Translation Memory Explained: How It Cuts Translation Costs

By 8 min read

If you have ever received a translation quote with lines like “100% matches”, “95 to 99%” or “repetitions”, you have already met a translation memory. It is the reason your second translation project with an agency often costs less than the first. It is also the reason your shop’s footer says the same thing on every page in every language. This guide explains what a translation memory is, how match rates are calculated, what discounts vendors actually publish and how you can use all of this to pay less for website translation.

What a translation memory is

A translation memory (TM) is a database of sentences you have already translated, stored as pairs: the source sentence and its approved translation. Translation software splits your text into segments (usually sentences or headings) and checks each new segment against that database. If it finds the same or a similar sentence, it offers the stored translation instead of starting from zero.

The concept is old and standardised. The file format most tools use to move a TM from one system to another is TMX (Translation Memory eXchange). Its version 1.4b specification, dated 26 April 2005, was published by the Localization Industry Standards Association (LISA) to describe TM data exchanged “among tools and/or translation vendors”. After LISA closed, the Globalization and Localization Association (GALA) took over the LISA standards and hosts them today.

Why does that matter to you? Because TMX means your translation memory is portable. If you paid an agency to build up thousands of translated segments, you can ask for the TMX file and take it to another vendor or tool.

How matching works

Every segment in a new text gets a match score against the memory. The vocabulary differs slightly between tools, but the logic is the same everywhere.

Exact and context matches

An exact match (100%) means the new segment is identical to one already stored. Phrase defines a 100% match as a segment that exactly matches one stored in the TM.

A context match goes one step further. RWS describes the Trados Studio context match in its developer documentation on TM lookups as an identical segment that also appears in the same sequence as in an earlier document, with the same translation for the preceding segment. Phrase calls this an in-context match and labels it 101% in its analysis documentation. The idea: if the sentence and its neighbours are unchanged, the stored translation is almost certainly right here too.

Fuzzy matches

A fuzzy match is a similar but not identical segment. Think “Free shipping on orders over 50 euros” versus “Free shipping on orders over 80 euros”. The software shows the stored translation and highlights what changed, so the translator only edits the difference.

Each tool sets a lower limit for what counts as useful. In Trados Studio the default minimum match value is 70%. Project managers can raise or lower that value. Phrase states that fuzzy matches below 75% are treated as new words and translated from scratch.

Repetitions

Repetitions are segments that appear more than once inside the same job, even if the memory was empty before. Phrase counts only the first occurrence in its match band and puts every later occurrence in the repetitions bucket. On a website this is a big deal: navigation labels, button text, cookie notices and product attributes repeat constantly.

Match bands at a glance

Analysis reports group segments into bands. The table below uses the bands documented by Phrase, which closely resemble what other CAT (computer assisted translation) tools report.

Band What it means Typical effort
101% (in context) Same sentence, same surrounding context Quick check
100% Same sentence, context not confirmed Check that it fits
Repetitions Repeated inside the current job Translated once, then reused
95 to 99% Tiny difference such as a number or punctuation Small edit
85 to 94% A word or two changed Moderate edit
75 to 84% Noticeably different Substantial edit
50 to 74% Loosely similar Often close to new work
0 to 49% No usable match New translation

Read the table from top to bottom as a falling scale of reuse. The higher a segment sits, the less human time it needs, which is exactly why vendors price the bands differently. Note that Phrase explains the label is the band’s lower bound, so “match85” covers 85 to 94% and not exactly 85%.

Why a translation memory cuts costs

Most professional translation is priced per word. When a CAT tool analyses your files, it produces a word count per band. The vendor then applies a different rate to each band. In Phrase this is called a net rate scheme. Its documentation gives the example that TM 100% matches “can be billed at 20 percent of the normal rate”. The Phrase support article on net rate schemes gives another illustrative setup: 0% for 101% matches, 10% for 100% matches and 100% for matches under 50%.

Those are examples of how the software can be configured, not prices. Real vendors publish their own grids. Translation company 1-StopAsia, for instance, lists a standard grid in which 50 to 74% matches cost the full rate, 75 to 99% matches cost 60%, 100% matches cost 30% and repetitions or context matches cost 30%.

Grids differ a lot between vendors, so treat any percentage you see, including the ones here, as one vendor’s policy. Discounts vary by vendor, language pair and contract.

A worked example

Say a vendor charges EUR 0.10 per new word and uses the 1-StopAsia grid above. You order a translation of 1,000 new words and 1,000 words that are 100% matches. The new words cost EUR 100. The matches cost 1,000 × 0.10 × 30% = EUR 30. Without a memory you would pay EUR 200; with it you pay EUR 130. The numbers are illustrative, but the arithmetic is the same on every quote that uses a grid.

Why translators push back on discounts

Discounted bands are not free work. Translators argue that 100% matches still have to be read in context and saved. A fuzzy match, they say, can take nearly as long as a fresh translation. As early as 2008, translator and author Corinne McKay described three common approaches: no TM discount at all, a discount only for 100% matches and repetitions or a graduated scale down to 75% or even 50%.

This matters for you as a buyer. If you push a grid too hard, good translators decline the job or rush the “cheap” segments. Quality standards also keep a human in the loop: ISO 17100, the international standard for translation services, requires revision by a second person. A TM speeds up work; it does not remove the need for review.

How to get the most out of a translation memory

You do not need to run a CAT tool yourself to profit from one. A few habits make a real difference.

  1. Own your memory. Ask your agency to deliver the TM as a TMX file at the end of each project. Because TMX is an open standard, you can hand it to the next vendor and keep your discounts.
  2. Write consistently in the source language. If your product pages say “Add to cart” in one place and “Add to basket” in another, the memory sees two different segments. Consistent source text produces more 100% matches. Our guide on translating brand terminology covers how a glossary helps here.
  3. Keep templates stable. Every rewording of a shared footer or checkout notice turns an exact match back into a fuzzy one.
  4. Ask for the analysis report. A quote should show word counts per band. If it only shows a total, ask how matches were priced.
  5. Fix errors in the memory, not only on the page. A wrong translation stored in the TM will be reused faithfully next time. When you correct something, make sure the correction flows back into the memory. Our guide to post-translation editing explains a sensible review routine.

Translation memory and machine translation

Translation memory is not limited to human translators. Phrase, for example, scores machine translation output by comparing it with the final post-edited version, using the same calculation as TM matches. Automated website translation tools often combine both: a machine produces the first version and a memory makes sure repeated text is reused consistently.

That is also how heylingo works. You add one script tag to your WordPress, Shopify, Webflow or custom HTML site. AI translation with a translation memory handles 30+ languages, so a repeated button or headline gets the same translation every time. You can correct a translation inline on the live site. Pricing is a flat EUR 20 per 1,000,000 translated characters per month, so there is no match grid to negotiate.

FAQ

Is a translation memory the same as machine translation?

No. A translation memory only stores and retrieves translations that already exist, usually approved by a human. Machine translation generates new text. Many tools use both together, with the memory taking priority for text that has been translated before.

Who owns the translation memory?

That depends on your contract. The TMX standard makes a memory easy to export, so ask your vendor up front whether you will receive the TMX file and put it in writing.

What is a good discount for 100% matches?

There is no industry standard. Published examples range from a Phrase configuration example of 10% of the normal rate to 30% in 1-StopAsia’s grid. Compare the full grid and the review process, not just one band.

Does a translation memory help small websites?

Yes, often more than you would expect. Websites repeat a lot: menus, buttons, legal notices, product specs. Even on a first project those repeated segments count as repetitions. On every update afterwards most of the site comes back as exact matches.

Conclusion

A translation memory is a simple idea with a direct effect on your bill: text you have already paid to translate should not be paid for in full again. Exact matches, context matches and repetitions are where the savings sit, while fuzzy matches below roughly 75% are usually priced as new work. Discount percentages vary widely by vendor, so ask for the analysis report, insist on receiving your TMX file and keep your source text consistent. If you want to estimate the size of your project first, try our website translation cost calculator.

Want consistent translations without negotiating match grids? See heylingo’s pricing.

Sources

JE

Written by

Jermaine builds and runs heylingo single-handedly from Germany and answers every support email personally.

More from the blog

Information pursuant to § 5 DDG (German Digital Services Act).

Jermaine J. Ernst
c/o Block Services
Stuttgarter Str. 106
70736 Fellbach
Germany

Email: info@heylingo.io
VAT ID: DE370296865

Responsible for content pursuant to § 18 (2) MStV: Jermaine J. Ernst, address as above.

Consumer dispute resolution

We are neither willing nor obliged to take part in dispute resolution proceedings before a consumer arbitration board.

Liability for content

As a service provider we are responsible for our own content on these pages under general law, pursuant to § 7 (1) DDG. Under §§ 8 to 10 DDG we are not obliged to monitor third-party information that is transmitted or stored, or to investigate circumstances that indicate illegal activity. Obligations to remove or block the use of information under general law remain unaffected. Liability in this respect is only possible from the moment we become aware of a specific infringement. As soon as we learn of such an infringement, we will remove the content immediately.

This website links to external websites of third parties whose content we cannot influence, so we cannot accept any liability for that content. The respective provider or operator of a linked page is always responsible for its content. The linked pages were checked for possible legal violations when the link was set, and no illegal content was apparent at that time. Permanent monitoring of linked pages is not reasonable without concrete indications of an infringement. If we become aware of an infringement, we will remove the link immediately.

The content and works created by the site operator on these pages are subject to German copyright law. Reproduction, editing, distribution and any kind of use beyond the limits of copyright law require the written consent of the respective author or creator. Where content on this site was not created by the operator, the copyrights of third parties are respected and such content is marked as such. If you nevertheless become aware of a copyright infringement, please let us know. As soon as we learn of an infringement, we will remove the content immediately.

Last updated 10 October 2026

Privacy policy

This privacy policy explains which personal data we process when you visit heylingo.io, contact us or buy heylingo. It also explains how data is handled when one of our customers uses the heylingo translation service on their own website.

Controller

The controller within the meaning of the EU General Data Protection Regulation (GDPR) is:

Jermaine J. Ernst
c/o Block Services
Stuttgarter Str. 106
70736 Fellbach
Germany

Email: info@heylingo.io

Please use this address for any question about data protection and to exercise your rights.

Overview

heylingo.io is a static website. When you browse it:

Personal data is processed only in the cases described below: delivery of the website by our host, images in blog posts, YouTube videos you choose to play, contact by email and the purchase of heylingo through Stripe. A separate section covers the heylingo translation service.

Hosting and server logs

heylingo.io is hosted by Cloudflare, Inc., 101 Townsend St., San Francisco, CA 94107, USA. Each time you open a page, Cloudflare processes the following server log data:

Cloudflare uses this data to deliver the website to you and to protect it from attacks and misuse. We do not use it to identify you or to create user profiles.

The legal basis is Art. 6(1)(f) GDPR. Our legitimate interest is the secure and reliable delivery of our website.

Cloudflare may process data outside the European Union, including in the USA. Cloudflare, Inc. is certified under the EU-U.S. Data Privacy Framework, so transfers to it are based on the adequacy decision of the European Commission (Art. 45 GDPR). Where the Data Privacy Framework does not apply, transfers are based on the EU standard contractual clauses (Art. 46(2)(c) GDPR).

More information: Cloudflare privacy policy.

Blog images and videos

Images

Images in our blog posts are loaded from assets.seobotai.com, an external image host. When you open a blog post, your browser requests these images directly from that host. The host thereby receives your IP address and the technical information that is part of every request, such as the requested file, the address of our website and your browser’s user agent.

The host needs your IP address to deliver the images. We have no influence on how the host processes this data.

The legal basis is Art. 6(1)(f) GDPR. Our legitimate interest is showing images in our blog posts.

YouTube videos

Some blog posts contain YouTube videos. They are not loaded when you open the page. No data is sent to YouTube until you click “play”. Only then is the video loaded from youtube-nocookie.com.

YouTube is operated by Google Ireland Limited, Gordon House, Barrow Street, Dublin 4, Ireland. Once the video is loaded, Google receives at least your IP address and the address of the page. Google may also store information on your device or read it from your device, for example cookies, and may process data about how you use the video. Google may process data in the USA; Google LLC is certified under the EU-U.S. Data Privacy Framework.

The legal basis is your consent, which you give by clicking “play” (Art. 6(1)(a) GDPR and, for storing and reading information on your device, § 25(1) of the German Telecommunications and Digital Services Data Protection Act, TDDDG).

Your consent applies only to the video you start. You can withdraw it at any time with effect for the future: do not start any further videos and close or reload the page. You can delete cookies set by Google in your browser settings.

More information: Google privacy policy.

Contact by email

If you email us, we process your email address, your message and any other information you provide, such as your name or company. We use this data only to answer your request.

If your request relates to a contract with us or to concluding one, the legal basis is Art. 6(1)(b) GDPR. Otherwise the legal basis is Art. 6(1)(f) GDPR; our legitimate interest is answering requests sent to us.

We delete the data once your request has been fully dealt with, unless statutory retention obligations apply (see “Retention”).

Purchasing heylingo via Stripe

You sign up for heylingo through Stripe Checkout, which you reach via a payment link on heylingo.io. Invoices, your payment method and cancellation are managed in the Stripe Customer Portal. There is no user account on heylingo.io.

The payment provider is Stripe Payments Europe, Ltd., 1 Grand Canal Street Lower, Grand Canal Dock, Dublin, D02 H210, Ireland. Stripe processes the data you enter during checkout and in the customer portal. This includes your name, email address, billing address, company name and VAT ID if you provide them, your payment details, and transaction and invoice data.

You enter your payment details directly with Stripe. We do not receive your full card or account details.

We receive from Stripe the data we need to manage your subscription and bill your usage, such as your name, email address, billing details and payment status. To bill you, we report your usage (the number of translated characters) to Stripe.

The legal basis is Art. 6(1)(b) GDPR, because the processing is necessary to conclude and perform the contract. We keep invoices and accounting records to meet our retention obligations under German commercial and tax law. The legal basis for this is Art. 6(1)(c) GDPR.

Stripe also processes some data under its own responsibility, for example to prevent fraud and to meet its legal obligations as a payment provider. Stripe may transfer data to Stripe, Inc. in the USA, which is certified under the EU-U.S. Data Privacy Framework.

More information: Stripe privacy policy.

The heylingo translation service

This section is for our customers and for visitors to websites that use heylingo.

Our role

heylingo makes a customer’s website multilingual. When a customer uses heylingo, we process the content of the customer’s website on the customer’s behalf. For any personal data of visitors to that website, the customer is the controller and we act as a processor (Art. 28 GDPR).

If you visit a website that uses heylingo, please read that website’s privacy policy and send requests about your data to its operator.

Customers can request a data processing agreement under Art. 28 GDPR at info@heylingo.io.

In the visitor’s browser

The customer adds the heylingo script to their website. The script shows a language switcher and picks the visitor’s preferred language. It stores the chosen language in the visitor’s browser (localStorage) so the website appears in that language on later visits. The script sets no cookies.

The entry stays in the browser until the visitor deletes it. Storing it is limited to what is strictly necessary for the language function the visitor uses (§ 25(2) no. 2 TDDDG).

Translation

To translate a page, the script sends the page’s text to our translation service. Because the visitor’s browser connects to the service, the service also receives the visitor’s IP address and the technical information that is part of every request.

Page text usually contains no personal data. If a page does contain personal data, for example names on a team page or text shown only to a logged-in user, this text is sent to us as well. The customer decides on which pages the script runs.

Text that is translated for the first time is translated by AI. For this we use a large language model of Anthropic PBC, USA, through its API, as a sub-processor. Transfers to Anthropic are based on the EU standard contractual clauses (Art. 46(2)(c) GDPR).

Each translation is stored in a translation memory for that website. Later visitors receive the stored translation, so the same text is not sent to the AI provider again.

The translation service and the translation memories run on servers of netcup GmbH in Germany.

Inline editing

Customers can correct translations directly on their live website. This edit mode can only be used by the site owner. Corrected sentences are stored in the translation memory and are never re-translated.

Deletion

When a customer cancels, the script stops translating and the website shows its original language. The website’s translation memory, including manual corrections, is deleted 30 days after the cancellation takes effect.

Retention

We keep personal data only as long as it is needed for the purpose it was collected for, unless a statutory retention obligation applies. While such an obligation applies, we use the data only for that purpose.

Your rights

You have the following rights regarding your personal data:

To exercise your rights, email info@heylingo.io. If your request concerns a website that uses heylingo, please contact the operator of that website first, as it is the controller. We support our customers in answering such requests.

Right to object

Where we process your data on the basis of Art. 6(1)(f) GDPR, you can object to the processing at any time on grounds relating to your particular situation (Art. 21 GDPR). We will then stop processing your data, unless we can demonstrate compelling legitimate grounds that override your interests, rights and freedoms, or the processing serves to establish, exercise or defend legal claims.

Right to lodge a complaint

You can lodge a complaint with a data protection supervisory authority (Art. 77 GDPR), in particular in the EU member state of your habitual residence, your place of work or the place of the alleged infringement. The authority responsible for us is Der Landesbeauftragte für den Datenschutz und die Informationsfreiheit Baden-Württemberg (the State Commissioner for Data Protection and Freedom of Information of Baden-Württemberg).

No automated decision-making

We do not use automated decision-making, including profiling, within the meaning of Art. 22 GDPR. The AI used by heylingo translates text; it does not make decisions about people.

Changes to this policy

We update this policy when our services or the legal requirements change. The current version is always available on heylingo.io. The date of the last update is shown at the end of this policy.

Last updated 10 October 2026

Terms of service

These terms of service govern the use of heylingo, a service that makes websites multilingual. heylingo is offered to businesses only.

1. Scope and parties

These terms apply to all contracts for the use of heylingo between Jermaine J. Ernst, c/o Block Services, Stuttgarter Str. 106, 70736 Fellbach, Germany, email info@heylingo.io (“we”, “us”) and our customers (“you”).

heylingo is offered exclusively to entrepreneurs (Unternehmer) within the meaning of § 14 BGB. These are natural or legal persons or partnerships with legal capacity who conclude the contract in the exercise of their commercial or independent professional activity. Consumers are not our target group. By concluding a contract, you confirm that you are acting as an entrepreneur.

Your own terms and conditions do not apply, even if we do not expressly object to them, unless we have agreed to them in text form.

2. The service

heylingo makes your website multilingual. You add one script tag to your website. The script shows a language switcher, selects the visitor’s preferred language, stores the chosen language in the visitor’s browser, sends the page’s text to our translation service and replaces it with the translation.

Text that has not been translated before is translated by AI (a large language model) and stored in a translation memory for your website. Later page views receive the stored translation.

All languages supported by heylingo are included. The currently supported languages are listed on heylingo.io.

You can correct translations directly on your live website, in an edit mode that only you as the site owner can use. Sentences you have edited manually are never re-translated.

There is no user account or dashboard on heylingo.io. Invoices, your payment method and cancellation are managed in the Stripe Customer Portal.

For the term of the contract, you receive a non-exclusive, non-transferable right to use the heylingo script on your website. All other rights to heylingo remain with us.

We may use subcontractors to provide the service, in particular for hosting and AI translation. We may develop heylingo further and change its features, provided the core function described in this section is preserved and the change is reasonable for you.

3. Conclusion of the contract

The description of heylingo on heylingo.io is not a binding offer.

You sign up through Stripe Checkout, which you reach via the payment link on heylingo.io. By completing the checkout, you make a binding offer to conclude the contract. We accept it by activating your subscription, which normally happens immediately after checkout.

The contract language is English. These terms can be viewed, saved and printed on heylingo.io at any time.

4. Prices and billing

heylingo has one pricing model: pay per use. The price is a monthly price per one million translated characters, as listed on heylingo.io (currently EUR 20 per 1,000,000 characters per month). It includes all supported languages.

The characters counted for a month are the characters of the translated text that heylingo keeps for your website in its translation memory, across all target languages. Page views and the number of visitors do not affect the price.

Billing runs through Stripe as a metered subscription. Your usage is billed monthly in arrears and charged to the payment method you stored with Stripe. Invoices are available in the Stripe Customer Portal.

All prices are net prices. VAT is added where applicable.

We announce price increases by email at least 30 days before they take effect. A new price applies from the first billing month after it takes effect. As you can cancel at any time, you can avoid a price increase by cancelling before it takes effect.

If a payment fails, we may suspend the service after notifying you until the outstanding amount has been paid. During a suspension, your website shows its original language.

5. Your obligations

You must hold all rights needed to have the content of your website translated, stored and displayed, including in translated form.

You grant us a non-exclusive right to process, translate, store and deliver your content and its translations, to the extent necessary to provide heylingo. This includes the right to have our subcontractors do so. The right is limited to the term of the contract and the deletion period in section 8.

You must not use heylingo for unlawful content, in particular content that infringes the rights of third parties or violates criminal law.

You are responsible for integrating the script correctly, as described by us, and only on websites that you operate or are authorised to operate.

You must not misuse heylingo. In particular, you must not:

If a third party asserts claims against us because of content on your website or because you breached these terms, you indemnify us against these claims, including the reasonable costs of legal defence. This does not apply if you are not responsible for the breach.

6. AI translations

Translations are produced automatically by AI. We do not review them before they are shown on your website.

Machine translations may contain errors, omissions or wording that does not fit the context. We provide machine translation, not a reviewed or certified translation.

You should review translations, especially where errors can have legal or financial consequences, such as legal texts, prices, product information and safety notices. You can correct any translation in the edit mode on your website.

You are responsible for all content published on your website, including its translations.

7. Availability

We aim to keep heylingo available and working, but we do not guarantee a specific availability or uptime. Interruptions can occur, for example due to maintenance, updates or disruptions at our subcontractors.

If heylingo is unavailable, your website shows its original language. As you pay only for translated characters, no charges arise for text that is not translated.

8. Term and cancellation

The contract starts when it is concluded and runs for an indefinite period.

You can cancel at any time in the Stripe Customer Portal or by email to info@heylingo.io. Usage up to the effective date of cancellation is billed.

We can cancel the contract with 30 days’ notice by email.

The right of both parties to terminate for good cause without notice remains unaffected. Good cause for us exists in particular if you are in default of payment despite a reminder, use heylingo for unlawful content or seriously breach your obligations under section 5. In these cases, we may instead suspend the service for your website.

After the contract ends, the script stops translating and your website shows its original language. We delete your website’s translation memory, including manual corrections, 30 days after the contract ends. We recommend removing the script tag from your website.

9. Data protection

We process personal data in accordance with the GDPR and our privacy policy.

When you use heylingo, we process the content of your website on your behalf. For any personal data of visitors to your website, you are the controller and we act as your processor. A data processing agreement under Art. 28 GDPR is available on request at info@heylingo.io.

You are responsible for meeting your own data protection obligations towards the visitors of your website, in particular for informing them about the use of heylingo in your privacy policy.

10. Liability

We are liable without limitation for intent and gross negligence, for injury to life, body or health, where we have given a guarantee, and under the German Product Liability Act (Produkthaftungsgesetz).

For slight negligence, we are liable only for breach of an essential contractual obligation. This is an obligation whose fulfilment makes the proper performance of the contract possible in the first place and on whose fulfilment you may regularly rely. In that case, our liability is limited to the damage that is typical for this type of contract and foreseeable when the contract was concluded.

Otherwise, liability for slight negligence is excluded. Strict liability for defects that already existed when the contract was concluded (§ 536a(1) BGB) is also excluded.

These limitations also apply in favour of our vicarious agents and subcontractors.

11. Changes to these terms

We may change these terms for a valid reason, in particular because of changes in the law or case law, or changes to heylingo. We will notify you of changes by email at least 30 days before they take effect.

You can object to the changes within this period. If you do not object, the changed terms are deemed accepted. We will point this out specifically in our notice. If you object, the contract continues on the previous terms, and we may cancel it in accordance with section 8.

This procedure does not apply to changes to the main obligations of either party. Price changes are governed by section 4, changes to the service by section 2.

12. Final provisions

German law applies, excluding the UN Convention on Contracts for the International Sale of Goods (CISG).

If you are a merchant (Kaufmann), a legal entity under public law or a special fund under public law, or if you have no general place of jurisdiction in Germany, the exclusive place of jurisdiction for all disputes arising from this contract is the competent court for Fellbach, Germany. We may also bring an action at your general place of jurisdiction.

If any provision of these terms is or becomes invalid, the remaining provisions remain valid. The invalid provision is replaced by the statutory provisions.

Last updated 10 October 2026