Everything still runs through you.

That is not a workload problem. It is an architecture problem, and architecture is fixable.

Joy Santos
AI-powered systems architect and operations manager  ·  Digital project management and instructional design

I build ecosystems that run without the architect.

How I think

The same scene, in business after business.

What I keep walking into

A founder still awake at 11pm answering questions only she can answer. Three assistants already hired. An operations manager already hired. And everything still pooling to her.

The problem was never how hard anyone was working. It was that nothing in the business had been built to absorb a decision.

It is never one broken thing either. It is four symptoms that all trace back to a single decision nobody else is allowed to make, which is why hiring another person rarely fixes it.

How I got here

I started as a virtual assistant nine years ago, because that was the on-ramp and not the destination. A young mother in the Philippines with a laptop and a stubborn belief that remote work could mean real work.

For the first few years I executed. Inbox zero, SOPs, task cards. Work that kept the trains running and never touched the tracks. Then I stopped executing tasks and started designing the thing underneath them.

Lean Six Sigma for finding where the work actually leaks. Business analysis for turning what a founder means into what a team can build. Apps Script, then automation, then AI, for the parts nobody should still be doing by hand.

What I build now is an ecosystem. Systems that hold, documentation that outlives me, people who know exactly what they own, and automation running quietly underneath all of it.

If the operation still needs the architect, the architecture failed.

That is the measure, and it is deliberately unforgiving. The ripple effect is the part I actually work for: founders stop being the last line of defense, the holiday finally gets taken, and Sunday evening stops feeling like a countdown.

The founder is the bottleneck, and the fix is architecture, not hustle. The rule everything else follows from
The method

Seven moves, always in this order.

The order is the method. Most operations work fails because somebody automated step four before anyone had done step one.

01

Map what actually happens

Not what the SOP says happens. I sit with the people doing the work and follow one real job from request to done.

Define · SIPOC
02

Find where the work waits

Work rarely fails while somebody is touching it. It fails in the gaps, sitting in an inbox, waiting on one approval, parked until Tuesday.

Measure
03

Name the real constraint

Usually it is one decision only the founder is allowed to make, hiding behind four symptoms that look like separate problems.

Analyze · Root cause
04

Delete before automating

Half of what a busy team does is a workaround for something that broke two years ago. Automating that just makes it fail faster.

Lean · 80/20
05

Design the decision, not the task

Who decides, on what evidence, and at what threshold they stop asking you. Ownership written down beats availability every time.

RACI
06

Build it so it outlives me

Automation, dashboards, templates, and documentation a new hire can follow in week one without a single call.

Improve
07

Put a number on it, weekly

A scorecard the team reads without me in the room. If a number goes red for two weeks, the system tells you before the client does.

Control · Scorecard

Step one is the step everybody fights me on.

Founders want to hand me the SOP folder and skip ahead. The SOP is a story about how the business is supposed to run, and everyone has quietly agreed to the story. The real process lives somewhere else, in a private Slack thread, a spreadsheet one person maintains, a step three people invented because the official one never worked.

I do not optimize the story. I map the thing that is actually happening, in the actual order it happens, including the embarrassing parts. That map is where every hour I ever give back to a founder comes from.

Skip a step and you get a beautiful system nobody uses. I have been called in to fix those too. Why the order matters
How I diagnose

Interviews alone never find the bottleneck.

The constraint is structurally the last thing in a business able to see itself. The person causing the delay describes it, sincerely, as somebody else’s delay. And almost nobody follows the SOP exactly as written, because people are natural rule-breakers and the written process usually deserves breaking. So I stack four kinds of evidence, and I trust the later ones more than the earlier ones.

  1. 01

    What they say

    Stakeholder interviews

    What it gives me

    Intent. How the business believes it runs, and where the pain is actually felt.

    Where it misleads

    The bottleneck cannot see itself. Necessary evidence, never sufficient.

  2. 02

    What they do

    Shadowing, screen-recorded, timed

    What it gives me

    The real path, workarounds included: the private spreadsheet, the side channel, the step three people invented because the official one never worked.

    Where it misleads

    People behave better while watched, and a week is a small sample. It shows me the route, not the volume.

  3. 03

    What the system logged

    Timestamps out of the tools they already pay for

    What it gives me

    Months of event data from Asana, ClickUp, the CRM, the shared inbox. Volume, rework rate, and exactly how long each step sat untouched. No memory bias.

    Where it misleads

    A log records only what a tool witnessed. Work done in DMs, voice notes and heads never enters the data.

  4. 04

    What the math says

    Operations arithmetic, applied to the other three

    What it gives me

    Which single step sets the ceiling, what removing it is worth in days, and which improvements would have changed nothing at all.

    Where it misleads

    It cannot flatter anyone, which is the point. It is also only ever as honest as the three tiers underneath it.

The SOP is the story a business agreed to tell about itself. The log is what actually happened. I optimize the log. How I decide what to trust
The arithmetic

Then I put a number on it.

Shadowing and logs tell me where to look. Arithmetic tells me what to fix first, what fixing it is worth, and what to leave alone. These are the standard tools of operations management and Lean Six Sigma. Knowing them is not the skill. Measuring honestly enough that they return something worth acting on is the skill.

Little’s Law
Lead time = WIP ÷ Throughput

Lead time is set by how much work is open at once, not by how fast anyone types. Halve the jobs in flight and you halve the wait, with nobody working harder.

Your team is not slow. They are carrying forty open jobs in a system that clears twelve a week.

Process cycle efficiency
PCE = Value-add time ÷ Lead time

What share of a job’s life is actual work. In founder-led operations it lands in low single digits, which means the problem is waiting, not working.

A five-day onboarding contains about ninety minutes of real work. The other four days are queue.

Rolled throughput yield
RTY = FPY₁ × FPY₂ × … × FPYₙ

Quality compounds downward across handoffs. Five steps that each go right ninety percent of the time deliver a clean result fifty-nine percent of the time.

Nobody on the team is failing. The handoffs between them are.

Utilization and queueing
Queue multiplier = U ÷ (1 − U)

Waiting explodes as a person approaches full load. At eighty percent utilized the multiplier is four. At ninety, nine. At ninety-five, nineteen. Slack is not waste, it is capacity.

Your best person is the bottleneck precisely because they are your best person.

Takt time
Takt = Available time ÷ Demand

The rhythm the operation has to hit to keep up. Any step slower than takt sets a hard ceiling on how many clients the business can hold at once.

This is the number that answers whether you can take ten more clients, or only two.

Theory of constraints
Throughput = the slowest step

Improving anything that is not the constraint changes nothing. Identify it, exploit it, subordinate everything else to it, elevate it, then go and find the next one.

Automating the fast part is the most common and most expensive mistake I get called in to undo.

Worked example

One onboarding, measured rather than described.

Six steps, timed from timestamps instead of from the SOP. This is what the first two weeks of an engagement produces, and it is usually the first time a founder sees their own operation as a number.

Teal = actual work  ·  Gray = waiting

1. Contract signed

Touch 5 min
No wait

2. Invoice raised

Touch 10 min
Waits 6 hours

3. Welcome pack sent

Touch 20 min
Waits 1 day

4. Portal access granted

Touch 10 min
Waits 2 days

5. Kickoff scheduled

Touch 15 min
Waits 3 days

6. First session held

Touch 60 min
Waits 1 day

Lead time
7.3 days
Actual work inside it
2.0 hours
Process cycle efficiency
1.1%

What the numbers name

Steps four and five hold sixty-eight percent of the entire lead time and contain twenty-five minutes of work between them. Both sit waiting on one person to do something by hand.

Nothing else on this map is worth touching first. Halving the hands-on time of every other step would save forty-eight minutes on a job that takes seven days.

This is the conversation the interviews never produce.

After the rebuild

Lead time7.3 days → 1.1 days
Hands-on time120 min → 67 min
Steps needing a human6 → 3
Cycle efficiency1.1% → 4.4%

Most of what is left is the client choosing their own kickoff slot. That is customer time, not process waste, and it is where I stop.

Method demonstration. The step times above are representative of a founder-led onboarding process rather than a report on one engagement, so that the arithmetic can be shown in full without disclosing a client’s operation.

The artifacts

The maps I actually hand over.

Both boards are live. Drag them, zoom in, read the notation. A flowchart shows the sequence and the decisions. A swimlane shows the handoffs, and handoffs are where the days disappear.

Inquiry to onboarded client

Fourteen steps, three decision points, two exit paths and one rework loop. Steps that run without a person are drawn differently on purpose.

Open in Miro

Who touches the work, and when

The same process across five lanes. The founder appears exactly once, which is the entire argument in one picture.

Open in Miro
Automation, in detail

Three live builds, opened up.

These are running in a client account right now. Click any of them to zoom and drag through the whole thing. The second and third are the same job crossing two platforms, which is the part that usually breaks and the part nobody shows you.

TRIGGERS · ANY ONE STARTS IT Survey Submitted Survey is any of Customer booked appointment Contact Mode is any of “contact” In calendar is “Path to Publication Enrollment” Customer booked appointment In calendar is “Consultation Call” Contact Mode is any of “contact” Customer booked appointment Contact Mode is any of “contact” In calendar is “Path to Publication” TAG Add Tag PIPELINE Create Or Update Opportunity INTERNAL Internal Notification – SMS INTERNAL Internal Notification – Email CLEANUP Remove from Workflow CLEANUP Remove from Workflow CLEANUP Remove from Workflow TO LEAD Email – Confirmation to lead TO LEAD SMS – asking for confirmation DELAY Wait 1 day TO LEAD Email Pulled out of three other sequences so they cannot be double-messaged. END Click to explore
Appointment booked, end to end. Four triggers, eleven actions, three cleanup steps. GoHighLevel.
GOHIGHLEVEL TRIGGER Survey Submitted Survey is any of “P2P Launch Ap…” OUTBOUNDWebhook POST to Zapier → END Three objects total. Its only job is to hand off cleanly, so the logic lives in one place instead of being duplicated across two tools. Click to explore
Writing survey submitted. One trigger, one webhook, done. GoHighLevel.
ZAPIER → arrives from GoHighLevel TRIGGER WEBHOOKS BY ZAPIERCatch Hook GMAILSend Email END The other half of the same job. GoHighLevel owns the trigger and the contact record; Zapier owns the delivery. Neither tool is asked to do the other’s work. Click to explore
Writing survey received. Catch hook, then Gmail. Zapier.
  1. Three removals, and this is the tell

    In the first build, the contact is pulled out of three other sequences so they cannot be double-messaged. Almost nobody builds this in first. It is the step that stops automation from turning into spam, and it is why that one has run for months without a complaint.

  2. The handoff is the hard part

    Builds two and three are one process split across two tools on purpose. GoHighLevel owns the trigger and the contact record. Zapier owns the delivery. Neither is asked to do the other one’s job, so when something breaks there is never a question about where to look.

  3. Small on purpose

    The second build has three objects in it. That is not a lack of ambition, it is the point. A workflow whose only job is to hand off cleanly can be read in five seconds by whoever inherits it.

Case studies

Three systems, documented end to end.

Real client work, written up the way I write everything up: situation, task, what I built, what changed. Each one opens in a new tab.

Each case study is a live document. If a link ever stops working, tell me and I will send the current one.

The stack

The tools are the easy part. Choosing between them is the job.

I do not lead with a tool list, because a tool list is not a strategy. But you should know what I can pick up without a learning curve, and what I have already built production systems inside. Hover to stop the scroll.

Zapier Make Apps Script GoHighLevel Airtable Google Sheets Kajabi Dubsado Stripe Typeform Calendly HubSpot GitHub Zapier Make Apps Script GoHighLevel Airtable Google Sheets Kajabi Dubsado Stripe Typeform Calendly HubSpot GitHub
Asana ClickUp Notion Miro Slack Google Drive Loom Canva Claude Claude Code Cowork Zoom Squarespace Asana ClickUp Notion Miro Slack Google Drive Loom Canva Claude Claude Code Cowork Zoom Squarespace

Automate

Zapier, Make, Google Apps Script and GoHighLevel, for the work that should never touch a human again.

Run

Asana, ClickUp and Notion, held together by scorecards rather than by check-in meetings.

Deliver

Kajabi, Dubsado, Airtable and Stripe, wired so onboarding and payment stop needing a chaperone.

Think and build

Claude, Cowork, Claude Code and GitHub, which is how I now ship in a day what used to take a developer a fortnight.

You have seen the method. Now see the machinery.

The next page is the deep version: a GoHighLevel pipeline end to end, the Zapier ecosystem behind a live business, an Asana student pipeline with its rules and custom fields, a ClickUp space, a dashboard you can actually click, and the meeting cadence that keeps all of it honest. It is long on purpose. This page was not.

See how I build it