Founder Brief

Three systems to install before Q4 planning

Founder Brief #001. Three client systems, and the real screens behind each one.

DashboardLim · Founder's desk

2026-07-014 min read

Three systems we built and still run. One keeps a client's week on rails, one runs a founder's inbox without ever sending for them, and one puts a whole sales floor on the office wall. Here is what each client came to us with, and what changed.

weekly reports in one client's folder

noise messages dropped in a single day

$1.9m

team revenue live on the office wall

The founder inbox command centre we built, showing 87 emails in the feed, 6 processed, 1 AI drafted, 5 AI flagged, 3 human needed, 0 refunds, 1 urgent, 81 dropped, a 0.94 average confidence and a 17 percent automation rate. — One day inside the inbox system below. 87 messages in, 81 dropped by plain rules before any AI ran, and the rest sitting in a named queue with an owner.

Every screenshot below came out of a system that was running that day.

System 01 · Weekly reporting

The report that arrives before anyone asks

One of our Australian clients kept asking the same question in different words. How are we actually doing? Status lived in people's heads and in whatever meeting happened next.

So we put the week on rails. A scheduled email opens it with every meeting booked, and the same status report closes it every Friday. We automate the assembly, not the words about what the week meant.

The moment it earned trust was not a good week. Their founder pushed back on a lead number in front of everyone, we answered in the thread, and the tick went on. A report earns trust the first time someone challenges it and gets a straight answer in the open.

A Slack message relaying a client founder's emailed question about a lead count, with the original report email attached as an image, a green check reaction and a five-reply thread beneath it. — The challenge, in the shared channel. Five replies later it was settled and the tick went on.

A real weekly status report open in Google Docs, showing Focus This Week with overdue items flagged Past and an In Review section beneath it. — The top of a real Friday report. Every visible line is what the client read that week, overdue flags included.

The n8n executions panel for the scheduled meeting report workflow, showing runs on a Monday and Thursday rhythm, each marked succeeded in about six seconds. — The run history behind the scheduled opener. A few seconds a run, every run. This is what "the report just arrives" looks like from the inside.

What to look at

Overdue work carries a Past flag. We report what slipped rather than quietly dropping it.
A change asked for in chat comes back as done work in the Friday report. The loop closes in writing.

A thin report that arrives every week beats a perfect one that doesn't.

System 02 · The founder inbox

The inbox that does the looking

A consumer brand came to us drowning. Seven mailboxes plus WhatsApp, Meta DMs and Shopify Inbox, around fifty customer emails a day in peak season, and someone clearing the backlog late at night.

We built one triage layer across all of it, and the first thing it does is not AI. Plain rules, which they edit themselves, drop newsletters and notifications before the model sees anything. What survives lands in a named queue with the reason it landed there and what a person should do next.

Every reply waits as a draft. Refunds over an agreed threshold, anything legal, and diet-critical questions never become drafts at all, they ping a named human. Their founder now works two queues and a morning digest instead of a midnight inbox.

The alert queue and human review queue from the inbox command centre, each row carrying the reason the message was flagged, a confidence score, its age and the recommended action. — Nothing is filed away silently. Every message the rules did not drop sits in a queue with a reason and a recommended action.

What to look at

Nothing on this screen can send itself. The system drafts, routes and escalates. People send.
It once quoted a discount it should not have, so we took pricing out of every AI reply. The fix was a sharper rule, not a smarter model.

The draft is the product. Auto-send is something a category of reply has to earn, and money, legal and safety never will.

System 03 · Sales visibility

The scoreboard everyone can see

Another Australian client, a trade-certification provider, ran commission out of one person's spreadsheet. Reps could not see where they stood, and nobody could show how a number had been reached.

We put the sales floor on the office wall. A live board cycles daily, weekly, monthly and year to date, and every rep can open the same numbers themselves in a portal instead of asking for them.

Commission is now worked out in the open, from real payments rather than a formula in someone's head. Finance pays from figures that trace back to the deal that earned them.

A sales leaderboard on an office TV, marked live, showing team revenue of A$1,896,954 against a A$2,430,000 target at 78 percent, 469 deals closed, and a ranked list of reps with their deal counts and average deal size. — The board the sales floor actually watches. Team revenue against target, deals closed, and every rep ranked, updating live.

A commission view showing a commission leaderboard for the last thirty days, a six-month commission trend chart, and a panel explaining how commission is calculated from gross profit after GST and allocated cost. — The same numbers, one level down. Commission per rep, the trend behind it, and the calculation written out where anyone can check it.

What to look at

The board is live. A closed deal shows up on the wall while people are still in the room.
Every commission figure traces back to a payment, so "how did you get that number" has an answer on screen.

People chase a number they can see. Putting the scoreboard on the wall changed behaviour before any of the reporting did.

What the three share

Same shape, three times

The shape repeats. Machines collect, route and draft on a schedule. People own the judgement, which is the explanation in the thread, the edit to the draft, and the approval before anything is written anywhere.

And every exception has an address. A challenged number becomes a thread, a strange message becomes a queue row. None of these systems is finished, and the ones that survive are the ones built to be questioned.

Steal the shape

None of this needs new tools. It needs a few decisions made in the right order.

Put the week on a schedule
One report, a fixed day, a fixed shape. Send it before it is polished.
Filter with boring rules first
Drop the noise deterministically and log every drop with its reason. Save the model for the messages that deserve one.
Draft, don't send
Let AI prepare the reply, the summary and the note. Keep a person on every send, and keep money, legal and safety out of its reach entirely.
Give every exception an address
A challenged number, a strange message, a refund over threshold. Each should land in a thread or a queue with an owner.
Put the scoreboard where people can see it
A number on a wall changes behaviour faster than a report nobody opens. Make sure it traces back to something real.

The proof behind this issue

System 02 in full. The architecture, the message flow and the honest outcomes of the inbox build, on the proof feed.

Case study

· Inbox automation

Founder Inbox OS

Seven inboxes and customer channels were feeding one overloaded team. We built a human-reviewed system that sorts every message, drafts the next reply and escalates anything that needs judgement.

mailboxes triaged

noise messages filtered in one morning

replies sent without human approval

Freckleberry

2026-06-18OpenClawn8nRailwayRead the case study →

Get the next issue in your inbox

Real systems we've shipped, with practical lessons you can use today.

Which of the three is your gap?

The report nobody reads. The inbox that runs the day. The scoreboard nobody can see. Tell us which one is costing you, and we will show you what the fix looked like.

Book a systems audit Discuss a system

Keep exploring the Intelligence Library

Newsletter

Reporting & Dashboards

The weekly reporting cadence that ends "how are we doing?"

A teardown of a reporting system that has run for a real client for a year and a half. The actual Monday emails, the actual Friday reports, the Slack thread where a number got challenged, and the workflow underneath.

The report that ends "how are we doing?" — the real client reporting folder of dated weekly status reports, with the n8n executions list showing the scheduled sends succeeding. Client name masked.

One client's week has closed with a Friday status report for a year and a half, and opens with a scheduled Monday email. This issue shows the real artifacts, masked and cropped but never rewritten.

7 min read

2026-06-17

Read the issue

Newsletter

Operations

Teardown: the inbox that drafts itself

The inbox system we built, on a real month of a client's mail.

The founder inbox command centre we built, on a thirty day window, showing 2256 messages processed, 1052 drafted by AI, 1204 flagged, 91 needing a human and 5536 dropped by filter rules, at a 47 percent automation rate.

Plain rules did the heavy lifting. In one month 5,536 messages were dropped before any AI ran, each logged against the rule that dropped it.

6 min read

2026-06-24

Read the issue

Case Study

Sales & CRM

Sales Command Center

Ads were working. The backend was not. We reset how the CRM was used, automated the record hygiene, built a trusted data layer on top of it, and only then built the Sales Command Center — so the founder and the sales floor finally read the same live numbers.