Feedback waits for the world to move. Feedforward tries to move first.
If a train is approaching an uphill gradient, the control system does not have to wait for speed to fall before adding traction. If a building knows tomorrow afternoon will be hotter, cooling can be prepared before indoor temperature drifts. If a warehouse knows a large inbound wave is scheduled, labour can be positioned before queues form.
Feedforward control uses information about a disturbance, command or future condition to act before the controlled output has developed an error.
It belongs beneath How Control Systems Work and beside How Disturbance Rejection Works. Feedback reacts to consequence. Feedforward reacts to cause.
Why Feedback Alone Can Be Late
A feedback loop needs evidence that the output has moved away from the desired state.
That means the disturbance has already had some effect.
In a slow system this may be fine. In a fast or high-consequence system, even a short delay can matter. A train needs braking before the speed restriction, not after. A power system may schedule generation before demand arrives. A learner preparing for an examination should repair a known prerequisite gap before the final mock exam proves again that it exists.
Feedforward exists because some causes are visible earlier than their consequences.
Feedforward Needs a Disturbance Measurement or Forecast
The controller must know something about what is coming.
- measured incoming load;
- known route gradient;
- scheduled demand;
- weather forecast;
- upstream flow rate;
- known change in product mix;
- planned policy or timetable change.
If the disturbance is invisible and unpredictable, feedback remains essential because the system can only respond after effects become observable.
Feedforward Needs a Model
Knowing that a disturbance is coming is not enough. The controller also needs some relationship between the disturbance and the required action.
If incoming load increases by this amount, how much additional actuator effort is needed? If outside temperature rises, how much extra cooling is appropriate? If demand is forecast to rise, how much capacity should be scheduled?
This relationship can come from physics, empirical data, a statistical model, engineering calibration or operational experience.
The quality of feedforward is therefore limited by model quality.
Feedforward Is Usually Paired With Feedback
A pure feedforward controller assumes its disturbance measurement and model are accurate enough to produce the right action directly.
Reality is rarely that cooperative.
The forecast can be wrong. The plant can age. An unmeasured disturbance can appear. The model can be approximate.
So mature control often combines both:
- feedforward supplies the first anticipatory action;
- feedback corrects the residual error that remains.
This is one of the most useful patterns in control: prediction reduces the size of the error; feedback prevents prediction from becoming blind faith.
Prediction Has a Horizon
Acting earlier is not always better.
A weather forecast several hours ahead may support building control. A forecast weeks ahead may be too uncertain for precise actuator commands. A logistics demand forecast one day ahead may help staffing; one year ahead belongs to capacity planning rather than real-time control.
Feedforward therefore needs a prediction horizon matched to the decision.
Timing Alignment Is Critical
Even a correct feedforward action can fail if it arrives at the wrong time.
If extra cooling begins too early, energy is wasted. If extra traction arrives after the gradient has already slowed the train, the controller has effectively become delayed feedback.
The system must know both disturbance magnitude and disturbance travel time.
Correct information delivered at the wrong phase can create overshoot instead of prevention.
Feedforward Can Overreact to False Forecasts
Anticipatory systems can create errors that the disturbance itself never would have created.
A predicted demand surge fails to arrive, but extra capacity was already activated. A learner is over-remediated for a weakness inferred from one noisy result. A public-service team is moved pre-emptively, leaving another service understaffed.
Feedforward quality therefore depends on forecast calibration and the reversibility of the anticipatory action.
Worked Example: Train Gradient
A train approaches an uphill gradient while following a speed target.
Feedback-only control waits for speed to fall, sees negative error and increases traction.
Feedforward control uses route data to anticipate the extra force required and increases traction as the train reaches the gradient. Feedback then corrects any remaining difference caused by load, adhesion or model error.
The result can be smoother tracking with smaller speed deviation.
Worked Example: Building Cooling
A large building knows occupancy will rise sharply at 8:30 a.m. and external temperature will increase later in the morning.
Instead of waiting for indoor temperature to exceed setpoint, the system can schedule cooling and airflow in anticipation.
If occupancy is lower than forecast, feedback reduces the correction. The two loops cooperate.
Worked Example: Logistics
A warehouse knows that four inbound trailers with unusually high carton counts will arrive between 2 p.m. and 3 p.m.
Feedforward action might include pre-positioning labour, clearing staging space and adjusting dock assignments before the arrival wave hits.
Once the actual arrivals occur, feedback updates the plan using real queue, unload-rate and staffing data.
The operations analogy remains useful because it separates anticipation from correction.
Worked Example: Finance
Financial institutions often prepare for known settlement dates, liquidity needs or predictable transaction peaks before balances become stressed.
This is not identical to a classical control loop, but the feedforward lens clarifies the mechanism: forecast an upcoming demand on the system, prepare resources in advance, then use actual state feedback to correct the forecast error.
The relevant owner remains How Finance Works; the control analogy is a way to inspect timing and anticipation, not a replacement for financial mechanisms.
Worked Example: Learning
A tutor knows that algebraic fractions next week depend on fraction fluency and factorisation.
Instead of waiting for the new topic to fail, the tutor can probe those prerequisites beforehand and repair obvious gaps.
This is a bounded feedforward analogy: known future demand is used to prepare current capability. Feedback remains necessary because the learner’s actual response can differ from prediction.
Worked Example: Public Administration
A policy deadline is expected to trigger a high volume of applications.
Staffing, digital capacity and guidance can be increased before the deadline rather than only after queues become unacceptable.
If actual demand is lower, the organisation should be able to release the temporary capacity. Reversibility makes anticipatory action safer.
Forecast Error Is Part of the Loop
A mature feedforward system records not only whether the output stayed near target, but whether the disturbance forecast was right.
That allows the model itself to improve. Prediction becomes a learning loop rather than a fixed assumption.
This is an important connection between control and evidence: the forecast should be tested against observed return.
Feedforward Checklist
- Which disturbances are observable before they affect the output?
- How accurately can their magnitude be measured or predicted?
- What model links disturbance to required correction?
- How much lead time is available?
- What delay exists between control action and physical effect?
- How reversible is the anticipatory action?
- What feedback loop corrects forecast error?
- How is model accuracy retested over time?
The CivDJ Rotation
- Forward: observe or predict disturbance → calculate anticipatory action → apply before output error → use feedback for residual correction.
- Backward: start from a failure that arrived “too fast” for feedback and ask what upstream signal existed earlier.
- Rotate: compare forecast owner, controller, operator, receiver and finance owner views of the anticipatory action.
What looks efficient to the controller can look wasteful to finance if forecasts are frequently wrong. What looks conservative to finance can look fragile to operations if no reserve is prepared before known peaks.
The Civilisation Lesson
Many of civilisation’s strongest systems do not wait for visible failure. They inspect leading indicators, seasonal patterns, known schedules, upstream conditions and approaching constraints.
But anticipation without feedback becomes dogma. The mature pattern is more disciplined:
Act early when the cause is visible; measure the return anyway, because the world is allowed to disagree with the forecast.
Continue through How Disturbance Rejection Works, How Feedback Works and the master How X Works hub. The next article asks how strongly a controller should react: controller tuning.