The reorg memo you have been avoiding writing for eighteen months is finally due, and it is not a memo about headcount. Your localization function ships forty languages across product, marketing, legal, and support, and for two decades it has run on a single, unquestioned shape: every job flows translate, then edit, then proofread, three humans in a line, each one starting where the last left off, each one paid by the word. That shape was not a strategy. It was the water the industry swam in. Then the engine arrived and pre-populated every segment before your first human opened the file, and the line did not break so much as quietly stop making sense: your translators were now editing a machine draft and calling it translation, your editors were re-reading content the machine had already made fluent, your proofreaders were polishing prose that was grammatically flawless and occasionally, silently, meant the opposite of the source. Throughput tripled and nobody could tell you whether quality had held, because the org was still built to catch the errors humans make (typos, awkwardness, omission through fatigue) and structurally blind to the error the machine makes (a confident, fluent, perfectly grammatical lie about a dosage or an indemnity). This lesson is about the memo. Not a tweak to the line, but a redesign of the function around a single organizing principle: people own quality and judgment, the machine owns throughput, by design and not by accident. We are going to take a real enterprise localization org from its legacy translate-edit-proofread shape to an MT-first, quality-owned operating model, name every role and handoff that changes, quantify the transition risks that sink these redesigns, and leave you with a worked org chart you could defend to your CFO and a certifying auditor in the same meeting.

The Line That Stopped Making Sense

Start by naming the two shapes precisely, because the entire redesign is the migration from one to the other and vague language here produces vague org charts. TEP is the legacy operating model: translate, edit, proofread, a linear sequence of three human roles where a translator renders the source from scratch, a bilingual editor (a reviser) compares the translation against the source and corrects it, and a monolingual proofreader polishes the target for surface errors. Each stage assumes the stage before it produced honest human work, each is priced per word, and the whole line is optimized to catch the failure modes of a tired human translator. An operating model is simply the arrangement of roles, handoffs, decision rights, and controls through which the function converts a source file into a shipped translation: who does what, in what order, who decides, and who is accountable when something ships wrong. MT-first means the operating model assumes machine translation (MT, and increasingly a large language model, LLM, doing the same job with more fluency and more confident errors) produces the first draft of every eligible segment before any human touches the file, so the human is never translating from a blank target and is always, structurally, editing a machine.

The failure of TEP in an MT-first world is not that it is slow. It is that its controls point at the wrong enemy. TEP was designed when the dangerous errors were the ones a human makes: a mistranslation from imperfect comprehension, an omission from fatigue, an inconsistency from a long file. Three sequential humans catch those well, because a second and third pair of fresh eyes is exactly the remedy for human error. But the machine does not make human errors. The machine produces machine-translation post-editing (MTPE, the workflow where a human edits machine output rather than translating from scratch) input that is fluent first and accurate second, and its signature failure is the silent critical error: a flipped negation, a moved decimal, an inverted indemnity, rendered in prose so clean that a second and third fresh reader skims past it precisely because it reads perfectly. Three sequential polish passes over fluent output do not catch a fluent lie. They validate it three times.

TEP was built to catch the errors a tired human makes. The machine does not make those errors. It makes fluent, confident, invisible ones, and a line of sequential polishers validates them rather than catching them.

So the redesign is not a productivity project. It is a re-pointing of the function's controls at the actual enemy. And that re-pointing has a structural consequence that most leaders underestimate: it changes what humans are for. In TEP, the human is a producer of words, and the org is a word factory with a quality inspection at the end. In the MT-first model, the machine is the producer of words, and the human is the owner of everything the machine cannot own: the judgment about what is correct against the source, what is on-brand, what is safe, what must never be machine-translated at all, and the accountable sign-off that a raw engine can never provide. The reorg memo, done right, moves your people up that value chain rather than out the door, and the rest of this lesson is how the org chart, the handoffs, and the change management actually execute that move.

The Organizing Principle: Humans on Quality, Machine on Throughput

Before we draw a single box, we need the principle the whole design serves, stated tightly enough to adjudicate the hard cases. The principle is a division of labor by comparative advantage: the machine owns throughput, the human owns quality and judgment, and the boundary between them is drawn by consequence, not by convenience. The machine is superb at volume, speed, and producing fluent first drafts at a marginal cost approaching zero. It is structurally incapable of knowing when its fluent output is wrong, because it optimizes for plausibility, not truth. The human is slow and expensive at producing volume and uniquely capable at the thing the machine cannot do: verifying the output against the source and the approved terminology, deciding what is safe to ship, and standing behind that decision with professional accountability.

Notice what this principle does to the old org's core metric. TEP measured people in words per day, because in a word factory the producer's value is words produced. The moment the machine produces the words, words per day measures the wrong thing entirely: it measures how fast a human can validate machine output, which incentivizes exactly the fluent-error-skimming the model exploits. The redesigned org measures its people on the quality they own, not the words they process: critical errors caught per file, terminology conformance, the defensibility of the quality record, the correctness of the risk-tier routing. A translator who cleared 8,000 words of light post-edit today did not necessarily create more value than the one who cleared 1,500 words of a contraindications section and caught two flipped negations. The redesign has to make the org chart, the incentives, and the titles reflect that, or it will build MT-first plumbing and keep rewarding people for behaving like a word factory, which is how these transformations fail while looking like they succeeded.

Where Humans Concentrate in the New Shape

If the machine owns first-draft throughput, the humans do not spread thinly across every segment. They concentrate at five points where judgment is load-bearing and the machine is structurally blind. These five are the spine of every role in the redesigned org, so name them before you assign anyone to a box:

• Risk-tiering at intake. The judgment of what a fluent error in this content would actually do to someone, and therefore what the machine may touch and how hard a human must look. A marketing string, a drug label, and an indemnity clause are not the same content, and the machine renders all three with identical, uninformative confidence. Someone human decides the tier, and that decision is the first control in the whole model.
• The quality gate. The severity-scored evaluation against an error typology (the MQM, Multidimensional Quality Metrics, framework formalized for translation output by ISO 5060:2024) that turns "looks fine" into a defensible go/no-go, where a single Critical error fails the file no matter how clean the rest reads.
• Terminology. Ownership of the approved terms the engine is instructed toward and drifts away from, because the machine prefers the common synonym and terminology is about identity and conformance, not plausibility.
• Transcreation and high-judgment content. Brand, culture, intent, and register, where a literal-but-fluent rendering kills a campaign and the required output is not a translation of the words but a re-creation of the effect.
• Exceptions. The segments and files the machine got confidently wrong, the ambiguous source, the novel content with no TM leverage, the MT-forbidden material. The exceptions are where the human's slow, expensive attention earns its cost, and the model's job is to route the exceptions to the human and let the machine handle the routine.

Everything the machine does well (the fluent first draft of routine, lower-consequence, high-volume content) it does at scale, and the humans stop competing with it there. Everything the machine cannot do (the five concentration points above) becomes the explicit, named, measured work of the function. The redesign is, at bottom, the act of building an org chart around those five concentration points instead of around the three sequential production stages of TEP.

The machine owns throughput; the human owns quality and judgment. The boundary is drawn by consequence, and the humans concentrate at five points the machine is structurally blind to: risk-tiering, the quality gate, terminology, transcreation, and exceptions.

The New Workflow Shape: From a Line to a Tiered Funnel

TEP is a line: every file enters at translate and exits at proofread, and every file gets the same three sequential passes regardless of what it is. The redesigned workflow is not a line at all. It is a funnel that sorts by consequence at the top and then routes each stream to the treatment its risk earns, so the function spends its expensive human attention where a fluent error is catastrophic and lets the machine run where a fluent error is recoverable. Walk the shape, because the org chart is downstream of it.

The Shape in Motion

At the top of the funnel sits risk-tiered intake, and it is the single most important structural change from TEP, because it is the decision TEP never made: TEP treated every file as the same kind of thing. The redesigned intake classifies content by consequence into tiers before any post-editing begins. A workable three-tier scheme, which you will map to your own content, runs like this. Tier 3, low consequence: marketing microcopy, UI strings, internal knowledge-base content, support macros, where a fluent error costs polish, not a life or a lawsuit. Tier 2, high consequence: user-facing documentation, help content with instructions a user acts on, product content where an error misleads but does not kill. Tier 1, critical consequence: regulated, life-safety, legal, and financial content, drug labels, contraindications, indemnity clauses, financial disclosures, where a fluent error can injure, sue, or ruin. And a fourth designation that sits outside the tiers: MT-forbidden, content the machine is not permitted to touch at all, routed to full human translation because the liability or the confidentiality forbids a machine draft.

Each tier flows to a different treatment, and this is where the funnel replaces the line:

• Tier 3 flows to grounded MT plus light post-editing: the engine drafts, a post-editor does a fast source-check and fixes what breaks meaning, and the file is scored on a sampling basis rather than segment by segment. The machine owns nearly all of this stream's throughput.
• Tier 2 flows to grounded MT plus full post-editing: the engine drafts, a post-editor works every segment against the source, and the file passes a full severity-scored gate before delivery.
• Tier 1 flows to grounded MT plus full post-editing by a competent domain linguist, or full human translation, plus a mandatory independent quality-gate pass by a second person, plus enforced terminology and locale checks, plus the full quality record. Nothing in this stream ships on a single human's say-so.
• MT-forbidden flows to full human translation with the standard revision, the machine explicitly out of the loop, routed by rule at intake.

The transcreation stream runs parallel to all of this: content flagged at intake as brand-critical or campaign-critical does not enter the post-editing funnel at all, because the required output is a re-creation of effect, not an edit of a draft. It goes to a transcreator, who may use the machine as an ideation tool but owns the creative judgment the funnel cannot mechanize.

The reason this shape is faster and safer than TEP is the same reason risk-tiering matters: the budget you save by running Tier 3 lightly is the budget you spend running Tier 1 deliberately. TEP spent the same three passes on a throwaway UI string and a contraindication, which meant it was simultaneously over-serving the trivial and under-protecting the lethal. The funnel fixes both by making consequence, not word count, the thing that decides how much human attention a segment earns.

The old model was a line where every file got the same three passes. The new model is a funnel that sorts by consequence at intake and routes each tier to the treatment its risk earns, so speed and defensibility fund each other instead of competing.

The Gate Is a Control, Not a Fourth Stage

A tempting mistake is to redraw TEP as translate-edit-proofread-gate, bolting a quality gate onto the end of the old line as a fourth sequential stage. That is not the redesign; it is TEP with an extra inspection. The gate in the new model is not a stage a file passes through on a conveyor. It is a control with decision rights: a person who owns the severity-scored evaluation and holds the authority to fail a file and send it back, and whose judgment on a single Critical error overrides how clean the rest of the file looks. The distinction is organizational, not semantic. A stage is a place work happens; a control is a place a decision with teeth gets made and owned. The redesign gives the gate an owner with the authority to say no, and that authority, not the position of the box in the diagram, is what makes the gate real.

How Roles and Handoffs Change

Now the part the reorg memo actually turns on: what happens to the three TEP roles and the handoffs between them. This is where transformations succeed or fail, because you cannot keep the old roles, add MT, and call it done. The roles change in kind, not just in tooling.

Translator Becomes Post-Editor and Risk Owner

The translator no longer produces words from a blank target. In the new model the translator is a post-editor who works machine drafts against the source, the termbase, and the memory, and the critical change is not that they edit instead of translate. It is that their value moved from words produced to errors caught and judgment applied. The revised ISO 18587, the post-editing standard, in DIS ballot with publication targeted for late 2025 into 2026, expanded to cover AI and LLM "non-human translation output," insists this person hold the same full professional-translator competence as a human translator, precisely because catching a fluent error in high-stakes content is a translator's judgment and not a button-pusher's reflex. So the role does not deskill. It concentrates: the post-editor spends their competence at the high-consequence tiers where a flip is catastrophic, reads Tier 3 fast, reads Tier 1 slow, and their throughput is expected to rise (a hybrid workflow lifts a linguist from roughly 2,000 words a day to 5,000 or more) precisely because the machine carries the routine and the human carries the judgment.

The Editor Splits Into Evaluator and Terminologist

The TEP editor's job, comparing target to source and correcting, does not vanish, but it splits along the fault line the machine exposed. Part of it collapses into the post-editor's own pass, because in MTPE the person editing the machine draft is already doing the source comparison the editor used to do downstream; a redundant second full source-comparison of fluent output is exactly the pass that validates fluent errors rather than catching them. The rest of the editor's value migrates up into two specialized owners the old line did not have. The quality evaluator owns the severity-scored gate: they do not re-edit the file, they score it against the MQM/ISO 5060 typology, assign category and severity to each error, and hold the authority to fail it on a single Critical. The terminologist owns the approved terms across the whole pipeline: they curate the termbase the engine is grounded on, they own the enforcement sweep that confirms the approved term held across every segment, and they adjudicate the term decisions the post-editor flags. Both are net-new decision-owning roles carved out of the old editor function, and both are quality-owning, not word-producing.

The Proofreader Role Thins, and Its Talent Moves to Transcreation and Exceptions

The monolingual surface-polish proofreading pass is the role most reduced by the redesign, and honesty about that is part of good change management. The machine produces fluent, grammatical surface prose for free; a dedicated human pass whose job is catching surface infelicity in already-fluent output is the lowest-value activity in the new model, and pretending otherwise wastes the transformation. But the people in proofreading are not surplus; their linguistic sensitivity is exactly what the two highest-judgment concentration points need. The redesign moves that talent into transcreation (the brand and campaign work the funnel routes around post-editing entirely, where the machine cannot produce the effect) and into exceptions handling (the ambiguous source, the no-leverage novel content, the segments the machine got confidently wrong). The role that thins is the surface-polish pass; the people move up into judgment work, which is the whole thesis of the redesign expressed as a career path.

The Handoffs Change From Sequential to Consequential

In TEP the handoffs were sequential and uniform: translate hands to edit, edit hands to proofread, every file, same order. In the redesigned model the handoffs are consequential: intake hands each file to a different path depending on its tier, and the handoffs that carry the most weight are the routing decision at the top and the go/no-go at the gate. This has a concrete governance implication. In TEP, accountability was diffuse (three people touched every file and none of them owned the ship decision), which is exactly why "the engine wrote it" felt like a defense when something shipped wrong. In the new model, accountability is explicit and located: the intake owner is accountable for the tier, the post-editor is accountable for the source-verified edit, the evaluator is accountable for the gate decision, the terminologist is accountable for term conformance, and a named person signs the delivery. The cardinal rule, accountability stays human and "the engine wrote it" is never an answer, is enforced by the org chart, not by exhortation, because every point where a Critical could ship has a named human owner with the authority and the accountability for that point.

The roles do not survive unchanged. Translator becomes post-editor and risk owner; the editor splits into quality evaluator and terminologist; the proofreading pass thins and its talent moves to transcreation and exceptions. The handoffs change from sequential-uniform to consequential-routed, and accountability moves from diffuse to located.

A Worked Org Redesign: From TEP Line to Quality-Owned Model

Abstract principles do not survive contact with a CFO. So take a concrete enterprise localization function and redesign it end to end, so you can see the boxes, the reporting lines, the decision rights, and the metrics change together. The function is real in the way every function is real: Meridian, the in-house localization group at a medical-device and health-software company, shipping into forty locales across product UI, user documentation, marketing, regulatory labeling, and legal. Before the redesign, Meridian ran pure TEP through a mix of in-house linguists and a vendor pool.

The Before: Meridian on TEP

Meridian's legacy org had a Director of Localization over three team leads, one each for translation, editing, and proofreading, mirroring the line. Forty in-house linguists and a large vendor pool were organized by language, each linguist paid and measured by words per day. Every file, whether a throwaway UI string or a contraindication in a device label, entered the same TEP line and got the same three passes. MT had crept in tactically (the TMS pre-populated segments) but the org shape had never changed to reflect it, so linguists were post-editing machine drafts while still being measured on words and still calling it translation. Quality was asserted, not scored: the proofreading sign-off was "looks good," and when a critical error slipped through, no one could reconstruct what scrutiny the segment had received, because the org produced no quality record. The regulatory labeling content, the most dangerous in the building, flowed through the same undifferentiated line as the marketing copy.

The After: Meridian on the Quality-Owned Model

The redesign keeps the same people and re-points the function around the five concentration points. The new org has a Head of Localization over four functions that replace the three TEP stages:

• Intake and Routing, owned by a small program-management group with a Risk-Tiering Lead. They own the top of the funnel: every file is classified Tier 1, 2, 3, or MT-forbidden at intake by rule, and the tier is recorded as the first entry in the quality record. This function did not exist in TEP, and it is now the most important decision the org makes on any file.
• Post-Editing and Transcreation, the largest function, staffed by the former translators and the strongest former proofreaders, organized by language and domain competence rather than by production stage. Post-editors work the tiered funnel; the transcreation sub-team takes the brand and campaign stream that routes around post-editing. Every member is a full-competence linguist under the revised ISO 18587, and they are measured on critical errors caught and tier-appropriate throughput, not raw words.
• Quality, owned by a Quality Lead over a small pool of quality evaluators drawn from the strongest former editors. They own the severity-scored gate, they score against ISO 5060, they hold the authority to fail a file on a single Critical, and they are independent of the post-editors whose work they score, so the gate is a real control and not a self-certification.
• Terminology and Linguistic Assets, owned by a Terminology Lead, who curates the termbases the engines are grounded on, owns the enforcement sweeps, and governs the translation memory so leverage compounds quality instead of propagating errors.

Across the top, a small Localization Engineering group owns the plumbing (grounding the engines on the TM, termbase, and style guide; placeholder and length safety; the CI/CD integration) so the machine's throughput is delivered reliably and the human roles are free to concentrate on judgment.

What Changed, and Why It Holds Together

Look at what the Meridian redesign actually moved. The three production stages (translate, edit, proofread) became four quality-owned functions (intake/routing, post-editing/transcreation, quality, terminology), plus engineering as an enabler. The metric moved from words per day to a quality-owned scorecard. The regulatory labeling content that used to flow through the same line as marketing now gets Tier 1 treatment by rule, with an independent gate and a full quality record, while the UI strings that used to consume three human passes now flow through light post-edit with sampled scoring, which is where the throughput and the cost savings come from. Accountability moved from diffuse (three people, no owner) to located (a named owner at every point a Critical could ship). And the org can now say the sentence a raw-MT competitor cannot: here is the tier each content type received, here is the ISO 5060 score with zero Criticals remaining, here is the terminology conformance, here is the record, all defensible under the revised ISO 18587. The redesign did not add people; it re-pointed the ones Meridian already had at the enemy the machine actually produces.

The Transition Risks and the Change Management That Beats Them

The org chart is the easy part. The transformation fails or succeeds on the transition, and the transition risks are predictable enough that not planning for them is negligence. Name them, because a redesign that ignores them ships a beautiful diagram into a demoralized function that quietly reverts to the old behavior.

The Morale and Deskilling Fear

The first and largest risk is that your linguists experience the redesign as demotion: the machine took the creative act of translation and left them cleaning up after it for less money. This fear is not irrational; it is a correct reading of a badly executed version of this transformation, where the org keeps the word-factory metric and just makes people validate machine output faster. The change management that beats it is to make the "up not out" thesis real and visible, not rhetorical. The post-editor's competence is more valued in the new model, not less, because the revised ISO 18587 requires full professional-translator competence to catch the fluent error, and the org's new metrics reward exactly the judgment the machine cannot replicate. The concrete moves: retitle and re-band the roles to reflect the quality ownership (a "Quality Evaluator" and a "Terminologist" are visibly senior to a "proofreader"), change the incentives so people are measured and paid on quality owned rather than words processed, and create the career path from post-editor to evaluator to terminology lead to head of localization that the new org chart makes possible. People revert to word-factory behavior when the org still rewards it; they own quality when the org visibly promotes and pays for it.

The Quality-Cliff Risk During Cutover

The second risk is a quality cliff during the transition: you dismantle the three-pass TEP safety net faster than you stand up the tiered gate, and for a window the function is running MT-first throughput with neither the old redundancy nor the new control fully in place. This is where a shipped Critical error during the transition can poison the whole redesign politically. The change management is to sequence the cutover so the control precedes the throughput, never the reverse. Stand up the risk-tiering intake and the severity-scored gate on the highest-consequence content first, prove they catch what TEP was catching (and the fluent errors TEP was missing), and only then expand MT-first throughput down the tiers. Pilot on a bounded content type and language, evidence the quality with the ISO 5060 scores and the record, and scale on the evidence, so you never have MT-first volume running ahead of the human control that owns its quality. The rule is that you earn the right to the throughput by first proving the gate.

The Accountability-Gap Risk

The third risk is that in the messy middle of a reorg, accountability falls into the gap between the old roles being dismantled and the new decision-owners being fully empowered. A file ships wrong and everyone points at the org chart in transition. The change management is to make the accountability map explicit and dated: from day one of the cutover, every point where a Critical could ship has a named owner, even if that person's title and reporting line are still settling. The intake owner owns the tier, the evaluator owns the gate, a named person signs each delivery. You do not wait for the reorg to finish to locate accountability; you locate it first, because the cardinal rule that "the engine wrote it" is never an answer requires a human owner at every risk point, and a transition is exactly when that requirement is most likely to lapse.

The Vendor and Client Alignment Risk

The fourth risk is external: your vendor pool and your internal clients are both still operating on TEP assumptions. Vendors quote and staff for the old line; clients expect either old-world human prices or new-world machine prices and do not understand the tiered model in between. The change management is to bring both onto the new operating model explicitly. Vendors are re-contracted around the tiered treatments and the quality record, so they deliver Tier 1 full-PE-plus-gate or Tier 3 light-PE-sampled rather than an undifferentiated per-word service. Clients are educated on the quality tiers so they are buying a provable tier ("Tier 1, ISO 5060-scored, zero Criticals, full record") instead of racing you to the bottom on a per-word price that hides the risk. A redesign that fixes the internal org but leaves vendors and clients on TEP assumptions will be dragged back toward the old shape by the parties it did not bring along.

The org chart is the easy part. The transition fails on four predictable risks: the deskilling fear, the quality cliff during cutover, the accountability gap in the messy middle, and misaligned vendors and clients. Each has a specific change-management answer, and the governing rule is that you earn the throughput by first proving the control.

The Sequence That Works

Put the change management into an order, because sequence is most of what separates a redesign that lands from one that thrashes. Lead with the principle and the "up not out" contract, so people understand the redesign moves them up the value chain before they experience any change to their work. Stand up risk-tiered intake and the severity-scored gate on the highest-consequence content first, and locate accountability at every risk point from day one. Pilot MT-first throughput on a bounded, lower-consequence stream, evidence the quality with the score and the record, and expand down the tiers on the evidence. Re-title, re-band, and re-incentivize the roles so the org visibly rewards quality ownership rather than word count, and build the career path that makes the new roles a promotion. Bring the vendors and clients onto the tiered model so the whole ecosystem operates on the new shape. Do it in that order and the function migrates from the line to the funnel with its people, its quality, and its client relationships intact. Do it out of order (throughput before control, tooling before incentives, org chart before the trust contract) and you will ship a diagram into a function that reverts the first time it is stressed.

Key Takeaways

• TEP (translate-edit-proofread) is a linear operating model whose three sequential human passes were designed to catch the errors a tired human makes; it is structurally blind to the machine's signature failure, the fluent, confident, invisible critical error, which three polish passes validate rather than catch. The redesign re-points the function's controls at the enemy the machine actually produces.
• The organizing principle of the MT-first, quality-owned operating model is a division of labor by comparative advantage: the machine owns throughput, the human owns quality and judgment, and the boundary is drawn by consequence, not convenience. This forces the core metric to move from words per day to quality owned (Criticals caught, terminology conformance, defensible record).
• Humans concentrate at five points the machine is structurally blind to: risk-tiering at intake, the severity-scored quality gate, terminology, transcreation, and exceptions. The redesigned org chart is built around those five concentration points, not around the three production stages of TEP.
• The workflow shape changes from a uniform line (every file, same three passes) to a tiered funnel that sorts by consequence at intake and routes each stream to the treatment its risk earns: Tier 3 light PE with sampled scoring, Tier 2 full PE with a full gate, Tier 1 full PE plus an independent gate and full record, and MT-forbidden content to human translation by rule. Transcreation runs parallel, routing around the funnel entirely.
• The roles change in kind: the translator becomes a post-editor and risk owner (still full-competence under the revised ISO 18587); the editor splits into an independent quality evaluator who owns the gate and a terminologist who owns approved terms; the surface-polish proofreading pass thins and its talent moves up into transcreation and exceptions. Handoffs move from sequential-uniform to consequential-routed, and accountability from diffuse to located.
• The gate is a control with decision rights and a named owner who can fail a file on a single Critical, not a fourth sequential inspection stage bolted onto the old line. What makes it real is the authority to say no, not its position in the diagram.
• The worked Meridian redesign turns three production stages into four quality-owned functions (intake/routing, post-editing/transcreation, quality, terminology) plus engineering as an enabler, keeping the same people and re-pointing them, so regulated labeling gets Tier 1 treatment by rule while UI strings flow light, and the org can state the tier, the ISO 5060 score with zero Criticals, the terminology conformance, and the record, all defensible under ISO 18587.
• The transformation succeeds or fails on the transition, against four predictable risks: the deskilling fear (beat it by making "up not out" real through titles, bands, incentives, and a career path), the quality cliff during cutover (beat it by standing up the control before the throughput), the accountability gap in the messy middle (beat it by locating a named owner at every risk point from day one), and misaligned vendors and clients (beat it by re-contracting and educating both onto the tiered model). The governing sequence: prove the control, then earn the throughput.