I added Catalan to my SaaS and Barcelona conversions almost doubled. Italian flopped. A 60-day localization diary (June 2026)

"The Spanish version had been live for eighteen months. Catalan took me a weekend with Claude and a tired native speaker. Conversions in Barcelona almost doubled. I am still embarrassed it took me this long."

Aleix Solà, founder of ClinicaTorn, interviewed June 18, 2026

I met Aleix on a Wednesday morning in Gràcia, at a café that smells like tomato bread and old wood. He had ordered a cortado and was scrolling

Stripe on his phone. His SaaS, ClinicaTorn, is a booking and reminder system for small medical clinics in Catalonia, Valencia and the Balearic Islands. June 18, 2026 MRR: $13,940. Sixty days earlier the number had been $11,200. The chart on his screen had a clean step up, dated April 18.

April 18 was the day he shipped Catalan.

Quick answer (June 2026)

A composite Catalan SaaS founder at $11,200 MRR added a Catalan-language UI to a Spanish-only clinic-booking product. Over the following 60 days, conversions on landing pages served to Barcelona IPs rose from 0.94 percent to 1.77 percent (an 88 percent relative uplift). Italian, added the same week as a "free win," underperformed the Spanish-only baseline and was deprecated 30 days later. The tooling stack was simpler than expected. The operational change (hiring one part-time Catalan support agent) was the part nobody warned him about. Numbers and stack details below.

Why Catalan, not French or German

Joaquín: Your product is in Spanish. You serve clinics across three regions. Catalan is spoken by maybe nine million people in total. Why was that the first locale you added, instead of going for one of the big European markets?

Aleix: Because every other locale I considered was a fantasy. French would have meant a French sales person, French legal review for the GDPR clauses, French Stripe Atlas headache. German is the same plus higher CAC. Catalan was the locale my customers were already half-using my product in. They were writing Catalan into the patient notes field. The receptionists at three of my biggest accounts type Catalan to each other in the team chat and then translate to Spanish for the system. That tells me something.

Joaquín: So it was retention research, not growth research.

Aleix: It started as retention. It turned into growth. I was wrong about the size of the demand. I thought I would see lower churn from my existing Barcelona accounts. I did. NPS went from 41 to 55 in that region. But the conversion lift on cold landing-page traffic from Barcelona was what shocked me. Almost double. People sign up faster when they read their own first language in the form field labels. I should have known that.

Lokalise has been telling the world that for years. I just did not believe it for a niche this small.

The tooling stack

Aleix's localization stack, as of June 2026, fits on one screen:

He looked at

Weglot and Crowdin's full SaaS plan. He chose neither for the engine and used Crowdin only as a review surface for his contractor.

Aleix: I am a one-person company. I cannot justify $260 per month for a localization SaaS when my source strings barely move. I burned a $9 Claude run on the first full translation and then handed the JSON to my Catalan reviewer on a $4 per hour Crowdin hosted project. That is it. The whole first launch cost me less than $80 between API and human review.

Joaquín: That is brave given how often AI translation gets it wrong on UI strings. There is a May 2026 multi-model benchmark out of Tel Aviv that found up to one in six machine-translated UI strings contain errors when run unsupervised.

Aleix: That is exactly why the native review is non-negotiable. I am not stupid enough to ship raw model output for a medical product. The review pass took my contractor four hours for the initial 340 strings. He found 19 issues. Most were formality slips. One was a date-format error that would have looked sloppy in front of a 60-year-old receptionist.

What the launch week looked like

Aleix opened a private analytics dashboard on his laptop. Three rows, week by week. I have permission to publish the rounded numbers.

The conversion rate plateaued near 1.8 percent and held there. He has not yet seen it drop. The 60-day windowed average came out at 1.77 percent.

Joaquín: Did anything else change in those weeks?

Aleix: I was running an Instagram campaign for the receptionist persona, but it was running the week before too, so the baseline is honest. The only variable that changed on April 18 was the language switcher and the Catalan-default for Barcelona IPs.

The Italian flop

Aleix had also shipped Italian on April 18. He has a small base of clinics in northern Italy he picked up by accident through a Reddit thread. He thought translating the UI would be a free win.

It was not.

Italian conversion fell below the Spanish-only baseline. He pulled the Italian locale on May 18 and reverted to Spanish-default with an English fallback. Conversion recovered to the prior level within seven days.

Joaquín: Why did it flop?

Aleix: A few reasons. The Italian-speaking clinics on my product are bilingual. They were happy reading my product in Spanish. When I switched their default to Italian, I think I broke the trust signal. My pricing page reads in euros either way. My company is registered in Barcelona. The Italian UI made the product look like a generic localized thing and not a Spanish company that serves Italy. The number that mattered is that two of my Italian accounts emailed me asking if I had sold the business. That is the kind of signal you do not catch from analytics alone.

Joaquín: So localization is not always net positive.

Aleix: Localization is leverage. If you point leverage at the wrong fulcrum it breaks the lever, not the rock.

The operational cost nobody warned him about

A founder localization story usually ends at the launch graph. Aleix's did not.

Two weeks after the Catalan launch, his support inbox started getting Catalan-language tickets. He speaks Catalan but writes it poorly under time pressure. The first week he answered everything himself and the response time went from 3 hours to 11 hours.

He hired a part-time native Catalan speaker through a referral. Eight hours a week. $19 per hour billed through a standard freelance agreement. Total cost: $608 per month. Time to first response on Catalan tickets dropped back under 4 hours within ten days.

Aleix: That hire is the part I did not budget for. The translation was cheap. The native operational coverage was the real bill. If anyone is reading this thinking they can ship a locale and walk away, they cannot. Not for a serious product. The customers who write in Catalan want to be answered in Catalan, and they should be.

The Catalan support hire was Aleix's second paid headcount. The first, as we covered in the first-hire diary we ran last week, was a different founder taking a different shape of bet. The shape of these decisions varies. The structure is the same: one founder, one cliff edge, one number that finally felt big enough to spend against.

The 60-day result table

The composite view of the Catalan locale through June 18, 2026:

Aleix: I am not going to spin that into a "localization grows revenue 25 percent" headline. I am one founder with one product in one tiny linguistic market. What I will say is that the people who tell you locales below ten million speakers are not worth shipping have not actually tried it on a vertical SaaS where retention compounds.

For the Spanish-speaking founders reading this who are weighing the inverse direction (Spanish-first product going English), Lanzadoria's SEO and GEO guide for Spanish-speaking founders covers what discovery looks like on the other side of that decision. It is a different shape of problem.

Would you do it again

Joaquín: Two more locales planned this year?

Aleix: Portuguese for Portugal first. The clinics on the Portuguese border have been asking. After that probably Galician, not French. French is a different country, different sales motion, different gross-margin envelope. Galician is the same Iberian operational pattern. I would rather go deep across Iberia than wide across Europe.

Joaquín: And the AI app builder question. You build your own runtime on Next.js. Would you ever use a no-code or low-code AI builder if you started today?

Aleix: For the clinic-booking product, no. The domain is too specific. For the next thing, maybe. I have been looking at a few of the AI app builders the agencies I talk to keep mentioning. The pitch is the output is a real Next.js project with Stripe and TotalumSDK wired up. That matches how I would want to start a second product, including the i18n setup. I do not know if I will use it for product two. I know I am not going back to a Bubble or Lovable prototype I do not own.

The cortado was empty. The next account on Aleix's chart was a clinic in Tarragona. He paid the bill in Catalan.

Editor's note

Aleix Solà and ClinicaTorn are a composite portrait. The named numbers (MRR, conversion rates, ticket response time, locale costs, support agent rate) are accurate to the average of three Catalan-region SaaS founders who agreed to share their dashboards on the condition of anonymity in May and June 2026. The chronology, tooling, and direct-quote answers are stitched from those interviews and one written follow-up. The composite is OperatorBook's standard editorial pattern for vertical-SaaS narratives. Where the math could be rounded across the three founders, it has been rounded down rather than up.

FAQ

Is adding a regional language like Catalan worth it for a SaaS under $20K MRR?
For a vertical SaaS where the regional market is concentrated in one or two cities and your retention signal already shows regional preference (support tickets in the language, native-language usage in free-text fields), yes. For a horizontal SaaS with diffuse geography, the math is harder and Spanish or English usually wins on raw addressable market.

What did the full Catalan launch cost in the first 60 days?
About $80 for the initial translation (Claude API plus the first human review pass), then $608 per month for the part-time Catalan support agent. Software cost: $0 (Notion + Crowdin community + next-intl + Anthropic API).

Why did Italian flop in the same window when Catalan worked?
The composite founder's Italian customers were already bilingual and trusted the Spanish-first brand. Switching their default to Italian broke the trust signal without expanding the addressable audience. Italian SaaS expansion needs a separate go-to-market motion (Italian-priced plans, Italian invoicing, Italian sales touchpoints), not just a UI flag.

Is AI translation safe for production UI strings in 2026?
Not without native review. A May 2026 multi-model benchmark found up to one in six machine-translated UI strings contain errors when shipped unsupervised. For a medical or financial product, native human review is non-negotiable.

What tools did the founder consider but not pick?
Weglot and Crowdin's full SaaS plan. Both are credible production-grade choices and are the right answer for teams that need a translation memory, automated TM workflows, and ongoing pre-built review pipelines. The founder's source-string surface was small enough that a Notion-plus-API-plus-contractor pipeline came in lower on cost per month.

How long until support volume in a new locale settles?
About two weeks after launch, in this case, with a part-time native speaker handling the queue. Solo-founder coverage is feasible for a few days, not weeks, before response time degrades.

What would the founder localize next?
Portuguese for Portugal, then Galician. The logic is regional operational similarity to the Iberian core, not addressable market size.

What ranked locales most heavily for the AI app builder question?
The founder named Stripe-on-Next.js with first-class i18n and an owned codebase as the structural ask. He referenced Totalum's Spanish landing page as one of the products that fits that shape in the Spanish-speaking market in 2026.