The email that starts every price war in localization arrives on a Thursday afternoon and it is only two sentences long. A procurement manager at a mid-market software company, a client you have served for three years, writes to say she has received a competing bid for the German, French, and Japanese work at "$0.06 a word, all in, MTPE." Your current rate on the same content is $0.11. She likes you, she says, but she cannot justify paying nearly double for what looks, on her spreadsheet, like the same line item: "translation, per word." She needs you to match it or explain the gap by Monday. You already know the vendor she is quoting. They run raw machine output through a light human skim and call it post-editing. You also know something she does not: two of the four content types in her account are a mistranslation away from a regulator letter or a lawsuit, and the $0.06 vendor has no way to prove they did not ship exactly that. This lesson is about the answer you send on Monday. Not "we'll match it," which starts a race you lose. Not "trust us, we're better," which she cannot put in a spreadsheet. The answer is a set of named quality tiers, each with a deliverable, a severity-score guarantee, and a defensible price, that turns "per word" from a commodity into a product line, and turns you from a vendor she compares on price into the only vendor who can prove what she is buying.

The Commodity Trap and Why "Per Word" Is Losing You

Start with the mechanics of the trap, because most language-service providers walk into it without noticing. A language-service provider (LSP) is a company that sells translation and localization as a service, whether a two-person boutique or a global firm. Machine translation (MT) is any engine that renders text from a source language into a target language with no human writing the words, whether a dedicated neural machine translation (NMT) engine or a general-purpose large language model (LLM) that translates as a side effect of predicting plausible text. Machine-translation post-editing (MTPE) is the workflow where the engine drafts every segment (the unit a translation tool breaks text into, usually a sentence) and a human revises it. And here is the structural problem: when you and your competitor both quote "MTPE, per word," you have agreed, without meaning to, to compete on a single number that hides everything that actually differs between you.

Per-word pricing is a measurement of input volume. It says nothing about what happens to that volume. Two vendors can both say "MTPE at $X/word" while doing radically different work: one runs the engine and skims for typos; the other runs the engine, verifies every number and negation against the source, scores the file against a formal error typology, blocks delivery on a single critical error, and hands the client an audit-grade record. Those are not the same service at different quality levels. They are different services entirely. But the unit of sale, the word, is identical, so on the client's spreadsheet they collapse into one comparable line, and the cheaper one wins by definition. You have let the pricing unit erase the thing you are actually good at.

This is not a new problem, but 2026 made it acute. MTPE adoption rose from 26% in 2022 to roughly 46% in 2024, and 81.1% of LSPs now offer MTPE. When a capability is offered by four out of five vendors, it stops being a differentiator and becomes table stakes, and table stakes get priced to the floor. Meanwhile the client has read a dozen articles telling her machine translation is now "good enough," so she arrives already believing the work is a commodity and already primed to pay commodity prices. The $0.06 email is the entirely predictable output of a market where everyone sells the same unit and the buyer cannot see inside it.

The mistake almost every LSP makes in response is to argue about the number. You explain that your linguists are better, that your process is more rigorous, that quality matters. All of that is true and none of it survives contact with a procurement spreadsheet, because it is unmeasured and therefore unpriceable. "Our quality is higher" is a claim. Procurement does not buy claims; it buys line items with attributes it can compare. The only way out of the commodity trap is to stop selling the word and start selling something the client can put in a cell next to a number: a quality tier, a named, defined level of service with a specified deliverable and a measurable guarantee attached.

Per-word pricing sells volume and hides value. The moment you and your competitor both quote "MTPE per word," you have agreed to compete only on the one attribute that erases your advantage. The escape is to sell a tier, not a word.

The Four Tiers and What Each One Actually Is

A defensible tier is not a marketing label like "gold" or "premium." It is a precise triple: a deliverable (what the client receives and what the output is permitted to read like), a guarantee (the measurable quality promise, expressed as a severity score, that you are willing to be held to), and a risk statement (an honest account of what the tier does not protect against). Four tiers cover the real range of localization work. Define each by all three attributes, because a tier missing any one of them is a slogan, not a product.

Tier One: Raw MT

Raw MT is machine output with no human touching it. The deliverable is exactly what the engine produced, delivered as-is. There is no post-editor, no verification, no score. The output is permitted to contain any error the engine makes, and the client is told this in writing. The honest use cases are narrow but real: internal gist translation so someone can decide whether a document is worth translating properly, high-volume user-generated content where the alternative is nothing at all, and disposable throwaway text with zero consequence and zero brand exposure. The guarantee is deliberately null: you guarantee nothing about accuracy, and you say so. The risk statement is the loudest of any tier: this output may contain silent critical errors, it has not been checked against the source, and it must never be used for anything a person acts on, publishes, or could be harmed by. Selling raw MT without that risk statement in writing is not a cheap tier; it is a liability you have taken on for free. Selling it with the risk statement is a legitimate low-cost product for the narrow content that genuinely warrants it.

Tier Two: Light Post-Editing

Light post-editing (light PE) aims at one target word: understandable. A human reads every segment against the source and fixes only what would mislead a reader: outright mistranslations, dropped or flipped meaning, anything that sends the reader to the wrong conclusion. Then they stop. The defining discipline of light PE is the prohibition on preferential edits, changes that improve the prose by the editor's taste but were not actually wrong. Clunky-but-accurate stays clunky. The deliverable is output that is correct in meaning but permitted to read like a machine produced it. The honest use cases are high-volume, low-shelf-life, low-consequence content: product catalogs, support articles, internal knowledge bases. The guarantee is meaning-level: no critical errors that mislead the reader, but no promise about style, register, or the polished feel of human translation. The risk statement is specific: because light PE deliberately limits verification depth, it is unsuitable for content where a subtle factual, legal, or safety error would cause harm; it protects against obvious mistranslation, not against a fluent error buried in a dosage, a negation, or an obligation.

Tier Three: Full Post-Editing

Full post-editing (full PE) aims at a different target word: indistinguishable. The output has to read as though a competent human translator produced it from a blank page, with no tell that a machine started it. The post-editor fixes everything light PE fixes and then keeps going: style, register, terminology consistency, locale conventions, flow, and the exhaustive source-against-target verification of every high-consequence element. The deliverable is publication-quality translation at the standard a client expects from full human work. This is the tier that carries a formal severity-score guarantee and blocks delivery on a single critical error. The guarantee is the strongest measurable promise you make: a stated error-typology score with zero critical errors, which we build in the next section. The risk statement is narrow and honest: full PE by a full-competence linguist catches the fluent silent error, but no human process is infallible, so the guarantee is expressed as a defined severity threshold on a defined sample, not as a claim of perfection. This is where the revised ISO 18587 lives: it requires the post-editor to hold the same linguistic competence as a professional translator, precisely because catching the silent fluent error is a translator's deepest judgment, not a button-pusher's reflex.

Tier Four: Full Human Translation with Revision

Full human translation is translation from the source by a professional linguist, followed by independent revision by a second linguist, with the machine either absent entirely or used only as a private reference the translator may consult and discard. The deliverable is the highest quality the field produces, and for a defined class of content it is the only defensible choice. The honest use cases are the content that must never be machine-drafted: regulated medical and pharmaceutical copy, binding legal instruments, financial disclosures, life-safety instructions, and high-stakes transcreation where brand and cultural meaning are the product. The guarantee is the full two-linguist standard of ISO 17100, the human-translation services baseline, with a severity score that expects zero critical and zero major errors. The risk statement is the shortest of any tier, because this tier exists to eliminate the risks the cheaper tiers carry: it is the correct routing for content where a single silent error costs a life, a license, or a lawsuit, and quoting a cheaper tier on that content is not a discount, it is malpractice.

A tier is a triple: the deliverable (what it reads like), the guarantee (the severity score you stand behind), and the risk statement (what it does not protect against). A tier missing the guarantee is a slogan. A tier missing the risk statement is a lawsuit waiting for a buyer.

The Severity Score That Makes a Tier Provable

Everything above is still just a nicer set of labels until you attach the thing that turns a label into a product: a measurable, defensible quality guarantee. This is where the $0.06 vendor cannot follow you, and it is the single most important move in the lesson. The instrument is a severity-scored error evaluation built on ISO 5060:2024, the standard that formalizes the MQM (Multidimensional Quality Metrics)-aligned scoring of translation output into Critical, Major, and Minor error severities across the dimensions of accuracy, terminology, locale, and fluency.

Here is how the score works in plain terms, because you will explain it to a client who has never heard of it. An evaluator, a qualified linguist who did not do the translation, reads a defined sample of the delivered output against the source and marks every error they find. Each error gets a category (was it an accuracy error, a terminology error, a locale error, a fluency error?) and a severity. A Critical error is one that carries safety, legal, or financial consequence, or that makes the output unusable for its purpose: a flipped dosage, an inverted indemnity clause, a dropped negation that reverses a permission. A Major error seriously distorts meaning or violates a firm requirement but stops short of that consequence. A Minor error is a real but low-impact flaw: a slightly awkward phrasing, a small locale slip that does not mislead. The severities get weights, the weighted errors get normalized against the sample size, and the file lands on a score. Crucially, one rule sits above the arithmetic: a single Critical error fails the file, regardless of how clean everything else is. That rule is what makes the score a guarantee and not an average that lets a catastrophe hide behind a thousand clean segments.

Now watch what the score does to your pricing conversation. When you sell "full PE," the abstract promise is "good quality." When you sell "full PE with an ISO 5060 severity score, guaranteed zero Critical errors on a representative sample, with the error report attached to delivery," you have converted the promise into a line item procurement can compare. The $0.06 vendor's spreadsheet cell says "MTPE, per word." Your cell says "MTPE, ISO 18587-conformant, ISO 5060 score attached, zero-Critical guarantee." Those are no longer the same product at two prices. They are two products, and the client can finally see the difference she is paying for.

Attaching a Number to Each Tier

Each tier carries a different severity guarantee, and the differences are the whole point. Raw MT carries no score, by definition. Light PE guarantees zero Critical errors that mislead the reader, while explicitly permitting Minor and some Major stylistic errors, because the tier's whole economic logic is that it does not chase them. Full PE guarantees zero Critical and a low, stated ceiling on Major errors, with the file failing on a single Critical. Full human translation guarantees zero Critical and zero Major, the tightest standard the field offers. Write these numbers into the tier definition and into the contract. The guarantee is not a marketing flourish you can quietly walk back; it is the specific, measurable thing that justifies the price, and the willingness to be held to it, with an evaluator's report proving it, is exactly what the commodity vendor cannot match. Their price is lower because their guarantee is nothing. Say that out loud, in those terms, and the gap stops looking like a markup and starts looking like the difference between a promise and a hope.

A quality tier without a severity score is a color. A quality tier with an ISO 5060 score and a zero-Critical guarantee is a product. The score is the thing the client can put in a spreadsheet cell, and the thing your cheapest competitor structurally cannot.

Pricing Each Tier So the Numbers Defend Themselves

Now the money, priced so that each number carries its own justification. Hold the industry anchors first, because clients will quote them and you must own them before they do. MTPE typically prices at roughly 50 to 75% of full human translation rates, which lands around $0.05 to $0.15 per word against full human work that runs higher. Light PE can be priced as low as about $0.02 per word, because the brief is genuinely lighter: meaning-only, no polishing, no preferential edits. And the productivity that makes the whole model viable is real: a hybrid MTPE workflow lifts a linguist from roughly 2,000 words a day toward 5,000 or more, which is precisely why the per-word price can fall while the linguist's daily earnings need not.

The strategic error is to derive your prices from your competitor's number. The correct method is to derive each tier's price from the effort and the guarantee it actually carries, then let the spread between tiers tell the client's own procurement logic. A worked tiered sheet, expressed as a percentage of your full-human-translation base rate so it travels across language pairs, looks like this. Read it as the skeleton of the Monday email, not as fixed cents.

• Raw MT: a nominal per-word or per-project handling fee, often under $0.01 equivalent, sometimes bundled as a flat engine-access charge. It carries no linguist cost and no guarantee, so it prices near the cost of running the engine plus your margin. It is not where you make money; it is where you honestly serve the narrow gist-and-triage use case without pretending it is quality work.
• Light PE: roughly $0.02 to $0.05 per word, or about 25 to 40% of your full-human base. The brief is meaning-only with a zero-misleading-Critical guarantee, and the linguist moves fast because preferential edits are forbidden. This is your honest floor. Price it confidently, because when a client demands your cheapest number you can give them a real one with a real (if modest) guarantee attached, which the commodity vendor's "MTPE" cannot claim.
• Full PE: roughly $0.05 to $0.15 per word, or about 50 to 75% of your full-human base. This is the tier the whole strategy defends. It carries the ISO 18587-conformant post-editing brief, the full-competence linguist, the ISO 5060 severity score, and the zero-Critical guarantee. The price is not a markup on light PE; it is the price of a fundamentally more demanding deliverable, and the severity report is the receipt.
• Full human translation with revision: your base rate, 100%, reflecting two linguists (translation plus independent revision) under the ISO 17100 baseline, with the tightest severity guarantee. For the content that legally or clinically requires it, this is not the expensive option; it is the only defensible one, and its price is justified by the catastrophe it prevents rather than the hours it consumes.

The Percentage Move That Wins the Argument

Notice what pricing as a percentage of a human base rate accomplishes. It reframes the entire conversation away from "is $0.11 too much" and toward "which tier does this content need." When the client sees full PE at 60% of human and full human at 100%, and she sees that her drug-label copy is routed to the 100% tier because a flipped dosage is a recall, the price stops being a negotiation and becomes a consequence of a routing decision she can understand and defend to her own boss. The $0.06 vendor, meanwhile, has no tiers. They have one number and one undifferentiated service, which means they are quoting light-PE-grade effort on every content type including the ones that can sue. You are not more expensive than them. You are correctly priced across four products where they have mispriced one.

One more discipline: never let the tiers blur at the edges to win a deal. The temptation, when a client pushes, is to quietly do full-PE verification but bill it as light PE to hit the target number. That destroys the whole architecture. It gives away the margin that funds your guarantee, it trains the client that your tiers are negotiable fictions, and it means the next competitor who actually does light-PE-grade work at light-PE prices undercuts you again. The tiers only defend your pricing if the wall between them is real: light PE gets light-PE effort and a light-PE guarantee, full PE gets full-PE effort and the severity score, and the price of each reflects exactly what was done. The integrity of the tier is the source of its pricing power.

Positioning Against the Race to the Bottom

Now assemble the Monday email, because the tiers and the score are the raw material and positioning is what turns them into a won deal. The positioning move is a single sentence you can say to any client, and it is the sentence a raw-MT vendor can never say: "We sell ISO 18587-conformant MTPE with an ISO 5060 severity score attached and a zero-Critical guarantee, routed by content risk into four defined tiers, and here is the record that proves each delivery." Every clause in that sentence is a wall the commodity vendor cannot climb.

Break down why it works against the $0.06 bid. First, it changes the comparison from price to product. The client wanted you to explain a 2x price gap on an identical line item. You respond by showing that the line items are not identical: their "$0.06 MTPE" is your light PE tier at best, and it is being quoted on content that needs full PE or full human. You have not defended a higher price for the same thing; you have revealed that they are selling the wrong thing cheaply.

Second, it moves the client's own risk onto the table, in her language. A procurement manager is accountable if a bad translation ships. Ask her, gently, one question: "When your competitor ships a flipped negation in the banking disclosure or a corrupted dosage in the pharma insert, who explains that to your regulator, and what do they hand over to prove they tried to catch it?" The $0.06 vendor has no answer, because they have no score, no gate, and no record. You have all three. You have just reframed your price as insurance she is buying against a liability she owns, and insurance is a category procurement understands and funds.

Third, it lets you concede gracefully where you should. This is the move that builds trust rather than defensiveness. Look at her four content types and say: "For your support articles and your marketing catalog, you are right, that content is genuinely light-PE work, and here is our light-PE price, which is competitive with their number and comes with a real guarantee theirs lacks. For your financial disclosures and your regulated product copy, that content cannot be light PE at any vendor, including us, and here is why, and here is the tier it requires." You have matched them where matching is honest and held the line where holding is defensible, and you have done both with the same tier architecture. That is not a vendor racing to the bottom. That is an advisor who priced the client's risk correctly, and it is the relationship that survives the next $0.06 email instead of being ended by it.

The Provable Tier as a Moat

Understand what you have built when this works. The commodity vendor competes on the one axis where someone can always be cheaper: price for undifferentiated MTPE. You have moved the competition to an axis where cheapness is a liability rather than an advantage: provable quality routed to risk. On that axis, the vendor who cannot produce an ISO 5060 score is not a cheaper version of you; they are a different, riskier product that a serious client cannot buy for serious content. The severity score, the tier definitions, the risk statements, and the conformance to the revised ISO 18587 and ISO 5060 together form a moat, not because they are secret, but because building them requires exactly the quality discipline that a race-to-the-bottom operation has spent years dismantling. Your competitor cannot cut their way to your position. They would have to rebuild the thing their pricing strategy destroyed.

Do not defend your price against a cheaper vendor. Change the axis. Move the competition from "cheapest per word" to "provable quality routed to risk," and the vendor who cannot attach a severity score stops being cheaper and starts being unbuyable for anything that matters.

The Worked Tiered Sheet and a Routing Example

Make it concrete with the client's actual account, because a tiered sheet only earns its keep when it routes real content. She has four content types across German, French, and Japanese: a marketing catalog, a set of support articles, a banking-app disclosure flow, and regulated pharmaceutical product copy. Here is the sheet you send, and the routing logic that makes each price self-defending.

Routing Each Content Type to Its Tier

• Marketing catalog: high volume, low shelf life, low consequence, but brand-facing. Route to light PE on style with pointed verification on any factual claim (material composition, safety marks, compatibility). Price at roughly 30 to 40% of the human base. Guarantee: zero misleading Critical errors. This is where you meet the $0.06 number honestly, and your version carries a guarantee theirs does not.
• Support articles: customer-facing, moderate consequence (a wrong instruction generates tickets, not lawsuits), moderate brand exposure. Route to light-to-full PE depending on the article: troubleshooting steps that could damage a device lean full; FAQ prose leans light. Price at 40 to 60% of base. Guarantee: zero Critical, low Major ceiling.
• Banking-app disclosure flow: client-facing, brand-critical, and legally consequential, because a flipped negation in a privacy or fee disclosure is a compliance failure. Route to full PE with the full ISO 5060 severity score and a hard zero-Critical gate, plus format and placeholder verification because these are UI strings with length budgets. Price at 60 to 75% of base. Guarantee: ISO 5060 score attached, zero Critical, delivery blocked on any Critical.
• Regulated pharmaceutical product copy: life-safety and regulated, the content where LLM error rates are frighteningly high, roughly 59% on drug names, 60% on dates and times, and 66% on adverse events, every one delivered in fluent prose. Route to full human translation with independent revision under the ISO 17100 baseline. Price at 100% of base. Guarantee: zero Critical and zero Major, full two-linguist record. This content leaves the MTPE conveyor entirely, and quoting it at MTPE rates, as the $0.06 vendor implicitly does, is the malpractice that ends relationships and invites regulators.

Now the email writes itself. Two of her four content types (catalog and much of support) get a competitive light-to-mid price that respects her budget. Two (banking disclosures and pharma) get full PE and full human, correctly, with guarantees attached, and the price gap on those is no longer a markup she has to justify but a risk decision she can defend to her own compliance and legal teams. You have not matched $0.06 across the board, which would have been a lie on half her content. You have shown her that a single flat number is itself the red flag, because no responsible vendor prices a drug insert and a sock description at the same rate. The tiered sheet did the persuading. You just sent it.

What to Put in the Client's Hands

Package three artifacts with every tiered proposal, because the tier is only as defensible as the evidence behind it. First, the tier definitions themselves: the four triples of deliverable, guarantee, and risk statement, written plainly so a non-linguist procurement manager can read them and route her own content. Second, a sample severity report from a real ISO 5060 evaluation on a representative file, so "zero-Critical guarantee" is not a phrase but a document she has seen the shape of. Third, a one-page risk-routing map that shows which of her content types belong in which tier and why, so the pricing looks like a consequence of her risk rather than a negotiation over your margin. Those three artifacts convert every future price challenge from an argument you have to win into a decision she has already been equipped to make. That is the deepest form of pricing power: not winning the negotiation, but making the negotiation unnecessary because the client can see, in her own terms, exactly what she is buying and why it costs what it costs.

Key Takeaways

• Per-word pricing sells input volume and hides the value you add, so when you and a competitor both quote "MTPE per word," you have agreed to compete on the one attribute that erases your advantage; with MTPE now offered by 81.1% of LSPs, the capability is table stakes and gets priced to the floor unless you sell a differentiated tier instead of a word.
• A defensible quality tier is a triple: a deliverable (what the output is permitted to read like), a guarantee (the measurable severity score you stand behind), and a risk statement (what the tier does not protect against). A tier missing the guarantee is a slogan; a tier missing the risk statement is a liability you took on for free.
• The four tiers are raw MT (no human, no score, loud risk statement, narrow gist-and-triage use), light PE (understandable, no preferential edits, zero-misleading-Critical guarantee), full PE (indistinguishable from human, ISO 18587-conformant, ISO 5060 score with a zero-Critical gate), and full human translation with revision (two linguists under ISO 17100, zero Critical and zero Major, for content that must never be machine-drafted).
• The severity score built on ISO 5060 is what turns a tier from a color into a product: an independent evaluator marks every error by category and severity (Critical, Major, Minor), one Critical fails the file regardless of how clean the rest is, and the attached report is the receipt a commodity vendor structurally cannot produce.
• Price each tier from the effort and guarantee it actually carries, expressed as a percentage of your full-human base rate: raw MT near engine cost, light PE at roughly 25 to 40% (about $0.02 to $0.05/word), full PE at 50 to 75% (about $0.05 to $0.15/word), full human at 100%. MTPE runs at 50 to 75% of human rates because a hybrid workflow lifts a linguist from ~2,000 to 5,000+ words a day, so per-word price falls without daily earnings collapsing.
• Never blur the tiers to win a deal: doing full-PE verification while billing light PE gives away the margin that funds your guarantee and trains the client that your tiers are fictions. The integrity of the wall between tiers is the source of their pricing power.
• Positioning against a cheaper vendor means changing the axis, not defending the price: sell "ISO 18587-conformant MTPE with an ISO 5060 score attached and a zero-Critical guarantee, routed by content risk," reframe your price as insurance against a liability the client owns, concede honestly on genuinely light-PE content, and hold the line on content that legally requires full PE or full human.
• Route the client's own content to tiers and let the routing defend the price: low-consequence marketing to light PE (meet the cheap number with a real guarantee), consequential disclosures to full PE with a hard Critical gate, and regulated or life-safety content to full human translation, where the ~59% drug-name, 60% date, and 66% adverse-event LLM error rates make MTPE-rate pricing outright malpractice. Hand over the tier definitions, a sample severity report, and a risk-routing map so the client can see exactly what she is buying.