Stepes review

There is a dedicated marketing translation page that does define transcreation and separates it from word-for-word work: 'Transcreation literally stands for the two-step process: translation followed by creative writing to adopt the translation to suite the local audience's cultural and linguistic taste', alongside 'Verbatim or word-for-word translations often fail to deliver the punch across languages'. That is the language of a vendor that understands the distinction rather than one that pasted a marketing label onto translation. Fourteen marketing-side client logos appear on that page, including AKQA, Burson Cohn & Wolfe, Constellation Agency, TheBrandAmp, Splash Box Marketing, Gramafilm and Tether, which is an unusually agency-heavy roster. But no attributable campaign work backs any of them: there is no case study, no described brief, no before-and-after copy, no stated language pairs on a specific engagement. The case study library at /resources/case-studies/ is a directory of nine collection headings that states 'Some case studies identify the customer and include an approved logo or quotation. Others are anonymized', and no individual study was reachable through it; /case-studies/digital-content/ returned unrelated Dutch boilerplate rather than the page its URL names. Named clients without described work is the rubric's 2-3 band, and rule 6 caps a portfolio that could not actually be viewed at Adequate.

What this dimension measures: Is marketing work a real, named practice with attributable campaign or content work — or a service-page claim on a general translation business? Named marketing clients with described campaign work across stated language pairs scores 4-5. A marketing services page with no attributable work scores 2-3. Distinguish transcreation (copy reworked for the market, briefed like creative) from translation with a marketing label; vendors that explain the difference in their own materials tend to actually do it. Counts for 25% of the weighted score.

Evidence: partly checkable — corroborated in part against the sources below; the remainder rests on the agency’s own account.

The review model is described with more specificity than most: the Spanish service page lays out independent bilingual review, automated QA checks, specialist validation, in-country validation, and in-context final-format QA as distinct stages, and linguists are said to be selected on 'native-language competence, market familiarity, subject matter, and content type'. Machine translation is disclosed rather than hidden, which the rubric asks for: 'Stepes uses AI to accelerate multilingual production while professional linguists review, edit, and validate translations', and MT post-editing is sold as its own named service line, so no MT is being passed off as human work. Against that, the pool is a scale claim of '10,000+ professional linguists' with no account of vetting, retention or in-country employment status. The certification claim cannot be verified: badges read 'ISO 17100', 'ISO 9001' and 'ISO 13485', and /help/certification-standards/ asserts 'Our commitment to quality and customer satisfaction is demonstrated through our adherence to internationally recognized certifications such as ISO 9001 and ISO 17100', but no certificate number, certifying body, audit date or registry link is published anywhere read, and note the hedge word 'adherence'. The rubric's top band requires certification that can be verified, so this caps at Adequate.

What this dimension measures: Who does the work and who checks it: in-country professional linguists versus anonymous pools, named review layers, and certifications that can be verified (ISO 17100, ISO 9001 — check the certificate claim, not the badge). A described model with named QA steps and verifiable certification scores 4-5. 'Native speakers' with no process detail scores 2-3. How the vendor positions machine translation matters: disclosed MT-plus-post-editing with stated boundaries is honest and scores on its own terms; undisclosed MT sold as human work is a red flag scored 1-2 where evidenced. Counts for 20% of the weighted score.

Evidence: vendor stated — the agency’s own claim, recorded as theirs rather than ours.

Actual per-word numbers are published by service tier at /translation-pricing/, including a marketing tier priced separately from the rest: Standard $0.09-$0.20 per word for manuals and technical documentation, Marketing $0.12-$0.24 per word for brochures and advertising materials, Legal/Medical $0.12-$0.24 per word. That is a real disclosure of the billing mechanism this category usually hides, and a buyer can budget a word count before speaking to anyone. It falls short of Excellent because the disclosure is incomplete in the ways that bite on transcreation specifically: no hourly or per-project transcreation rate, which is how creative reworking is normally billed since it is not a per-word activity; no per-language minimums; and no desktop-publishing or review surcharges. Individual language pages pull back from the published grid, with the Spanish page stating pricing 'depends on the content, language direction, Spanish variant, production workflow, file format, volume, review requirements, and delivery schedule' and routing to a quote instead. The unqualified boast 'We charge one of the lowest per word fees within the translation and localization industry' is sales copy and was not scored.

What this dimension measures: Published per-word rates, hourly transcreation rates, or stated pricing models by service tier score 4-5. A stated model without numbers scores 2-3. Bespoke-only quoting scores 1-2. This category bills by mechanisms buyers rarely see up front (per-word, per-hour, minimums per language); a vendor that explains its own billing mechanics scores toward the top of whatever band its disclosure earns. Counts for 15% of the weighted score.

Evidence: verified — checked against a named source you can open; the links under Sources below are where to check it yourself.

The workflow is described concretely rather than gestured at. Named entry paths include a Translation API, a Software Localization API and a website translation proxy described as 'A managed layer can create and serve multilingual versions without storing every translation in the source CMS', and the website page lists intake via 'CMS workflows, APIs, scheduled exports, structured files, proxy technology, or direct content submission'. The platform page describes translation request intake, multilingual workflow management, project milestones and status visibility, reviewer collaboration, version tracking and delivery management. Continuous localization is supported in the way that matters for content that ships weekly, with content comparison plus translation memory used to 'Translate only what changed'. The marketing page names Marketo and Salesforce as marketing-automation integrations. Not Excellent because the connector list is never enumerated: CMS support is described only as 'enterprise, open-source, cloud-based, and custom CMS environments through the most practical content exchange model', no design-stack integration such as Figma is named, and file formats are deferred to a separate resource rather than listed on any page read. Standards-based formats TMX and XLIFF are supported, which limits lock-in exposure.

What this dimension measures: How work actually flows: TMS use, connectors into the client's CMS or design stack, stated file-format coverage, and whether continuous localization is supported for content that ships weekly. Named integrations and a described workflow score 4-5. Email-the-files-back workflow scores 2-3 — workable for a campaign, a ceiling for a content operation. Proprietary-platform lock-in with no export path should be named in the verdict. Counts for 15% of the weighted score.

Evidence: vendor stated — the agency’s own claim, recorded as theirs rather than ours.

This is the clearest gap. The translation memory page is built entirely around ownership language without ever stating ownership: it promises to 'Turn Every Approved Translation Into a Reusable Enterprise Language Asset', to 'Retain Control of Your Multilingual Intellectual Property', and to 'Centralize multilingual intellectual property instead of leaving it scattered across projects and providers'. Nowhere does it say the client owns the TM, the glossaries or the deliverables, and nowhere is there an exit provision, a stated export-on-termination right, or a link to terms that would settle it. The 'Portable by Design' section that mentions TMX, XLIFF, bilingual files and legacy assets describes bringing existing resources in, not taking accumulated memory out. Standards-based formats are at least a practical hedge, so this is not the vendor-retains-the-TM case the rubric scores 1-2. But a buyer funding memory build-up across many languages cannot tell from the published material whether that asset leaves with them, and the rubric scores silence on ownership at 2.

What this dimension measures: The translation memory and glossaries built on the client's money are the client's accumulated asset, and whether they leave with the client at exit is the switching question of this category. Explicit client ownership of TM, glossaries and deliverables scores 4-5. Silence scores 2. A stated policy that the vendor retains the TM scores 1-2 — that is the client's spend converted into the vendor's moat, and it should be named plainly. Counts for 10% of the weighted score.

Evidence: inferred — our reading of indirect evidence, not a documented fact.

Stronger here: Argos Multilingual scores Strong on the same dimension.

No turnaround commitment is published on any page read. There is no words-per-day throughput figure, no delivery window by volume or language pair, no rush or expedited surcharge terms, and no stated coverage model across time zones beyond a generic continuous-availability claim. The most concrete statements are conditional rather than committed: 'expedited delivery may be available', and larger or regulated projects 'benefit from structured schedules with planned review stages'. That restraint is at least honest, and it avoids the same-day-everything-in-forty-languages overclaim the rubric warns against rewarding. But honest silence is still silence, and the rubric scores undisclosed scheduling at 2. A buyer with a campaign flight date cannot plan against anything published here.

What this dimension measures: Stated turnaround by volume and language pair, rush terms, and stated capacity across time zones. Published, realistic turnaround commitments score 4-5. Undisclosed scheduling scores 2. Score honesty over speed: same-day everything at marketing quality across forty languages is a claim to be sceptical of, not to reward. Counts for 10% of the weighted score.

Evidence: inferred — our reading of indirect evidence, not a documented fact.

Stronger here: Translated scores Strong on the same dimension.

No independent client review base was located or readable. Searches surfaced no Clutch or G2 profile for Stepes; the ConsumerAffairs page returned HTTP 403 and was not read, so nothing from it is cited. The one independent page read first-hand, Trustpilot, shows 2.0 out of 5 across 16 reviews, but the reviewer population is freelance translators disputing payment rather than buyers assessing delivered work, so it is supply-side evidence and is deliberately not treated as client evidence for this dimension. The homepage displays enterprise logos including Deloitte, Meta, Netflix, PepsiCo and Zoom, but logos are not references: none is attached to a described engagement or a contactable referee. This weight is low by design because enterprise localization is bought on procurement references rather than public reviews, so the absence is weak evidence and the practice-evidence dimension carries the judgment.

What this dimension measures: Independent, verified reviews or named checkable references. A substantial verified base scores 4-5; a handful scores 2-3; none scores 2. Low weight deliberately — enterprise localization runs on procurement references rather than public reviews, so absence is weak evidence, and the practice evidence above matters more. Counts for 5% of the weighted score.

Evidence: partly checkable — corroborated in part against the sources below; the remainder rests on the agency’s own account.

Stronger here: Translated scores Strong on the same dimension.