← The Decision Window

Module 2 · 25 min

Seeing the system

Why institutions produce the results they produce.

Watch & reflect · optional

In A World of Systems

Donella Meadows · External learning resource

Video thumbnail: In A World of Systems
YouTube · loads only when playedWatch on YouTube

An introduction to relationships and feedback in everyday systems. Use the example to practise seeing connections, then test your institutional map against present evidence.

Where does an action in your institution feed back into the problem it was meant to solve?

The hook

A ministry is worried about data quality at facility level. The numbers are implausible, too round, too stable, too flattering. The response is a familiar one: a data quality audit, a training workshop, a stricter verification protocol.

Six months later the numbers are still implausible. Often they are worse, and now the facilities are also resentful.

Here is why. Facility staff are assessed against targets. The data they report is the instrument of their own assessment. Reporting honestly when you are behind produces a penalty; reporting favourably produces none, because verification is rare and late. So the numbers drift toward what is safe. Poor numbers reduce trust in the data. Reduced trust means less use of the data for anything except assessment. And data used only for assessment is data people have every reason to manage.

The audit did not fail because it was badly run. It failed because it treated a loop as if it were an event.

What a system is

A system is not an organisational chart and not a piece of software. It is the set of parts and relationships that produces a pattern of behaviour over time. There is an old line in quality improvement, usually credited to Paul Batalden: every system is perfectly designed to get the results it gets. Its exact origin is disputed, but the idea is sound and it is the most useful sentence a director can carry around. If your institution reliably produces late, unused analysis, that outcome is not an accident or a failure of effort. Something is producing it, dependably, and you can find out what.

Four ideas do most of the work.

Stocks and flows

A stock is an accumulation: trained analysts in post, unspent budget, patients on treatment, trust in the data.

A flow is a rate that changes it: hiring and attrition, disbursement and spend, enrolment and loss to follow up.

The mistake is to intervene on a stock when the problem is a flow. You train forty analysts, a one off addition to the stock, while attrition runs at fifteen a year and nothing in the system changed. In three years you are where you started, having spent the budget. The training was not wrong; it was aimed at the wrong variable.

Ask of any intervention: am I topping up a stock, or changing a flow? Stock interventions produce a visible bump and then decay. Flow interventions look like nothing is happening and then compound.

Feedback loops

A reinforcing loop amplifies: more of A produces more of B produces more of A. Growth and collapse both run on reinforcing loops. The data quality example above is one.

A balancing loop stabilises: the system pushes back toward a set point. Every target regime creates balancing loops, and most of them are invisible until you go looking.

When an intervention produces less effect than expected, investigate whether a balancing loop is absorbing it. When it produces far more than expected, you have found a reinforcing loop, and you should check which direction it runs, because they are indifferent to your intentions.

Delay

Delay can contribute to oscillation. One possible pattern is:

  1. You act.
  2. Nothing visible happens, because the effect is lagged.
  3. You conclude the action was too weak, and act harder.
  4. The first action lands.
  5. The second action lands on top of it.
  6. You overshoot, and now you are correcting in the other direction.

This is how commodity stock outs alternate with commodity expiry in the same supply chain, year after year, with competent people at every desk. Nobody is wrong at any individual step. The delay does it.

The practical discipline: before you intervene, write down how long you expect the effect to take and what you will see at the halfway point. Then hold your nerve until that date.

The system has a goal, and it may not be the stated one

Every system behaves as if it is optimising something. Sometimes that matches the mission statement. Frequently it does not. A monitoring system whose real goal is "produce an audit proof record of compliance" will behave very differently from one whose real goal is "help a district manager allocate next month", even if both are described in the same strategy document.

The question to ask: looking only at the behaviour, what does this system appear to be for? The honest answer is often uncomfortable and always informative.

The tool

Choose a persistent institutional problem. Sketch three to six factors and the relationships between them. Mark arrows + when a change tends to move the next factor in the same direction, or − when it tends to move it in the opposite direction. Mark delays with ⏱.

Treat the map as a hypothesis. Discuss it with colleagues, identify evidence for each relationship and note external influences. A map that does not close into a loop may still be useful.

The tool · Sketch one loop

Three to six boxes, arrows between them, an arrow back to the start. Mark each arrow + or −, and any delayed arrow with a clock.

Saved on this device as you type. It comes back in your systems brief at the end.

Onetsa · The Decision Window · Module 2

Sketch one loop

Three to six boxes, arrows between them, an arrow back to the start. Mark each arrow + or −, and any delayed arrow with a clock.

onetsa.co · info@onetsa.co

What to take away

  • Persistent problems are produced by structures, not by people's effort.
  • Ask whether you are topping up a stock or changing a flow.
  • Delay creates oscillation that looks like incompetence and is not.
  • What a system optimises for is visible in its behaviour, not in its strategy document.

Check your understanding

Nothing here is recorded and there is no score. The questions are for you.

1.A ministry trains 40 new data analysts. Three years later the number of analysts in post is roughly what it was before the training. What is the most likely explanation?
2.Facilities report figures that are too round and too favourable. The response is a verification protocol and a data quality workshop. Six months on, nothing has improved. Why?
3.A supply chain alternates between stock outs and expiry, year after year, despite competent management at every level. Which systems concept best explains this?
4.You examine a monitoring system's actual behaviour and conclude it is optimising for an audit proof compliance record rather than for operational decisions. The strategy document says the opposite. Which should you believe?

The Decision Dispatch

Practical thinking on evidence and public decisions.

Short essays on data systems, MEL, DPI and AI in public institutions. Register your interest in the planned monthly dispatch.