Güray Uzun | Founder-operator | London

AI Operations for owner-managed UK businesses

I build supervised AI operating systems for owner-managed businesses. I test them first inside my own real UK companies.

The operating model is being tested inside property management and hospitality businesses that I own. Property-investment workflows are documented here as candidates while that worker layer is on hold. It is for owner-managed teams where repetitive admin is expensive and the human decision still matters.

AI Operations Review

The review identifies one repetitive or expensive workflow that may suit AI-assisted automation, maps its human gate, and may end with a recommendation not to automate.

See the review ↓

A practical starting point

Start with one costly workflow

The first problem is not the whole company. It is one repetitive or expensive workflow involving enquiries, quotes, documents, reporting, scheduling preparation, or follow-up. The aim is to find where supervised AI can prepare the work and where a person must still decide.

Who this is for

  • Property management
  • Hospitality
  • Removals and logistics
  • Local service businesses
  • Other admin-heavy SMEs

Candidate map

Workflows worth reviewing

These are workflow candidates, not claims of live integrations or current automation.

  • Enquiries and lead triage
  • Quotation drafting
  • Customer-message drafting
  • Document processing
  • Owner or client reporting
  • Task generation
  • Scheduling preparation
  • Pricing or revenue analysis
  • Follow-up workflows
  • Internal knowledge retrieval

Truth labels

Four states, stated plainly

AI-assisted

AI prepares a draft, analysis, or recommendation. A person still owns the decision and the action.

Human-approved

An AI work product can move only after a named person reviews and accepts it.

Automated

A bounded workflow runs without a manual step inside defined controls. This label requires public evidence.

Experimental

A constrained test used to learn whether the workflow is suitable. It is not treated as production operation.

No workflow is presented as automated unless public evidence supports it. None currently does on this service page.

The method

Map, gate, pilot, measure, decide

  1. 01

    Map the work

    Document the trigger, handoffs, systems touched, exceptions, and failure cost.

  2. 02

    Define the human gate

    Name what a person must review, decide, apply, send, sign, or publish.

  3. 03

    Run a bounded pilot

    Test one constrained workflow without widening its authority.

  4. 04

    Measure the result

    Compare the pilot with the present process using agreed evidence.

  5. 05

    Decide whether to widen

    Keep, revise, stop, or widen only after the evidence is reviewed.

Evidence boundaries

Inspect the operating record

Evidence is separated into four categories so published system activity is not mistaken for an outcome.

Business identity
Existing public company links and site records support business identity.
System state
Status, agent, and operating-record pages describe the published system state.
Dated public work products
Dispatches and notes can be inspected separately, but they are not case-result evidence or proof of human clearance.
Measured outcomes
No measured outcomes are published.