Glossary
Straight-through processing
Also: STP
A transaction runs end to end without human intervention, from intake to booking.
The term comes from finance and describes the goal of any automation: no touching in the normal case, handling only for exceptions.
It is measured as a rate: what share of transactions completes without intervention.
For that number to mean anything, the counting method has to be agreed. Does a transaction where somebody only confirmed something count as straight through? Does an automatic retry after a failure count? Is the rate based on incoming or on completed transactions? Without that definition two rates appear in the same company that differ by twenty points, and the discussion turns to the measurement instead of the process.
The rate is also a distribution, not a point. What matters is less the average than the question of which cases systematically drop out: a particular customer, a particular document type, a particular weekday. One exception reason that explains a third of the cases is a piece of work; a hundred different reasons are a hint that the rules are not clean.
The rate rarely goes up through more rules in the code. What usually works better is improving the incoming data: a mandatory field in the form, a validation at the supplier, a structured format instead of a PDF file. Repairing a messy input behind the boundary instead builds special-case logic that wants maintaining forever.
A hundred per cent is almost never the economic optimum, and in some areas it is not legally permitted either. Where a decision has a significant effect on a person, a rejection for instance, the GDPR requires that a human can intervene. That is not an obstacle to automation but an argument for planning the human in at exactly that point and taking them out everywhere else.
The actual purpose of the metric is not the competition for the highest number but steering. It shows whether a change had an effect, and it shows early when an automation is quietly getting worse, for example because a supplier has changed its document format.
How you notice it
- A large share of transactions are uniform.
- Processing time consists mostly of waiting.
- Rules can be stated clearly.
- There is an error list, but no reasons attached to it.
- Nobody can say how high the rate currently is.
Not to be confused with
- Process automation
- The means. Straight-through processing is the result, measured as the share of transactions without intervention.
- Partial automation
- Individual steps run automatically, the transaction still gets touched. Does not count towards the rate, but does save time.
- Human in the loop
- The deliberately planned human at a decision point. Not a failure of straight-through processing but a design decision.
When it fits
- A large share of transactions are uniform and the rules can be stated.
- The volume is large enough that percentage points are measurable in working time.
- The incoming data is structured or can be structured.
When it does not
- When every case needs a judgement: then the groundwork is the goal, not the completion rate.
- For decisions with a significant effect on people, where a human has to be able to intervene.
- When the last few percentage points cost more maintenance than handling the cases manually.
How to approach it
- Agree the counting methodWhat counts as an intervention, what counts as a transaction, which period. Without a definition you get two rates that contradict each other.
- Measure the baseline before buildingOtherwise there is no evidence of the effect later and no basis for the next project.
- Record exception reasons, do not just count themEvery transaction that drops out gets a reason from a fixed list. The distribution shows where the next step lies.
- Start at the intakeMandatory field, validation, a structured format at the sender. Cheaper and more durable than special-case logic behind the boundary.
- Set the line deliberatelyWeigh the effort of the next stage against the handling it saves. Where it no longer carries itself, the rate is finished.
- Watch the rate continuouslyA drop of ten points is a sign that something has changed, usually a format at the sender.
Frequently asked
Which rate is realistic?
Depending on the area 60 to 90 per cent, and that is enough. The last few per cent are exceptions whose automation costs more than handling them manually. The goal is that people only see the cases that need their judgement.
How do you measure it correctly?
Agree the counting method first: what counts as an intervention, does a plain confirmation count, is the rate based on incoming or completed transactions, over which period. Then give every transaction that drops out a reason from a fixed list. Without reasons the rate is a number, with reasons it is a work list.
What is the cheapest way to raise the rate?
At the intake, not in the processing. A mandatory field in the order form, a validation at the supplier, a structured format instead of a PDF: that removes an entire exception reason instead of maintaining it forever in special-case logic.
Are there legal limits?
Yes. For decisions with a significant effect on a person, rejections for instance, the GDPR requires the possibility of human intervention. In practice that does not mean less automation but a deliberately placed point: the human decides where it counts and is out of the way everywhere else.