The repetition rate is the most concrete number that shows the return on the translation infrastructure. Read on its own, however, it can be misleading.
The rate varies according to the type of document
It is normal for it to be high in documents based on a template and low in original texts. An average taken without breaking the types down gives no information about any type of document.
A rise is not always a gain
If the memory is not updated, old and uncorrected sentences come back as matches. The rate rises, but every match carries a burden of correction.
It is read together with the trend in findings
If the number of findings is rising while the repetition rate is rising too, the quality of the entries in the memory should be questioned. The two numbers acquire meaning together.
The match tiers are separated
An exact match and a fuzzy match carry a different workload. When they are added up in a single rate, the real gain cannot be seen.
A low-percentage match can be expensive
In a sentence that bears little resemblance to the one in the memory, finding the difference can take longer than translating from scratch. Counting this tier as a gain produces a wrong result.
Internal repetition is shown separately
Sentences that occur many times within the document itself are a type of repetition that does not come from the memory, and they are assessed as a separate item.
The rate can be raised through the source text
When consistent writing is applied inside the organisation, the same expressions are written in the same form and the match rate rises naturally. This is a real improvement.
Counter-view: this is a technical detail
It may be thought that the rate concerns not the buyer but the supplier. But the rate is reflected directly in the cost and is the only number that shows the return on the infrastructure investment; it is not a technical detail, it is a commercial indicator.
It is expected to be low in the first year
The memory is still small. An assessment made early comes out in favour of cutting the investment, and the gain is never seen.
Ownership determines the value of the rate
If the memory stays with the supplier, the accumulated gain is lost when the relationship ends. Whether the rate is lasting depends on ownership.
It is followed by direction
Every language direction has its own memory, and their rates are different. A combined number hides the weak direction behind the strong one.
Conflicting entries spoil the rate
When there is more than one rendering for the same sentence, the tool cannot know which one to suggest. A clean-up is what makes the rate useful.
A very high rate is examined
A rate far above what is expected may also show that the same content is being translated again unnecessarily. In that case the real problem is in the volume.
The report is obtained before the job starts
The analysis report is produced before the job starts. A measurement made afterwards describes not the workload but the work that has been done.
Takeaway
The repetition rate shows the return on the infrastructure, but it misleads when it is read on its own. Breaking it down by type of document and by language direction, showing the match tiers separately, reading it together with the trend in findings and following the maintenance of the memory: these four tell apart whether a rising rate is really a gain or an accumulated maintenance debt.