Running your company as a system, not in reaction

Most companies of 20 to 500 people move forward in reaction: a customer chases, someone digs through their inbox, a meeting is improvised, and things get fixed until the next incident. Running your company as a system means doing the opposite: describing each flow as a loop, from a customer’s request to their payment, making it run without manual relays, measuring it and adjusting it on a fixed schedule. This article from our section on process automation and business bots explains what this shift requires, what automation and AI bring to it, and where people remain essential.

What is a company designed as a system?

American researcher Donella Meadows, a pioneer of systems analysis, defined a system as “an interconnected set of elements that is coherently organized around some purpose”. In a company, the elements are the people, the tools and the files; the interconnections are the information passing from one to another; the purpose is to serve a customer and get paid.

Thinking of the company this way changes the question you ask when facing a problem. In reaction, you look for who should act. As a system, you look for which flow let the problem through, and how to change it so the problem does not come back. The flow becomes a loop when what you measure at the output comes back to change the way work is done at the input.

The best-known version of this loop is the PDCA cycle: plan, do, check, act, then start again. The US health research agency AHRQ describes it as a four-step cycle for carrying out change, solving problems and continuously improving processes. The idea is nothing new. What has changed is the cost of running this loop every day rather than once a year.

What does the difference look like in practice?

Take the order flow in a service company, from the incoming enquiry to payment. The diagram shows it twice: switch from “In reaction” to “As a system” to compare.

In reaction, each post is held by a person, and a file waits for them to check their inbox before it moves: queues form at every step, and nobody sees them until a customer chases. As a system, automations relay information as soon as it arrives, and the dashboard shows what is waiting, how long each file has taken and the step where work piles up: here, quote approval. The durations in the diagram are example values, but the queues are indeed the ones those durations produce.

Why does a company running in reaction wear itself out?

A company running in reaction does not lack energy: it wears itself out spending it. In 2001, Nelson Repenning and John Sterman, researchers at MIT, studied more than a dozen companies, from telecoms to cars, to understand why their improvement programmes failed. Their finding: pushing teams to work harder to hit results locks them into a “capability trap”, where ever more effort is needed to maintain the same level. Time spent firefighting is taken from the time that would prevent the next fires, and most organisations reward last-minute rescues rather than prevention.

A large share of available time also goes into moving information around, as two major surveys show:

Looking for who has the latest version of a quote, chasing to find out where an order stands, copying a figure from one tool to another: this is precisely the relay work an automated flow does in place of people.

What do automation and AI change?

For a long time, running a loop without a human relay required custom-built software, out of reach for a 50-person company. Two things have changed.

First, connecting your tools no longer requires a big project. An orchestrator such as n8n plugs into hundreds of software tools, from the CRM to invoicing, and passes information on as soon as it arrives: the enquiry creates the record, the record notifies the right person, the order triggers the invoice. Our article Connecting your business tools: enter data only once details the ways to do it.

Second, AI can read what no rule could handle: a freely written email, an attachment whose layout changes, an ambiguous request. And it keeps getting cheaper: according to Stanford University’s AI Index 2025, the price of a GPT-3.5-level answer fell more than 280-fold between November 2022 and October 2024.

That does not mean you need AI everywhere. In a well-built loop, most relays are simple rules, faster, cheaper and always identical; AI only comes in at the steps where language needs reading or writing. Our guide Automation without AI: when a rule is enough helps you sort this out, step by step.

Where do people remain essential?

Thinking in systems does not take people out of the loop: it puts them back in the right place. Statistician W. Edwards Deming, a leading figure of continuous improvement, estimated from experience that about 94% of troubles and possibilities for improvement belong to the system, and therefore to management, and only 6% to special causes. It is an estimate, not a measurement, but it says the essential thing: when a file drags, looking for someone to blame helps less than looking at the flow.

In an automated loop, people keep three roles:

• judging what commits the company: an out-of-the-ordinary quote, a commercial gesture, a publication;
• handling the exceptions no rule foresaw, and deciding whether a new rule should be written;
• improving the loop itself, based on what the dashboard shows.

One client learned this the hard way: on Palmora Property, a model once published three property developments that did not exist, and they had to be taken off the site. Today, every imported listing waits for human approval before going live. This split of roles is developed in The human role alongside AI: knowing how to judge.

There is a flip side: a person who approves everything quickly becomes the point where work piles up, like quote approval in the diagram above. Spotting that point and clearing it without losing control is the subject of our article on business bottlenecks; seeing it coming is the subject of our guide Dashboards and KPIs: defining them with leadership.

Where to start?

A loop is built one flow at a time:

1. 01Draw the flow.

   Choose one that matters, from a customer’s request to their payment for example, and describe it with the people who run it, as it really happens.

2. 02Measure what is waiting.

   Three figures are enough to start: files in progress, end-to-end lead time and the step where they pile up.

3. 03Automate the relays.

   Hand automations what requires no judgement: copying, passing on, chasing, notifying.

4. 04Review on a fixed schedule.

   A short, regular review decides the next improvement, instead of waiting for the next incident.

The fourth step is the one most often skipped. Our guide to management rituals suggests a rhythm that holds without adding meetings.

The first two steps are the work of our digital and AI maturity audit: in two to four weeks, it delivers a map of your processes as they really run, and a written GO/NOGO for each project. Once the first flow is chosen, our automation offer removes a retyping task in two weeks, with its cost per run.

Frequently asked questions

Sources

1. Donella Meadows, Harvesting One Hundredfold, 1989 (new tab)donellameadows.org
2. A four-step cycle for carrying out change, solving problems and continuously improving processes (new tab)digital.ahrq.gov
3. Repenning and Sterman, California Management Review, 2001 (new tab)mitmgmtfaculty.mit.edu
4. Microsoft’s 2023 Work Trend Index (new tab)microsoft.com
5. Atlassian’s State of Teams 2025 survey (new tab)atlassian.com
6. Fell more than 280-fold between November 2022 and October 2024 (new tab)hai.stanford.edu
7. SPC for Excel, Deming’s Profound Knowledge (new tab)spcforexcel.com