In short
- Automating the wrong step can make results worse: it just builds a bigger pile in front of the real limit.
- Stocks are piles, flows are rates, loops are cause-and-effect circles, bottlenecks are limits.
- Donella Meadows, Eliyahu Goldratt and W. Edwards Deming each offer a lens MitHub uses in plain terms.
- MitHub's rule: map the system → find the constraint → automate there → measure → repeat.
Why automation needs systems thinking
AI makes it easy to build automations. That is exactly the danger. When building is cheap, people automate whatever is most annoying or most visible, not whatever is actually limiting results.
For example, imagine a company that automates prospecting with AI and triples the number of leads it generates. Sales stays flat. The reason: the team can only hold 40 discovery calls a week, and it was already booking 40. The new leads sit in the CRM, go cold and waste the budget spent to get them. The automation worked. The system did not improve.
Systems thinking prevents this by asking "what does the whole thing produce, and what limits it?" before asking "what can I automate?"
The four ideas, in plain language
Stocks: what piles up
A stock is anything that accumulates and can be counted at a moment in time: leads in a CRM, open support tickets, unpaid invoices, cash in the bank, trained team members. In her essay on leverage points, systems scientist Donella Meadows uses everyday examples like water behind a dam or money in a bank account.
In a business, stocks tell you where work is waiting. A growing stock of "leads not contacted" is a warning light.
Flows: what moves in and out
A flow is a rate that fills or drains a stock: leads per day coming in, leads per day contacted, invoices paid per week. Meadows compares it to a bathtub: the level only changes through what pours in and what drains out.
This is the most practical idea for automation. If the inflow of leads is 100 a day and the outflow of "contacted" is 60 a day, the stock of uncontacted leads grows by 40 every day, no matter how good your lead generation is.
Feedback loops: what reinforces or corrects
A feedback loop is a circle where a result influences its own cause. Meadows describes two kinds:
- Reinforcing loops amplify: more happy customers bring more referrals, which bring more customers. They also work in reverse: slow follow-up loses deals, which lowers morale, which slows follow-up.
- Balancing loops push toward a target, like a thermostat. A rule that re-contacts a lead until it responds or reaches five attempts is a balancing loop.
Meadows also points out that delays inside loops cause swings and overcorrection. Her example is a shower where hot water arrives late, so you keep turning the tap too far. In business, if you only look at conversion numbers once a quarter, you will overreact the same way.
Her book Thinking in Systems: A Primer, edited by Diana Wright, is the most accessible introduction to these ideas if you want to go deeper.
Bottlenecks: what limits everything
Management thinker Eliyahu Goldratt built the Theory of Constraints around one observation, popularized in his business novel The Goal: every system has a constraint, and the output of that constraint sets the output of the whole organization. The Theory of Constraints Institute uses the image of a chain: strengthening any link other than the weakest is wasted effort.
Goldratt's method is summarized as five focusing steps: identify the constraint, exploit it (get the most from it without big investment), subordinate everything else to it, elevate it (add capacity) and then prevent inertia, meaning go back to the start, because the constraint will move.
For AI automation, this is gold. It tells you where to point your tools.
Deming: fix the system, not the person
W. Edwards Deming, the quality pioneer, adds a lens that matters for teams. Among his 14 Points for Management, he argued that most causes of low quality and low productivity belong to the system, and so lie beyond the power of the individual worker. He also urged leaders to stop relying on inspection at the end and build quality into the process from the start, and to improve the system constantly.
In MitHub's terms: when an automation sends a wrong message, don't just blame whoever wrote the prompt and fix that one message. Ask what in the system allowed it, then add the checkpoint, the data rule or the test that prevents it next time.
The MitHub System Map (one page)
Before you build any automation, fill this in for the workflow you want to improve.
| Question | Your answer |
|---|---|
| Outcome: what does this system produce, per week, as a number? | |
| Stocks: what piles up along the way? (list 3–5) | |
| Flows: what is the in-rate and out-rate of each stock? | |
| Waiting: which stock is growing fastest? | |
| Constraint: which step has the least capacity compared to demand? | |
| Loops: what reinforces growth? What corrects errors? Where are the delays? | |
| Human decisions: where must a person judge? | |
| Signal: how will you see, weekly, whether the system improved? |
Then apply the MitHub rule: map → constraint → automate there → measure → repeat.
Worked example: a sales follow-up system
Imagine a service business with these weekly numbers (hypothetical):
- 300 new leads come in (inflow).
- Sales reps manually research and contact 120 (outflow from "new").
- The stock of uncontacted leads grows by 180 a week.
- Of the leads contacted, 30 book meetings; 10 become customers.
Where is the constraint? Not lead generation: there are more leads than the team can handle. Not closing: reps close a third of meetings. The constraint is first contact, because research and writing each message by hand takes too long.
Exploit it: stop reps from researching low-fit leads. An AI step scores each lead against a written ideal customer profile, so reps only touch the top ones.
Subordinate everything to it: pause spending on more lead generation until first contact catches up. More inflow only grows the pile.
Elevate it: automate enrichment and a first-draft personalized message, with a human approving high-value leads before sending. Tools like n8n often orchestrate this kind of workflow; see n8n for AI automation.
Add a balancing loop: automatic follow-up until a reply or a set number of attempts.
Measure and repeat: if first contact rises to 300 a week, the constraint moves, perhaps to meetings reps can attend. Go back to step one.
This is the logic behind MitHub's approach to revenue engineering: trace the money backwards, find where it gets stuck, and build the system there.
Mistakes to avoid
- Automating the loudest pain, not the constraint. Annoying is not the same as limiting.
- Adding inflow to a clogged system. More leads, tickets or orders just grow the stock.
- Ignoring delays. Measure often enough to see effects before you overcorrect.
- Inspecting at the end. Put checks inside the workflow, not only in a monthly audit.
- Declaring victory once. Constraints move. A good system is improved, not finished.
Where to go next
Systems thinking is what separates a Director from a Designer in MitHub's ladder. To see how systems turn into a business that runs without constant supervision, read Build systems, not just tasks. To practice this method on real revenue processes, start with Diagnose in the Revenue Reverse Engineering faculty.