The Spreadsheet Shuffle: Fixing How Translations Actually Get Done

Here is how multilingual content dies in practice, reconstructed from sites we have been called to revive.

An editor exports sixty pages to a spreadsheet, one row per field, and emails it to the translation vendor. Three weeks later the spreadsheet returns, translated. Someone (an intern, a coordinator, whoever drew the short straw) pastes sixty pages’ worth of cells back into the CMS by hand. Along the way come the broken formatting, missed fields, and version mismatches that a launch-week scramble then makes permanent.

Meanwhile the English site moved on. Four of the sixty pages changed while the spreadsheet was abroad. Nobody logged which. The drift our internationalization series has chronicled since 2008 begins its work on day one.

The tragedy is that every step was rational and the whole was ruinous. The good news, which is this post, is that the tooling grew up while nobody was watching. The pipeline that replaces the shuffle exists today on every platform we run. The remaining work, the part no software takes off your hands, is the governance humans still own.

The pipeline, platform by platform

The modern shape, available now. Content flagged for translation inside the CMS. Bundled into jobs. Routed to the translation layer: vendor, machine-plus-human workflow, or internal reviewers. And returned automatically to its proper places with IDs intact, review states tracked throughout.

On Drupal, the TMGMT module suite is the mature version. Translation jobs created from content lists. Connectors to the major vendors and machine services. Review workflows before publishing. And the Drupal 8 multilingual core our platform coverage celebrated gives it the cleanest base it has had.

On WordPress, the multilingual stacks have grown vendor connectors and translation-management integrations of genuinely usable quality, with the configuration care our hreflang audits recommend. One implementation path, verified both directions.

On AEM, the translation framework is enterprise-grade and connector-rich, befitting the tier, with the governance requirements befitting the tier too.

In every case the vendor conversation our architecture guidance prescribed years ago now has its technical half. Good translation firms consume these formats natively. Translation memory, the vendor-side reuse of already-translated segments, compounds savings on every consistent phrase. The round-trip that took three weeks and a short straw takes days and a workflow state.

The shuffle is not a staffing problem. It is a tooling decision that has been made obsolete. Sites still running it are paying artisan prices for error-prone logistics.

The governance humans still own

• Which content localizes. The deliberate-core rule from our architecture series, still the highest-leverage decision. The audience-driven subset maintained well beats the full mirror maintained hopefully. The pipeline makes the chosen scope cheap without making ‘everything’ wise.

• What happens when the source changes. The pipeline’s flags and states finally give the sync question, the one that kills multilingual sites, its instrument. Changed source content marks its translations outdated automatically. The instrument still needs its owner per language, its review cadence, and its monthly sync report read by someone with authority, per the workflow practice. Software flags. Humans decide.

• Who approves the local voice. Machine-assisted workflows and translation memory raise output. The review layer (the bilingual staff member, the regional reviewer) is what keeps the voice and the terms true, per the glossary discipline. Budget the review as the quality gate it is. The sites that skip it discover machine fluency is not institutional fluency, usually via a stakeholder’s relative.

• The year-two line item, still. The pipeline cuts the unit cost of staying current. It does not repeal the operating-cost truth we have repeated since 2009. Translation is a subscription to your own accuracy. The pipeline just made the subscription affordable.

The round-trip, priced before and after

Numbers make the case faster than principles, so the comparison from a recent engagement, anonymized and rounded.

Before: a professional-services firm maintaining four languages by spreadsheet shuffle. Roughly sixty updated pages per quarter. Each cycle ate about three weeks of calendar time. By a careful estimate, it also ate forty staff-hours of export, paste-back, and correction. A standing error backlog (formatting breakage, missed fields, stale pages nobody flagged) that audits routinely measure in the dozens.

After the pipeline: a TMGMT build against their existing vendor, who tellingly preferred it. The same quarterly volume routed as translation jobs. Calendar time under a week. Staff involvement cut to review and approval measured in hours. The error backlog gone by structure, because content returns to its fields rather than through an intern’s clipboard. And translation memory already trimming the vendor invoice as the institutional phrases repeat. Boilerplate, program blurbs, and legal language recur far more than anyone expects. The memory savings compound quarter over quarter.

Build cost: recovered inside the first year on staff-hours alone, before pricing the errors that no longer ship or the pages that no longer drift.

The spreadsheet shuffle survives in so many sites because its costs hide in salaried time and quiet inaccuracy. Put them in a column, and the pipeline stops being a technology decision and becomes a cost calculation.

The stack, assembled

This completes a stack we have been assembling for eight years, and the assembled view is worth a paragraph.

The architecture layer: URL strategy, encoding, the decisions before any translation invoice, 2009. The platform layer: language systems in core, Drupal 6 through 8. The annotation layer: hreflang, audited both directions, 2014. And now the operations layer: the pipeline that keeps the whole structure true over years.

Together they answer the question our international clients actually have. It was never ‘can we launch a multilingual site.’ Anyone can launch one. It was ‘can we still trust it in year three?’ With the stack assembled, yes, at operating costs a mid-market institution can carry. That was not true when we started. It is the quiet good news of the decade’s tooling growth.

The universities recruiting across continents, the manufacturers with regional catalogs, the health systems serving multilingual communities. The machinery is ready, the guidance is written, and the remaining variable is the one it always was. Multilingual sites rot in asymmetry or stay current by workflow. Choose workflow. Staff the owners. Read the sync report. The spreadsheet, at long last, can retire.