A client asks why they cannot have it cheaper
A client has read that machine translation is nearly free and asks why your quote for their product documentation is not lower. Another client, a marketing team, has explicitly said they do not want MT used on their brand copy. A third has confidential legal content that must not go near a public engine. Your PMs make these calls one job at a time, based on experience and whatever they remember about each client's wishes.
Meanwhile your post-editors are paid a flat discounted rate whether the machine output needed a light touch or a near rewrite, and some of them have quietly stopped accepting MTPE jobs.
Why MT decisions are messy
- Client permissions about MT are held in emails and PMs' memories.
- The suitability of content varies within one client: manuals may work well, marketing copy may not.
- Engine quality varies by language pair and subject, and nobody measures it systematically.
- Post-editing effort is not measured, so rates are guesses.
- Confidentiality rules about which engines can be used are not enforced by any system.
What it costs
Jobs machine translated against a client's wishes, which is a serious breach of trust. Post-editors underpaid for heavy edits, who then decline MTPE work or rush it. Jobs quoted as MTPE that turn out to need full translation, eating the margin. And clients who press for MT discounts on content where the output is poor, with no data to explain why that is.
| Question | Today | With the build |
|---|---|---|
| Has the client allowed MT? | Ask the PM | Recorded per client and content type |
| Is this content suitable? | Gut feel | Pre-check plus a sample run |
| Which engine is permitted? | Remembered | Only approved engines offered |
| How much did post-editors change? | Unknown | Edit distance measured per job |
| What should we charge and pay? | Flat discount | Rules based on measured effort |
How we build MT decision support
- Each client's MT permissions are recorded: allowed or not, which content types, which engines, and any confidentiality requirements such as no public engines.
- When a job is created, the content is checked against those permissions. Jobs where MT is not allowed are simply never routed through it.
- For permitted jobs, a sample is run through the engines you use, such as DeepL, Google or a custom engine through your CAT tool or TMS, and a reviewer rates the sample quickly.
- After post-editing, the difference between the machine output and the final text is measured per segment and per job, giving a record of actual effort.
- Over time, effort data by client, content type, language pair and engine builds up, so you can see where MTPE genuinely saves effort and where it does not.
- Your pricing rules for clients and pay rules for post-editors can use effort bands based on that data, rather than a single flat discount. You set the bands.
Whether to use MT on a given client's content is always a decision for you and the client. The build records permissions, provides evidence and enforces the rules you set.
What changes
MT is never used where a client has said no. PMs make decisions with a sample and data behind them. Post-editors are paid in a way that reflects the work, which keeps good ones willing to take MTPE jobs. And when clients ask about MT discounts, you can show them what the data says for their content, which is a better conversation than a flat refusal or a price you cannot sustain.
The effort data also helps you choose between engines for a given pair and subject, based on your own content rather than generic claims.
Is your agency here?
- MT permissions per client are not recorded in a system.
- Post-editors are paid a flat rate regardless of effort.
- Some post-editors now decline MTPE work.
- Clients push for MT discounts and you have no data to respond with.
- You are not sure which engines are allowed for which clients.