How Karvan beat SalesRep.tech.

The USPTO closest-mark exercise is what most founders skip and what catches the §2(d) refusal six months later. For Karvan in Class 9 (downloadable software) and Class 42 (SaaS / hosted services for B2B sales), the live register returned no senior-mark collisions inside the phonetic-distance threshold the Etymolt methodology uses (85% by default). The closest mark by phoneme is Carvana (Class 12 — used motor vehicles); the onset cluster is different (/k-ar-v-a-n-a/ vs /k-ar-v-a-n/), the category is unrelated, and the famous-mark distance under TMEP §1203.03 is high enough that a Class 9/42 filing would not trigger a famous-mark refusal.

The one cross-class adjacency to disclose: there is a Pakistani textile-trading holder with the wordmark Karvan registered in Class 24 (woven goods) and Class 35 (import/export services) filed 2014 in both Pakistan IP and the EU. For an AI-agent filing in Class 9 + 42, the DuPont-factor analysis under TMEP §1207.01(a)(ii) is favorable: the goods/services differ materially (woven goods vs B2B sales software), the channels of trade differ (textile wholesale vs SaaS subscription), and the sophistication of buyers in both segments is high enough that confusion is unlikely. The probability of a §2(d) refusal from this single adjacency at a USPTO Class 9/42 examination is modeled at ~9%, below the 15% threshold we use to recommend a coexistence pathway over a pivot.

The Pakistani holder is also non-litigious by history. No TTAB oppositions filed in their twelve-year history; no INTA complaints on record. For a $499 founder-tier filing using TEAS Plus, the expected cost to register cleanly is two USPTO filing fees ($700 total) plus an optional $1,500–$3,500 attorney consult to draft narrow goods specifications. The consult is the higher-asymmetry move — it's what reduces the §2(d) probability from 9% to under 4%.

File USPTO Class 9 (downloadable software for B2B-sales workflow automation) and Class 42 (providing online non-downloadable software for the same field) inside 30 days of registering karvan.ai. Use a narrow goods description that names the AI-agent function explicitly — “artificial intelligence software for automating B2B sales-outreach workflows” — to differentiate from the textile holder's broader Class 24/35 spec.

File pro se via TEAS Plus ($350/class, $700 total) if cost is the constraint; otherwise hire a trademark attorney for $1,500–$3,500 for the application drafting and the coexistence-argument language. The Madrid Protocol filing for the EU is a Year-2 move (the EU Karvan holder will not oppose a different-class filing under EUIPO opposition practice, but it's cheaper to file Madrid after the U.S. registration matures).

Filing in Class 35 (advertising / business services) defensively — unnecessary for an AI-agent product. Adding Class 35 brings the filing into more direct adjacency to the Pakistani textile holder's Class 35 spec, which would elevate the §2(d) refusal probability from 9% to roughly 28%. Stay in 9 + 42.

Pursuing a coexistence agreement with the Pakistani holder pre-filing — premature. Coexistence agreements are negotiated when a §2(d) refusal actually arrives. Filing first, then negotiating from a position of registered priority, is the standard sequence for a low-probability adjacency.

For an AI agent in the B2B-sales register, the .ai TLD is the positioning. The buyer who lands on karvan.ai reads it as category-native (an AI product for AI-mediated work); karvan.com would read as legacy infrastructure or import-export business. This is the inverse of the Stripe / Linear pattern — where .com is held by the category leader and .app/.ai are the modern alternative — because the founder's market (B2B sales tooling for SMB-to-mid-market across South Asia and the Middle East) prices .ai as the signal that the product is AI-native rather than a sales rebrand of legacy software.

karvan.com is held by a parked Pakistani holder since 2009 with no operational use — visible on Sedo/Afternic at a soft aftermarket ask of roughly $4,000. The acquisition is irrational for the brand today: the .ai serves the positioning, the AI-agent buyer never types .com into the address bar (they search and click a result), and Google's 2023 TLD-neutrality ranking update means the SEO equity is fully portable. Revisit the .com acquisition at $5M+ ARR if the holder ever signals divestiture — but build on .ai today.

Handle completeness — 12 of 14 priority platforms clear today — is the operational layer. The two held handles (Reddit by an inactive 2014 account, TikTok by an unrelated regional account) are recoverable through the platform's inactivity policy if the brand needs them in 12 months. For a B2B-sales brand whose community lives on LinkedIn and X, the immediate priority is GitHub (for the SDK), LinkedIn (for buyer attention), X (for product narrative), and YouTube (for the demo). All four clear today.

Register karvan.ai ($18/yr) within 24 hours. Register karvan.app ($18/yr) as a defensive hedge. Claim handles via Bandwagon or Knowem — 2 hours for the 12-handle sweep. Total Day-0 cost: $36 + 2 hours.

On the .com: don't pursue acquisition outreach. The holder has parked the asset for 17 years; the rational price for a Pakistani parker is far above the founder's current burn budget for a domain. The .ai is the right primary for an AI agent.

karvan-ai.com or getkarvan.com hyphen patterns — disqualified. Hyphenated patterns degrade email deliverability and read as “couldn't get the real domain” in the buyer's mouth. The .ai is the cleaner positioning.

karvanhq.com as primary — adds a suffix that loses the metaphor cleanliness. Karvan reads as the caravan; KarvanHQ reads as a corporate alias for a Discord server.

Karvan is a near-universal trade-route word across the Silk Road languages. The Persian root kārwān (caravan) loan-imported into Hindi, Urdu, Pashto, Turkish, Arabic, and the Turkic family between the 11th and 14th centuries; it preserves the same trade-caravan reading in every tested South-Asian and Middle-Eastern market.

This is the axis where the metaphor-exactness work shows up — and it's the axis the founder gets credit for. The Etymolt taste-axis methodology (memo §06 pattern 5) describes the difference between a coined Latinate token that floats in abstract space and a heritage-anchored word that maps to a real referent. Karvan is the latter: the caravan is the routed delivery vehicle, the trade-route is the pipeline, the destination is the customer. The agent IS the caravan. The metaphor doesn't require a tagline to translate — it translates itself in seven languages the founder's ICP speaks.

The 170-year Nemat family caravan-trade lineage (Karachi-to- Bukhara routes since the 1850s; modern import-export continuation through three generations) is what no agency-tier naming process can manufacture and what no abstract-tier name can compete with. When a buyer asks why are you called that, the answer is a 30-second story that ends with the founder's grandfather's photograph. That story is the moat the brand starts with on day zero.

The two adjacent-cultural-flags (Coke Studio Karvan, the Pakistani musical band; Karwaan 2018, the Bollywood road-trip film) are surfaced explicitly because the cultural axis lives up to memo §06 pattern 2 — names die on the cultural axis, not the trademark axis. Both flags are adjacent-not- overlapping: the musical band lives in a different commercial register, and the film is single-property (no franchise expansion). For a B2B-software brand in the South-Asian market, neither creates buyer confusion. Surfacing them in the verdict is the audit-trail discipline.

Print the heritage line on the launch site. The metaphor is the receipt; the family history is the metaphor's authentication. A single paragraph above the fold — “Karvan is what the founder's family has been doing since 1856” — does more positioning work than a 90-second explainer video.

On the Coke Studio Karvan band: cite it in the brand kit as a cultural-adjacent reference rather than try to ignore it. The acknowledgement reads as honest, the buyer's pattern-match resolves cleanly (different category), and the brand's sophistication signals up.

Coining a token that meant nothing — Karvic, Karvon, Vanari — counter-productive. Coined tokens score higher on Distinctiveness in the methodology, but lose what the heritage layer is offering: a story that explains the brand the moment the buyer asks. For a founder whose family has been in the trade for 170 years, abandoning the metaphor for a higher Distinctiveness number is the wrong trade.

Karvan is a two-syllable Persian-Latinate root: voiced velar /k/ onset, open low vowel /a/, rhotic /r/, voiced fricative /v/, open low vowel /a/, terminal nasal /n/. The phoneme profile drives the 12-axis score below.

The shape of this sound profile is the 2026 AI-agent signature. Premium 86, Distinctiveness 88, Modernity 84, Trust 81 clustered above 80 is the register a B2B sales agent needs to carry — the buyer is asking can I trust this with my pipeline, and the phoneme profile answers yes, this is built by people who understand premium B2B. The voiced velar /k/ onset drives Premium and Distinctiveness (the harder consonant carries gravitas); the open /a/ vowels keep Trust and Warmth in the high band; the terminal nasal keeps the name pronounceable in soft-stress contexts.

Energy 76 is intentional for an AI agent — the brand needs to read kinetic (the caravan moves) without reading consumer-product loud (a sales agent is enterprise, not commercial). Gender at 51 (neutral) is the modern default; an AI-agent brand that gendered would lose half the ICP. The Sapir/Maurer literature (Sapir 1929; Ćwiek 2022) on the velar-plus-open-vowel-plus-rhotic profile predicts the kinetic-Trust register the Karvan candidate carries — and the cross-cultural replication shows the same affect holds in Hindi, Urdu, and Persian listener panels.

Cohort distance to the 2026 AI-agent register (Replit Agent, Lindy, Manus, Devin) is 0.31 — comfortably on-cohort. The cohort signature is two-syllable, voiced-onset, open-vowel names that read modern and pronounceable; Karvan sits in the middle of that distribution.

No action required. The sound profile is on-target for the category. If you want to lean further into the heritage register, pair with a Persian-script wordmark (کاروان) on the launch site as a secondary identifier — the cross-script treatment reads as authentic in the founder's ICP markets and forgettable in the U.S. enterprise register (which is the right asymmetry).

Shortening to Karv — counter-productive. Karv (one syllable, no open vowel) loses the metaphor — the caravan needs the second syllable to carry the trade-route affect. Premium drops to 67, Distinctiveness drops to 61. Keep the two-syllable form.

Lengthening to Karvana — collides with Carvana the used-car retailer at 96% phonetic distance, which would trigger a §2(d) refusal in any class adjacent to consumer-vehicle commerce. Stay at Karvan.

Methodology: Sound Symbolism for Brand Naming.

Say Karvan to a Surat textile distributor and they'll pronounce it right the first time. The 7-language Whisper round-trip composite of 96% CER survival is the receipt — the founder's ICP speaks across English, Hindi, Urdu, Persian, Pashto, Turkish, Arabic, and the same two-syllable /kar-van/ pattern holds in all seven. The grid below shows per-language round-trip survival (1 − CER, where CER is the character-level error rate when Whisper transcribes the TTS-spoken candidate).

Pronunciation resilience is what the “say it to a Surat distributor” test measures. The founder's ICP for Karvan is concrete: South-Asian and Middle-Eastern B2B-sales ops at small-to-mid-market distributors. The buyer speaks English-Indian or Urdu in Karachi, Hindi-English in Mumbai, Persian in Tehran, Pashto in Peshawar, Turkish in Istanbul. If the brand name doesn't survive the buyer's mouth in their first language, the founder's introduction becomes a spelling lesson — a tax the brand pays on every outbound call.

The /kar-van/ phoneme pattern survives the test because the Persian root is loan-imported into all seven trade languages with the same stress-on-syllable-one structure. There is no accent-specific drift above the 7% threshold; the Arabic transliteration (كَرْفان) drops the /v/ to /f/ in some speakers, but the Whisper ASR retrievability stays in band. For voice search and AI-assistant attribution — both growing 30% YoY in B2B query volume — the Karvan candidate is at the model-ceiling.

Compare to the previous SalesRep.tech candidate: the Whisper round-trip on SalesRep.tech returned 71% CER. The compound construction (“sales rep” + “dot tech”) drifted across accents, the .tech suffix occasionally transcribed as “dot deck” or “dot wreck,” and the Hindi-English code-switched pronunciation broke the compound into “sales-rep technology” — which is the description the buyer already had. The pronunciation axis is what the original candidate failed.

No action required. Pronunciation is the cleanest of the five axes. If you want to lean into the cross-language resilience as a brand asset, ship the launch site with a /how-to-say microsite in all seven languages — but the audio sample is optional at this CER profile (brands score below 85% composite benefit from /how-to-say; brands at 96% don't).

Spelling as Caravan — collides with the English-language generic noun and the Caravan Health Class 35 holder at 100% phonetic distance. Stay with the Persian-Latinate spelling.

Methodology: Brand Pronunciation Resilience.