ANCIENT CITY AI / DELIVERY
A working system.
A team that owns it.
The goal isn’t an impressive demo. It’s a process your people can use on a busy Tuesday—and understand when something needs attention.
Start with your project brief ↗
01Find the handoff
that breaks.
Follow one real request through the business. Establish the current tools, the person responsible, and the specific delay or repeated work.
You should be able to see: the current process, a baseline measure, and a clearly bounded first project.
02Make the work
explicit.
Agree on the scope, permissions, price, timeline, review rules, and the checks that will show whether the setup works. Check software compatibility before committing to an integration.
You should be able to see: what is included, what costs extra, who approves changes, and who owns the accounts.
03Test the awkward
cases, too.
A missing phone number. A lead that arrives twice. A client who has already replied. We test the agreed workflow with representative examples before live use.
You should be able to see: test results, unresolved issues, a pause procedure, and your launch decision.
04Hand over something
your team can run.
Train the people doing the work, document the steps, and name the owner of exceptions. Compare the agreed measure with the starting point instead of assuming that activity equals improvement.
You should be able to see: operating instructions, account ownership, a handoff date, and the support arrangements in your agreement.
WHAT A TEST RECORD LOOKS LIKE
Proof before
promises.
Illustrative acceptance checks for a follow-up workflow. These are examples of what to test, not reported client results.
| Situation | Expected behavior | Evidence to record |
|---|
| The same inquiry arrives twice | Duplicate is handled by the agreed rule; no duplicate outreach | Record IDs, action log, reviewer result |
| The client replies | Scheduled follow-up stops and the owner is notified | Reply time, stopped action, owner notification |
| A connection fails | Issue is visible and manual recovery is possible | Error, alert, recovery step and outcome |
| Required information is missing | Record is flagged for review; no invented details | Missing field and assigned next action |
Download the first-project preparation guide ↓
CLEAR RESPONSIBILITY
You know
what’s running.
People stay in charge
We agree on which actions are automatic, which are drafts, and which require a person. Technical, legal, and other professional judgments stay with qualified people.
Access follows the job
Use approved accounts and the access needed for the work. Document where information goes and how the client can pause the process or remove access.
Support has a defined scope
The implementation packages include 30 days of fixes to the agreed setup. Optional $350/month support includes a monthly workflow check, a review call, and up to one hour of minor changes. It is not continuous monitoring or an emergency response service.
Evidence earns the claim
Our public examples are simulations. We will publish client results only with permission and enough context to explain what was measured.
Bring us the bottleneck.
We’ll work out whether the right answer is a clearer process, existing software, automation, or AI.
Book an AI consultation ↗