the complete result, including the effort your team spends correcting or administering it.
Be cautious with guaranteed rankings, unsupported savings claims or a promise that no human involvement will ever be needed. Ask for the underlying measurement and the conditions under which any result was achieved. Useful evidence includes a defined baseline, sample period and task boundary.
What Orwell brings to the conversation
Orwell works across website development, custom software, AI agents and automation, with a London and Dubai presence. Our selected projects explain the scope of work for Eastmans, Walthams, Socialroom, Cranbrook Legal and Klientflo. Our delivery process covers discovery through support.
A free 30-minute discovery call is a way to establish fit. Bring one process, a few examples and the result you want to improve. For questions about ownership, timing and ongoing arrangements, visit working with Orwell FAQs.
Written for Orwell Lab’s practical guide collection. Examples and calculations are illustrative unless a project is specifically identified.
Discuss your own requirements ↗