A real working day, company names hidden
One day, nobody watching.
ARO Solutions runs its own business development with a team of AI agents: a chief of staff and 17 specialists that research companies, check facts, write, send and keep records on a schedule, with nobody watching. This page replays one real day, Thursday 19 November 2026, from the files those agents wrote: what each one did, which rules stopped it, and what it passed to a person. Company names are hidden, and the times, counts and reasons are as logged. It is the same pattern ARO offers to build for a partner's client: one process run end to end, with checks, a person where a decision needs one, and a record of every action.
Last updated · ARO Solutions, Amman
Taken from that day's run logs. Company names are hidden. Counts and reasons are as logged, and times are Amman time.
- 08:05
The sending engine
Reads the mailbox before it does anything else.
Eighteen messages are waiting. None is from a company it has written to, and none is a bounce or a request to stop, so nothing needs an answer.
The sending engine
Plans the day inside its limits.
Fourteen emails: six follow-ups, six first emails for one campaign and two about search visibility. The ceiling for the day is 50 emails, 20 of them to companies that have never heard from the studio.
The sending engine
Finds an address that leads nowhere useful.
One clinic's listed email address turns out to be its jobs mailbox, not the business contact. It never enters the plan.
Passed to a person
No email is written. The lead goes to a person with one line: call instead.
The quality check
Reads all fourteen emails and fails one.
Thirteen pass. The fourteenth is for a clinic whose website now shows an expired-domain page, so its address would probably bounce.
Passed to a person
That email is not sent. The lead goes to a person with the phone number and the reason.
- 08:23
The sending engine
Sends ten emails, then stops itself.
After the tenth send it checks for bounces. One has come back: 10% of the day's sends, against a limit of 5%.
A rule stepped in
First emails to new companies are switched off until tomorrow. Three stay unsent.
- 10:43
The partner desk
Looks for replies in two mailboxes.
No person has replied. The only item on file is an automatic acknowledgement from the day before, already logged as not a reply.
- 10:46
The partner desk
Writes five follow-ups.
Each adds one use case for that company's own platform, runs 82 to 89 words, stays in the original thread and carries the line that lets the reader say no.
The partner desk, research
Researches new companies with about 50 web lookups.
Five qualify, and for each one the contact page and the page behind every fact were opened during this run. Three are parked because their site hides its email address. Seven are turned away with a reason on file, such as a large delivery team of its own abroad, its own AI product, or being a development shop and so a competitor.
The partner desk, country check
Checks each country's email rule before a company abroad can be written to.
Six companies in five countries pass, each with its proof on the record: a UK limited company confirmed on the official register, a general company mailbox only for Ireland and the Netherlands, an email address printed on the company's own site for Australia, a postal address for the United States.
A rule stepped in
One more is dropped: it sells the same service to agencies, so it is a competitor.
- 10:57
The partner desk, quality check
Checks eleven first emails, one at a time.
The opening observation is true and its page was opened today. The address is printed on the company's own site. The company has never been emailed, checked against four records. Nothing in the email is invented.
A rule stepped in
All eleven pass, and still none goes out today: the morning's bounce stop holds first emails until the next day.
- 12:40
The sending engine
Sends the five follow-ups.
Five sent, none skipped, none failed, between 12:40 and 12:43.
- 15:43
The partner desk
Checks its own work and stays quiet.
The afternoon run matches the five sends against the send log and finds no bounce and no reply. Its rule for a day with no news is to send the founder nothing. Every change of the day is saved as a dated entry the team can read back.
By the end of the day: 15 emails sent, 14 held by a rule, 2 leads passed to a person.
The rules did the important work
An agent that can send email needs something that can stop it. These are the rules behind the day above.
The bounce stop
When undelivered mail passes 5% of a day's sends, first emails to new companies end for the day. It is code, not a model's judgement. That day it held fourteen emails.
Each country's rule, checked first
A company abroad is emailed only when the proof ARO's rule for that country asks for is on the record: a register entry, a published email address, a postal address in the footer.
Never twice
Before a first email, the company is matched against four records. One company gets one first email.
No sentence without a source
Every opening line must point to a page opened during that run. An agent that cannot show the page does not get to write the sentence.
A person for money
Prices outside the published list, contracts and anything signed go to the founder. The agents prepare, and a person commits.
Quiet when nothing happened
A run with no news sends no message. The log still records what was checked, so silence can be audited too.
The same pattern on a client's process
Outreach is only the process ARO runs for itself. The parts carry over: agents that do the reading and drafting, rules that can stop them, and a person where a decision needs one.
| Process | What the agents do | Where a person decides | How it is measured |
|---|---|---|---|
| Support ticket triage | Read each ticket, find the customer and earlier cases, then draft the answer or route it with a summary. | Refunds, complaints and anything the rules mark as sensitive. | Minutes saved per ticket, and the share routed correctly on agreed test tickets. |
| KYC document checks | Read the documents, scanned ones included, extract the fields, compare them with the application and list what is missing. | Every approval and every rejection. | Extraction accuracy on an agreed test set. |
| Quote drafting | Read the request, pull prices and terms from the systems of record, and draft the quote. | Sending the quote, and any discount. | Time from request to draft, and drafts sent without edits. |
| Invoice and purchase-order matching | Read invoices and purchase orders, match them to the ERP record and draft the entry. | Posting the entry, and every mismatch. | The share matched correctly on agreed test documents. |
Swipe the table sideways to see every column.
These four are designs, not past projects. A first pilot fixes the real scope, the rules and the test cases with you.
What a four-week pilot hands over
One process running end to end in your client's environment.
The rules written in plain language, where a manager can read them.
A person's decision built in wherever money, contracts or exceptions appear.
A record of every action, with the reason for it.
An evaluation report on test cases agreed with you before the build starts.
A demo every week, and documentation your own team can operate from.
What this page does not claim
It shows how the system operates. It quotes no sales results: ARO's partner outreach began in November 2026, and it is too early for any.
It is one day, not a record of uptime. Other days that month did not run in full.
The agents do not close deals. They research, check, write, send and record, and they stop when a rule says stop. A person takes the calls and signs.
It is one studio's own operation, shown as an example of the pattern. A client's process will have different tools, different rules and different points where a person decides.
What we build with
- Claude
- Python
- Postgres
- Git
What we've built
- HerShape. An AI fitness app we built and run, live on the App Store and Google Play.
- Ring Studio. A real-time 3D ring builder, live at a jeweller in Amman.
- Websites and stores for brands in Jordan and the Gulf.
Questions partners ask
Is this a simulation?
No. Every entry comes from the files ARO's agents and its sending engine wrote on 19 November 2026. Company names and addresses are removed. The times, counts and reasons are unchanged.
What does the agent team run on?
Claude models do the reading, research and writing, on a schedule. A small Python service on a dedicated machine does the sending and enforces the limits, so the rules that matter most are code and do not depend on a model's judgement. Leads live in a Postgres database, and every change is saved in a git repository.
What happens when an agent is wrong?
It is expected, and the system is built around it. A fact needs a page behind it, a separate check reads every first email, the limits are enforced in code, and anything about money or contracts goes to a person. On 19 November those checks held fourteen emails and sent two leads to a person.
Can a system like this run inside our client's environment?
Yes. For partner work the code lives in your repositories or the client's, the data stays in the client's environment, and the model can be an in-region or self-hosted one when the client requires it.
Does it work in Arabic?
Yes. The same agents read and write Arabic and English, and ARO also builds assistants that answer from Arabic contracts, policies and scanned files.
How do we start?
Send one live bid or client request that needs an agent workflow. Within 48 hours you get a solution outline with the architecture, the timeline and our price to you. A fixed-price pilot of about four weeks follows when you win the work.
Related
- For integratorsARO Solutions is a white-label AI team for Gulf IT integrators and vendor partners: agent workflows, document assistants in Arabic and English, and integrations on the platforms you sell. Free bid outline in 48 hours.
- For agenciesARO Solutions is a white-label development partner for branding, UX and marketing agencies: web apps, portals, integrations, mobile apps and AI chat, built from your designs. You keep the client.
- AI agentsARO Solutions builds production AI agents and agentic workflows with LangGraph, Claude and MCP: ticket triage, document checks, quote drafting and support. A first agent is a pilot of about four weeks, with a written scope and price within 48 hours.