Skip to main content
Success
[AI in business]

Build an AI-Driven Enterprise

An AI-driven enterprise uses AI agents to carry out work across the business, connected to its records and systems. We help you design and build those workflows. Your leaders set the goals, decide what to delegate and remain responsible for the results.

[The starting point]

Find the work that keeps waiting

A quote sits in an inbox while someone looks up the price. An invoice waits for a project code. A customer repeats information because sales and operations use different records.

Cheek calls this delay latency. His framework asks leaders to redesign how work passes between people and AI agents. People define the outcome and the limits. Agents carry out the parts you can delegate.

We start with one process and follow it from the first request to the completed job. We record where it waits, who needs to decide, and which information they need. That gives us a specific piece of work to improve.

[Illustrative workflows]

A request through to a completed job

These examples show how we'd divide the work. We'd agree the approval rules with your team before connecting anything to live customer records.

Quoting and sales

Your sales manager defines pricing rules and which promises need approval. We connect the enquiry inbox to the customer records and quoting system.

  1. 01 · AI AGENTRead the request

    Identify the customer and required work. Flag missing details for the sales team.

  2. 02 · SOFTWARECheck the records

    Retrieve approved prices and availability. Calculate totals using your pricing rules.

  3. 03 · AI AGENTPrepare the quote

    Draft the scope from the request and source records. Show assumptions beside the draft.

  4. 04 · SALES OWNERApprove and issue

    Review scope, price and delivery commitments. The system sends the approved version and records it.

We log corrections and quote outcomes. Your sales owner can then approve changes to the rules or examples before the next version goes live.

Supplier invoices

Your finance lead owns coding rules and payment authority. We connect invoice processing to the purchase and project records your team already maintains.

  1. 01 · AI AGENTRead the invoice

    Extract supplier details, line items and references. Keep the original document attached.

  2. 02 · SOFTWAREMatch and check

    Check totals, duplicate references and purchase orders. Flag discrepancies.

  3. 03 · AI AGENTPrepare the coding

    Suggest a project and expense code using approved examples. Explain uncertain matches.

  4. 04 · FINANCE OWNERReview exceptions

    Check discrepancies and approve the posting. Payment approval stays in your finance process.

We track how often staff correct the coding, which suppliers cause exceptions, and how long each invoice waits. Those records guide the next change.

Customer onboarding

Your operations lead defines the handover from sales. We connect the agreed order to your project system and document checklist.

  1. 01 · SOFTWAREOpen the job

    Create the project from the approved order and link it to the customer record.

  2. 02 · AI AGENTPrepare the handover

    Summarise the agreed scope and highlight missing information with links to its source.

  3. 03 · SOFTWAREAssign the work

    Create the agreed tasks, owners and reminders. Flag overdue items for a named person.

  4. 04 · OPERATIONS OWNERResolve the gaps

    Handle unusual terms, missing documents and changes to the customer commitment.

We record where handovers stall and what the team has to correct. Your operations lead decides which changes to carry into the next onboarding.

[Working with Vu]

From the first session to a working process

We work with the people who run the function, as well as whoever owns your systems. They know which exceptions matter and which steps exist for a reason.

  1. Give the leaders something to try

    We run a hands-on session using a small, safe example from your business. Each process owner sees where the agent helps and where it gets things wrong. This follows Cheek's Start with You advice.

  2. Follow a recent job

    Bring a completed quote, invoice or customer handover. We map the steps and measure waiting time separately from time spent doing the work. You get a baseline and a proposed workflow.

  3. Agree who can do what

    We define the agent's inputs, permitted actions and escalation rules. Your team names an owner for each decision. We agree access to records, spending limits and the evidence a reviewer needs.

  4. Build and test the connections

    We connect the required systems and test against past examples, including failures. Staff review the results before we allow live actions. The pilot includes a way to stop it and return work to the team.

  5. Review the results before extending it

    We compare completed work with the baseline, including corrections and running costs. Your process owner approves any changes. We then decide which adjacent workflow is worth connecting next.

You get the agreed process map and access rules, working software connected to your systems, and instructions for the people who run it. We include an activity record and a review plan so someone can see what happened and decide what to change.
[Responsibility]

Your leaders still own the decisions

In Cheek's framework, each leader designs how their function works with people and agents. We help turn those decisions into software. Your finance lead still owns financial controls. Your sales lead still owns pricing and customer commitments.

For regulated work, we agree the boundaries with the responsible specialists before building. We can help prepare evidence and coordinate work around a review. Required approvals and professional judgements stay with the people authorised to make them.

We record agent actions and the information behind them. A named person handles exceptions, and the team can pause the workflow if the results stop meeting the agreed standard.

[Measurement]

Check the whole process

Cheek makes a longer-term argument about a business's ability to compete. For the first implementation, we also need evidence that the work has improved.

Time to completion

Track elapsed time from request to finished job, including queues and approvals.

Review and rework

Count corrections, exceptions and the time people spend checking the output.

Cost per completed job

Include AI usage, licences and support alongside the staff time the process still needs.

Use of the freed capacity

Agree what the team will do with the time, then check whether that work happens.

[Start with one process]

Bring us a job that takes too much chasing

Tell us where it starts, who touches it and what a completed job looks like. A director will reply in writing within two working days with the shortest route, a price where possible and any open questions.

Source: Paul Cheek's No One Works Here video library. The workflow examples on this page are illustrative. For more on the business design behind the framework, see The Architect and the Moat.

Message us on WhatsApp