AI Workflow Pilot

Who this is for

Teams with one clear use case, such as classifying enquiries, summarising approved documents or preparing a first draft for human review.

What is included

• One AI task inside one workflow with up to three action steps.

• Up to two tools connected through supported connectors.

• One selected AI model or service and documented reasons for the choice.

• Up to ten representative evaluation cases, including failures and escalation.

• Human review before external or consequential actions, usage limits and failure notification.

• One revision, a handover session and a review two weeks after agreed launch.

How it works

We agree the task, examples, cost limit and success checks. You review a test version before launch. Target: one to two working weeks once scope, access and examples are ready.

What you take away

A tested pilot, editable prompts and configuration, evaluation results and an operating guide. You can decide whether to expand based on measured results.

Price

£995 one-off for the starter scope. 50% at booking and 50% after the acceptance demonstration, before final handover. Model usage and software subscriptions are separate.

When you need a different scope

Custom models, large document libraries, bespoke integrations and autonomous regulated decisions are outside the starter. A normal handover is included; there is no need to buy AI Handover as an extra.

Ownership, costs and next step

Start with a short fit call. We confirm the scope, total price, access needed and delivery dates in writing before work begins. Prices are in GBP, excluding VAT where applicable. Software subscriptions, hosting and AI usage are separate and disclosed before you commit. A Hub subscription is only needed if it is the platform you choose. Ongoing support is optional.

You keep the agreed deliverables and editable documentation. Bespoke code transfers after payment, subject to third-party licences; existing platforms remain licensed. We explain any limitations on export or moving providers. Results depend on the process and how the tools are used; no particular savings are guaranteed.

Builds include 30 days of fixes for defects against the agreed scope from handover. New features, third-party outages and platform changes are separate from that cover.

Before we begin

Define one task and describe how a person decides whether its output is acceptable. Bring representative examples, including cases that should be rejected, escalated or handled manually. Approved inputs and clear expected results are more useful than a large, unreviewed collection of documents with no agreed purpose.

Identify the workflow owner, the tools involved and the cost limit to discuss during scoping. Decide where human review belongs before an external or consequential action. The starter is a bounded pilot for learning about a task in practice, so it needs an evaluation question as well as a technical connection between tools.

An illustrative walkthrough

An illustrative pilot might classify an incoming enquiry and prepare a draft for a colleague. The selected model performs the agreed task, but a person reviews the result before it produces an external or consequential action. Cases outside the expected pattern should have an escalation route rather than be treated as a confident answer.

Use the evaluation cases to inspect both useful outputs and failures. A short, plausible draft may still omit a required fact or use the wrong category. Reviewing those outcomes against the agreed success checks helps the team decide whether the pilot is ready for its limited use and which assumptions need further work.

Using the result

Keep the prompts, configuration and evaluation results together. When a prompt or input changes, the existing examples provide a starting point for checking what has improved and what may have become less reliable. The operating guide should remain useful to the person who reviews results and handles notifications.

The included review two weeks after agreed launch provides a point to discuss the pilot against the chosen measures. Expansion is a separate decision, not an automatic part of the starter. A normal handover is already included, so you do not need to purchase AI Handover simply to understand a new Labs pilot.