An agency lives on its memories. Years of approved translations for dozens of clients are a quiet asset that makes every new project faster and cheaper. But a messy memory structure turns that asset into a liability, with leaks between clients, duplicated entries and matches nobody trusts.

One memory per client or per subject

The simplest rule is to keep a separate memory for each client, and for large clients, one per subject area. This keeps terminology clean and protects confidentiality.

Mixing clients in one database may increase matches, but it also risks wrong terms and exposure of sensitive text.

Naming conventions

Agree on a clear naming pattern that includes the client, the language pair and the subject. A name like ClientA_EN-DE_Legal tells everyone what the memory holds.

Without naming rules, memories pile up with unclear names, and project managers choose the wrong one under time pressure.

Choosing tools that support the structure

The best translation memory software for an agency should support multiple memories per project, penalties for lower priority memories and easy export. Check these functions before you decide.

Read our overview of the best CAT tools as well, since many tools differ in how they handle memory hierarchies.

Read and write permissions

Not everyone should change every memory. Give translators write access only to the memory for their project, and make reference memories read only.

This prevents accidental pollution and keeps a clear record of who changed what.

Confidentiality between clients

Clients expect their content to stay private. Never reuse a memory across competitors, and review contracts for restrictions on data use.

Document your rules and share them with clients when asked. Clear policies build trust and win contracts.

The confidentiality rules matter most where the content is legally sensitive. A law firm's memory may hold contract wording, certificates and court documents that must never leak into another client's work, which is precisely the environment that legal translation services operate in day to day. For such accounts, agencies typically keep the memory locked to named linguists, log every export and strip identifying details before archiving. Treating legal memories as a separate, higher-security class keeps both the client and the agency safe.

Cleaning and maintenance

Memories need regular care. Remove duplicates, fix known errors and retire entries linked to old product names or outdated terms.

Schedule cleanups each quarter, and assign them to a reviewer who knows the client and the language.

Handling client feedback

When a client corrects a term, update the glossary and the memory at once. Otherwise the same mistake returns in the next project.

Keep a log of changes and share it with all linguists who work on the account, so everyone applies the new rule.

Working with freelancers

Freelancers often use their own tools. Provide them with project packages that include the right memory and glossary, and ask for the updated memory back.

A shared platform makes this simpler, because everything stays in one place and updates arrive automatically.

Analysis and pricing

Match analysis shows how much text is new, repeated or fuzzy. It drives quotes and deadlines, so it needs to be accurate.

Be transparent with clients about discount rules. Clear terms avoid disputes and keep relationships healthy.

Learning from industry data

Research by TAUS and similar groups describes how agencies use data and technology to improve efficiency and quality.

Use these sources to benchmark your own practice and to spot opportunities for improvement.

Backups and export

Back up memories regularly and store copies in a secure place. Export them in an open format so they can move between tools.

A tested backup is worth far more than an untested one, so restore a sample from time to time.

Training project managers

Project managers choose memories for each job, so they need clear guidance. A short checklist helps them pick the right resources every time.

Include steps for confirming language pair, subject, client and permissions before the project starts.

Quality gates

Define when a segment may enter the memory. Many agencies only allow reviewed and approved text, which keeps data reliable.

Unreviewed text can be stored in a separate working memory and promoted after review.

Onboarding a new client

Create the memory and glossary before the first file arrives. Ask the client for past translations, style guides and preferred terms, and load them after a quick quality check.

A strong start means better matches in the very first project and fewer corrections later.

Offboarding and archiving

When a client leaves, archive the memory and follow the retention rules in your contract. Remove access for freelancers who no longer need it.

Keep a record of what was archived and when, so you can answer questions if the client returns.

Reporting on memory value

Track the percentage of reused text by client and language. These figures show which memories deliver savings and which need attention.

Share summaries with account managers, who can use them in conversations about pricing and long term contracts.

Common mistakes to avoid

Common errors include mixing clients, using unclear names and never cleaning old entries. Another is allowing every user to write to every memory.

A simple, documented structure prevents most of these problems and makes onboarding new staff easier.

A practical cleanup plan

List all memories, group them by client and language pair, merge duplicates and archive unused ones. Apply the naming rules and set permissions.

The effort pays back quickly in cleaner matches, faster projects and fewer questions from translators.