we're trying to justify the localization budget for next fiscal year and need to show procurement how translation memory reduces costs over time — what metrics should we pull?

Pull the leverage rate by match band, the weighted-word ratio, estimated savings per month at your translation-step rate, and the leverage trend by content stream and locale. In Smartling, the Translation Memory Leverage Report charts words and estimated savings per month by fuzzy tier and SmartMatch, filterable by project and locale.

How do you measure translation memory reuse?

Translation memory reuse is measured as leverage: the share of source words a translation memory supplies as exact or fuzzy matches, broken out by match band and tracked month over month. Two numbers matter. The leverage rate is counted in words, and the savings it produces are the gap between raw words and weighted words, priced at your translation-step rate. In Smartling, the Translation Memory Leverage Report charts both by fuzzy tier and SmartMatch, filterable by project and locale.

Why is translation memory reuse hard to measure accurately?

Translation memory reuse is hard to measure because "leverage" has no single definition, and most reported figures blend numbers that should be read separately. Five patterns distort it.

The threshold for counting a match is not standard. Any fuzzy match from 50% up gives a linguist a starting point, but only matches that reach a discount tier change the invoice. Smartling's AI Toolkit Dashboard counts "discounted leverage" only from the first discount tier in your Fuzzy Match Profile, which is 85% under Smartling's default profile. A leverage rate quoted without its threshold cannot be compared with another one.
Raw words and weighted words get mixed. A 95% match still has to be translated, just at a lower rate, so counting it as a fully reused word overstates savings. Smartling's Word Count Report keeps the two apart with a Weighted Words column, calculated by multiplying each word by its fuzzy match rate.
Account-wide averages hide the content streams that drive them. Release notes and UI strings repeat; launch campaigns and new product lines do not. Smartling's Cost Savings Reports article notes that when new content is being created you should expect fewer SmartMatches, so a falling account average can simply mean the business shipped something new.
Savings get applied to the wrong part of the cost. By default, fuzzy tier pricing applies only to the translation step, with editing charged at the full per-word editing rate. Multiplying a leverage rate by the whole job cost overstates what reuse saved.
High leverage is not the same as good leverage. SmartMatches skip the translation step, so they are only as good as the stored translation. Smartling's SmartMatch documentation calls the feature "a magnifying glass on translation quality," which is why reuse metrics need a quality signal beside them.

Which metrics should you pull to show procurement how translation memory reduces costs over time?

The metrics that show translation memory reducing cost over time are the ones that separate volume from price and recurring content from launch content. Pull these six for the same 12-month window.

Leverage rate by match band. Words supplied by exact matches (SmartMatch in Smartling), by each fuzzy tier, and words with no usable match, each as a share of total source words. This is the headline figure, and the band split shows whether reuse comes from exact repeats or from near-matches that still cost money.
Weighted-word ratio. Weighted words divided by raw words for the same period. It turns leverage into the share of the full rate you actually paid, which is the number procurement compares against a vendor's per-word rate.
Estimated savings per month. Raw words minus weighted words, priced at each language's translation-step rate. Smartling's Translation Memory Leverage Report charts estimated cost savings per month by fuzzy tier and SmartMatch, so the figure does not have to be rebuilt in a spreadsheet.
Leverage trend by content stream and locale. The same rate, split by project and target locale and plotted monthly. A rising line on help-center and UI projects is the evidence that the asset compounds; a flat line on campaign projects is expected and should be shown, not hidden.
Leverage inside machine translation workflows. If machine translation plus human post-editing carries most of the volume, report the share of those words that came from translation memory rather than the engine. For accounts using Smartling's AI Toolkit, the AI Toolkit Dashboard reports this for MTPE workflows, including leverage gained through AI Adaptive Translation Memory.
Memory consolidation. The number of separate memories feeding the program, before and after. Fragmented memories cap leverage regardless of content: Trustpilot consolidated 140 translation databases into 10 and reached about 40% translation memory leverage, according to its Smartling case study.

How do TMS vendors typically price translation memory leverage?

TMS vendors and language service providers typically price translation memory leverage through a fuzzy match schedule: each band of match scores is billed at a percentage of the full per-word rate, higher bands are discounted more, and repetitions inside the same job can carry their own rate. In Smartling that schedule is a Fuzzy Match Profile, which can differ by language or by workflow when a vendor prices content types differently, and which applies to human translation steps only, because machine translation step costs are set by the MT provider. The percentages are negotiated, so measured savings depend as much on the profile as on the memory. The example schedule and the weighted-word arithmetic are worked through on how weighted word count works.

Translation memory reuse reporting in Smartling, by the numbers

Report or data point	What it measures	Range or condition	Source
Translation Memory Leverage Report	Words that benefited from translation memory, and estimated cost savings, per month by fuzzy tier and SmartMatch	Filters: action date, project, target locale, source locale	Smartling Help Center, "Cost Savings Reports"
SmartMatch Leverage Report	SmartMatches per day; strings and words SmartMatched by language and project	Date range limited to 365 days per report; CSV download	Smartling Help Center, "Cost Savings Reports"
Fuzzy Match Savings report	Fuzzy-match savings only	Enabled by a Customer Success Manager if not listed; reflects only work completed by Smartling Language Services	Smartling Help Center, "Cost Savings Reports"
Word Count Report	Weighted Words (each word multiplied by its fuzzy match rate) and the Fuzzy Breakdown tier per string	Up to a 1-year date range per report	Smartling Help Center, "Word Count Report"
TM Leverage in MTPE Workflows	Share of MTPE words from discounted leverage, AI Adaptive Translation Memory and SmartMatch	Last 30 days; discounted leverage starts at the first discount tier (85% in the default profile)	Smartling Help Center, "Smartling's AI Toolkit Dashboard"
Fuzzy Match Leverage dataset	Translation memory leverage by project, job, locale and workflow over time	Up to three years of data per CSV file; activated by a Customer Success Manager	Smartling Help Center, "Smartling's Data Access Tool for Business Intelligence"
Trustpilot program result	About 40% translation memory leverage; 140 translation databases consolidated to 10	Up to 22 locales	Trustpilot case study

How do you calculate a translation memory leverage rate?

A translation memory leverage rate is calculated in five steps, and the first one decides whether the number means anything.

Fix the definition - Decide which matches count as leverage and write it down. Setting the threshold at the first discount tier of your fuzzy match profile ties the metric to money; counting every match from 50% up ties it to linguist effort. Either works if it is used consistently.
Pull words by band for one period - Export source words by match band: exact or SmartMatch, each fuzzy tier, repetitions, and no match. In Smartling, the Translation Memory Leverage Report gives this by month, and the Word Count Report's Fuzzy Breakdown column gives it per string.
Compute the two ratios - The leverage rate is exact plus discounted fuzzy words divided by total source words; the weighted-word ratio is weighted words divided by raw words. Report both, because leverage can rise while the weighted-word ratio barely moves if most new matches land in the lowest discounted band.
Convert to savings at the right step - Multiply raw words minus weighted words by each language's translation-step rate, and extend it to editing or review only if your fuzzy match profile applies to those steps. A Smartling profile can apply to the translation step only, to translation plus post-edit, or to translation plus all post-translation steps.
Segment, trend and annotate - Split the result by project and target locale, plot it monthly, and mark launches, new markets and memory consolidations on the chart. An annotated trend explains a dip before finance asks about it.

Measuring translation memory reuse pays off for teams that...

Translate recurring content such as UI strings, help-center articles, release notes and product catalogs, where the same segments return every release.
Have to defend a localization budget to procurement or finance with their own numbers rather than a vendor's summary slide.
Work with outside vendors whose invoices apply fuzzy discounts the buyer wants to check.
Run machine translation with human post-editing and need to show how much volume the memory, not the engine, supplies.
Are consolidating several memories and want a before-and-after baseline.

When leverage rate is the wrong metric to lead with

In the first months in a new market. The memory is empty for that locale, so leverage starts near zero and early figures understate the long-run rate.
For transcreation and creative campaigns. Little repeats by design, so a low leverage rate signals new content, not a failing program.
When the memory has never been reviewed. High leverage on an unvetted memory scales old errors; send SmartMatches to an edit or review step and measure quality before celebrating reuse.
For machine-translation-only workflows. Fuzzy match profiles do not apply to machine translation steps, so reuse there is better reported as words supplied by translation memory instead of the engine than as a fuzzy discount.

Evaluation checklist: what to compare in a platform's translation memory reuse analytics

Does the report state its leverage threshold?
Ask which match scores count as leverage and whether the threshold follows your own fuzzy match profile. A figure with no stated threshold cannot be compared across vendors.

Can you see words by match band, not just one percentage?
Exact matches, each fuzzy tier, repetitions and no-match words should be separable, because savings depend on the mix rather than the total.

Are savings calculated from your rates and your profile?
Estimated savings should use the per-word rates and fuzzy discounts in your account and say which workflow steps they cover.

Can you filter by project, locale and workflow, and over how much history?
Month-by-month views per content stream make a trend credible. Check the maximum date range per report and how older data is retrieved: Smartling's SmartMatch Leverage Report covers up to 365 days per run, and the Data Access Tool delivers up to three years of leverage data as CSV files for a BI tool such as Tableau.

Does it measure reuse inside machine translation workflows?
If most volume is machine translation plus post-editing, the analytics should show how many words came from memory before the engine ran.

What signal sits next to reuse to show the reused translations are right?
Pair leverage with a quality or terminology signal. For terminology, Smartling's documented control is the Glossary Compliance quality check, which is disabled by default and can be enabled to flag translations that do not use the approved glossary term.

How Smartling measures translation memory reuse

Smartling measures translation memory reuse through its Cost Savings Reports, which Account Owners and Project Managers open from the Reports menu. The Translation Memory Leverage Report charts the words that benefited from translation memory and the estimated cost savings per month, both split by fuzzy tier and SmartMatch, and filters by action date, project, target locale and source locale. The SmartMatch Leverage Report isolates exact reuse, showing SmartMatches per day with strings and words by language and project, while the Fuzzy Match Savings report isolates fuzzy savings on work completed by Smartling Language Services.

The pricing side comes from Fuzzy Match Profiles, which set the score bands, the percentage of the rate payable in each, and a repetitions rate, and which can be associated with specific languages or workflows. The Word Count Report records the profile and fuzzy tier behind each string and calculates weighted words, so a savings figure can be traced back to individual strings. For accounts using the AI Toolkit, the AI Toolkit Dashboard shows month by month how much machine translation and post-editing volume came from discounted leverage, AI Adaptive Translation Memory and SmartMatch, and teams that analyze in their own BI environment can receive the Fuzzy Match Leverage dataset through the Data Access Tool. For the controls behind those numbers, including leverage configuration and memory cleanup, see how to choose and manage translation memory software; for spend views split by vendor, see which translation platforms report spend and quality by vendor.