Skip to content
HumanExpert or SmartEdit? A Practical Guide to Choosing the Right Workflow
For direct clients and LSP partners — a clear comparison of Alafranga's two translation workflows
 
The question we are asked most often before a project begins is some version of this: do you use AI, and if so, how?

The honest answer is: it depends on what you are sending us. We run two distinct workflows — HumanExpert and SmartEdit — and the choice between them is not a matter of preference or budget alone. It is a matter of content type, risk profile, and what the document will be used for.

This page explains both workflows, how they differ, and how to determine which one your content requires.


The two workflows at a glance

  HumanExpert   SmartEdit
AI involvement None — no AI at any stage of the workflow   AI draft generated via API (GPT, Claude, or DeepL) inside MemoQ, SDL Trados or any platform such as Smartcat.
Human stages Stage 1: Specialist translator. Stage 2: Independent reviewer working from source — not from translator's output   Stage 1: AI draft generation. Stage 2: Domain specialist full edit — every segment reviewed and corrected before delivery
Number of specialists Two independent human specialists per project, minimum

   One domain specialist reviewer
ISO 17100:2015 compliant ✅ Yes — fully compliant, both stages meet ISO 17100 requirements

   ✅ Yes — human review stage meets ISO 17100 requirements
Audited by Bureau Veritas

   Bureau Veritas
CAT tool MemoQ or SDL Trados — client-specific configuration   MemoQ or SDL Trados — AI connected via API inside the CAT tool, not as standalone
Translation memory ✅ Applied — client-specific TM built and maintained across all projects   ✅ Applied at prompt level before first segment is generated, and enforced inside CAT tool during human review
Glossary / terminology governance ✅ Client-specific glossary applied and maintained — terminology decisions locked across all translators   ✅ Glossary applied at prompt level before AI draft generation — constrains AI output from the first segment
Shared TM for team projects ✅ Real-time shared TM — approved segments immediately available across all translators on the project   ✅ Real-time shared TM — all translators on the same AI engine, same prompt architecture, same reference assets
Client terminology contribution ✅ In-house engineers, legal teams, or technical experts can contribute to shared TM — decisions applied automatically   ✅ Client terminology decisions written to shared TM and applied at prompt level for all future AI draft generation
AI engine selection Not applicable   GPT, Claude, or DeepL — selected based on content type, language pair, and domain
Prompt engineering Not applicable   ✅ Custom prompts built around content type, domain, and client terminology — not generic MT
QA process Independent reviewer checks terminology consistency, numerical accuracy, formatting, and source fidelity — working from source, not from translator's output

   Centralised QA across all translator outputs before delivery — terminology, numbers, formatting
Data privacy Client content processed entirely inside CAT tool — no external API calls

   AI applied via API in controlled environment — client content never used to train public models
NDA availability ✅ Standard for all projects   ✅ Standard for all projects
Speed profile Standard — two full human stages require more time than AI-assisted workflows   Faster on volume — AI draft reduces first-pass time significantly on repetitive or high-volume content
Cost profile Higher — two specialist fees per project   Lower on repetitive and high-volume content — AI draft reduces production cost, CAT discount matrix applied to human review stage
CAT discount matrix ✅ Applied — repetition and fuzzy match discounts applied to human review   ✅ Applied — discounts applied to human review component; no separate per-word AI rate
Best for Legal contracts and agreements, regulatory submissions, compliance documentation, safety-critical machinery and equipment manuals, sworn translation, medical device documentation under regulatory review, patent translation, any content requiring a guarantee of zero AI involvement

   Repetitive technical documentation, large-volume content updates, product documentation where prior translation exists, knowledge base and internal documentation, multilingual programmes requiring cross-language consistency
Sworn translation / notarisation ✅ Supported — fully human workflow required for legal validity   ❌ Not suitable — AI involvement is incompatible with sworn translation requirements
Regulatory submissions ✅ Default workflow — CE technical files, KVKK compliance, BDDK/SPK/EPDK submissions, pharmaceutical regulatory files

   ❌ Not recommended without explicit client risk assessment
Safety-critical documentation ✅ Default workflow — machinery manuals, safety procedures, hazardous materials documentation

   ❌ Not recommended as default — SmartEdit Pro assessed case by case
Language pair coverage All language pairs in Alafranga's active network — for pairs where AI quality does not meet standards, HumanExpert is applied regardless of client workflow preference

   Major European, Middle Eastern, and Asian language pairs where AI draft quality meets Alafranga's accuracy standards
LSP / white-label use ✅ Available under client brand, within client tools, NDA standard   ✅ Available under client brand, within client tools, NDA standard
Project management ]Project Open[ — full project history, client and translator portals, file sharing, role-based access   ]Project Open[ — same infrastructure, AI workflow steps logged per project
Workflow transparency Full — translator and reviewer identities available on request   Full — AI engine used, prompt version, and reviewer identity available on request
When we recommend switching workflows When volume and terminology stability make SmartEdit viable — we will propose this proactively   When content type shifts to regulatory, legal, or safety-critical — we flag this before translation begins