A 90-Day Video Localization Rollout Plan

Why First Quarters Fail

Localization programmes that stall in their first ninety days usually fail for one of three reasons, and all three are avoidable.

They start with the hardest content. Someone picks the flagship brand film or the regulated product explainer, because that is the content leadership cares about. It requires the most review, the most approvals, and the most terminology work, all before any process exists. The programme spends its entire initial goodwill on one asset.

Nobody owns quality. The tooling gets procured, content gets processed, and no named person is accountable for whether the output is good. Either it ships unreviewed and someone complains, or it sits waiting for a review that nobody scheduled.

Nothing gets measured. The programme produces localized content, cannot demonstrate an effect, and loses its budget at the next planning cycle to something that can.

This plan is structured to avoid those three failures specifically. It assumes a team of one to three people with other responsibilities, not a dedicated department.

Weeks 1–2: Audit and Scope

The goal is to know what you have and what it is for, before touching any tooling.

Inventory the library. Every video asset, with duration, age, view or usage data, and owner. Most organisations discover they have substantially more content than they thought and that a small fraction of it carries almost all the value.

Sort by risk tier. Regulated or brand-critical, customer-facing, internal, and archive. This determines review requirements later and prevents applying maximal process to everything.

Sort by volatility. Evergreen content that will be useful in two years, versus content that expires. Localization spend on volatile content is largely wasted.

Identify what is localizable. Check whether you have edit masters, audio stems, textless versions, and project files. Some of your library will turn out to be effectively unlocalizable, and knowing that now prevents promising it.

Choose target languages with evidence. Traffic data, existing customer or user distribution, sales pipeline by market, support ticket languages, and where the business has actually committed to operating. Two languages chosen well beat six chosen by intuition.

Name an owner. One person accountable for output quality. Not a committee.

Deliverable: a scored content inventory and a two-language target.

Weeks 3–4: Terminology and Source Preparation

This is the phase that determines everything downstream, and the one most often skipped.

Build the glossary. Product names, feature names, brand terms, technical vocabulary, role titles, and legal or regulatory phrases. For each, decide whether it is translated, retained in the source language, or has a specific approved rendering per market.

Source it from what already exists. If your product interface, documentation, or website is already localized, the terminology decisions have been made. Use them. Building a separate content glossary that contradicts the product is a common and costly mistake.

Get it approved once, properly. Product marketing and legal should sign off the glossary as an artefact, rather than reviewing every asset that uses it.

Assess source content quality. Transcribe two-minute samples from several assets and read the output. This tells you your realistic starting accuracy and identifies content that needs re-recording rather than localizing.

Write the source-content guidance. A one-page note for anyone producing video: leave headroom in text containers, keep text as editable layers, script with pauses, avoid idiom, avoid narration tied precisely to visual events, archive stems and textless versions. This costs nothing and improves everything produced from now on.

Deliverable: an approved glossary and a source-content guideline.

Weeks 5–7: Pilot

Pick deliberately, and pick for learning rather than for impact.

Choose five to eight assets from the internal or moderate-risk tier. Evergreen, reasonably well-produced, with clean audio. Not the flagship. Not anything regulated.

Run them end to end in both target languages, including the steps you intend to keep permanently: transcript review, terminology application, translation review, generation, and delivery.

Track effort honestly. Hours spent on transcript review, terminology additions, translation review, graphics, mixing, and delivery, per asset. This is the number your entire business case depends on and the number most teams never collect.

Log every correction by category. Proper nouns, technical terms, register, grammar, timing, numbers. Recurring categories tell you what to add to the glossary and where the process needs a control.

Get real reviewers. Native speakers who know the subject, ideally from inside the business. A colleague in the target market who uses the product is worth more than a general translator.

Do not skip review to save time. The pilot exists to calibrate the process, and a pilot that skipped review has calibrated nothing.

Deliverable: measured effort per asset, a correction log, and an expanded glossary.

Weeks 8–9: Process and Controls

Turn what the pilot taught you into something repeatable.

Write the workflow down. Who does what, in what order, with what handoffs. A page is enough.

Define review by tier. Full review for customer-facing, sampled review for internal, mechanical checks only for archive. Base the thresholds on your correction log rather than on instinct.

Build the mechanical checks. Reading speed and line length for subtitles, loudness for audio, numeric verification against source, character rendering, and terminology compliance. These are cheap, catch a lot, and free reviewers to focus on judgement.

Set the versioning rule. The reviewed source transcript is authoritative; localized versions derive from it. When the source changes, every language is flagged stale.

Set expiry on volatile content. Anything with pricing, competitive claims, or roadmap content gets an automatic withdrawal date.

Agree the escalation path. What happens when a language pair performs poorly or an asset fails review.

Deliverable: a documented workflow and an automated pre-review check.

Weeks 10–12: Scale and Measure

Process a real batch. Twenty to forty assets, both languages, using the documented process. This tests whether the workflow survives volume rather than attention.

Publish properly. Localized metadata written for how the target market searches, not translated from the source. Separate discoverable destinations per language rather than versions hidden behind a selector. Subtitle tracks and transcripts published alongside.

Instrument from day one. The metrics that matter depend on content type, but pick them now rather than retrofitting: watch time and completion by language against the source version, support ticket volume in target-language markets, conversion or engagement rate by visitor language, and for internal content, completion rates and time-to-competence.

Report the effort numbers. Cost per finished minute per language, review hours per source hour, and how both changed between the pilot and the batch. The trend is the story — review effort should be falling noticeably as terminology matures.

Decide the next phase on evidence. More languages, more content, or deeper review — whichever the data supports.

Deliverable: a published batch, a measurement baseline, and a funded next phase.

What Not to Do in the First Ninety Days

Do not start with six languages. Two languages done properly teaches you everything six would, at a third of the effort, and leaves you able to fix what you learn.

Do not localize the flagship asset first. It carries the most approval overhead and the least process learning.

Do not skip the glossary. Every hour here saves several downstream, and the glossary is the asset that persists when everything else changes.

Do not measure only output. Assets localized is an activity metric. It tells you nothing about whether the programme should continue.

Do not build the whole workflow before testing any of it. The pilot should surface process design, not confirm it.

Do not promise turnaround times before you have measured any. Your pilot numbers will be worse than your steady-state numbers, and committing to either prematurely creates problems.

Do not localize content nobody watches in the source language. Localization multiplies reach; it does not create it. A video with no domestic audience will not acquire a foreign one.

A Working Checklist

Inventory the library and sort by risk tier and volatility.
Verify what is actually localizable — masters, stems, textless versions, project files.
Choose two target languages on evidence, not intuition.
Name one person accountable for output quality.
Build and approve the glossary before processing anything, sourced from existing localized assets.
Issue a one-page source-content guideline to everyone producing video.
Pilot five to eight moderate-risk assets end to end, with real review.
Track effort per asset and log corrections by category.
Automate mechanical checks before human review.
Instrument outcome metrics from the first published batch, not retrospectively.

After the First Ninety Days

The plan above gets a programme launched. What determines whether it survives the following year is a different set of habits.

Keep feeding the glossary. The correction log should update terminology continuously. A programme whose review effort stops falling has almost always stopped doing this.

Add languages one at a time. Each new language surfaces its own rendering, register, and terminology issues, and adding several at once means diagnosing them simultaneously.

Expand content tiers deliberately. Moving from internal content to customer-facing content is a bigger step than it appears, because the review standard and the approval chain both change.

Watch staleness. The proportion of localized assets whose source has changed grows quietly and is the failure that catches most programmes in their second year.

Re-baseline annually. Cost per finished minute and review hours per source hour should both be trending down. If they are flat, something in the process has stopped improving.

Keep the source-content guideline alive. New people join production teams constantly, and the guidance that improves every downstream language needs periodic reissuing rather than a one-time email.

Report outcomes, not activity. The programmes that get funded for a third year are the ones that can show what changed in the business, not how many assets were produced.

Frequently Asked Questions

How many languages should a first phase cover?

Two. Two languages surface essentially every process problem that six would — terminology gaps, review capacity, rendering issues, timing constraints — at a fraction of the effort, and crucially they leave you able to fix what you find. Programmes that launch with six spend the first quarter firefighting rather than learning.

What should the pilot content be?

Five to eight evergreen assets from the internal or moderate-risk tier, with clean source audio. Deliberately not the flagship brand film and not anything regulated. The pilot exists to calibrate effort and surface process gaps, and high-stakes content buries that learning under approval overhead.

What is the single most important thing to do early?

Build the glossary, sourced from wherever your product, documentation, or website has already been localized. It determines output quality more than any other input, it is the artefact that persists as tools and vendors change, and every hour spent on it saves several hours of downstream review.

What should we measure?

Both effort and outcome. On effort: cost per finished minute per language and review hours per source hour, tracked over time — the trend matters more than the level, and review effort should fall noticeably as terminology matures. On outcome: watch time and completion by language against the source, support volume in target markets, and engagement or conversion by visitor language.

Why do localization programmes lose funding?

Because they report activity rather than outcome. Assets localized is not evidence that anything improved. Programmes that instrument outcome metrics from the first published batch, and can show both a falling cost curve and a measurable audience effect, survive planning cycles that activity-reporting programmes do not.

What if we do not have enough content to justify a programme?

Then run a project rather than a programme. Localize a defined set of assets, measure the result, and revisit in six months. The failure mode to avoid is standing up process, tooling, and governance for a volume that does not need it, which produces overhead nobody maintains and a programme that is quietly abandoned.

Who should own localization inside the organisation?

Whoever owns the content it serves, in most cases — marketing for brand content, learning for training, support for help content — with a single person accountable for output quality across them. Placing it in a central function detached from content ownership tends to produce a service nobody prioritises; placing it entirely with content owners produces inconsistent terminology and duplicated effort.