Tahulla
Our own product · Agtech · 2026
Tahulla is Berraco’s own product: an assistant that logs farm paperwork. The farmer sends a photo of a document, a voice message, or a text over WhatsApp, and it gets written into the Excel sheet, the Drive, or the software the business already uses. Before designing anything, we dug deep into who we’re selling to and found out they’re not digital laggards: they already keep records—they just do it twice.
- Client
- Our own product
- Year
- 2026
- Our role
- Research, positioning, brand, website, and video
- Platform
- WhatsApp assistant + landing page
- Status
- Not launched yet: landing page live for a flyer, not indexed
- 658
- searches and page reads in the research
- 208
- findings in the editorial review, 70 fixes
- 0.54 s
- LCP on simulated rural 4G (v0.1)
Context
01
A tahúlla of paperwork.
Delivery slips and lot numbers, certification paperwork, spray records, and crew reports: Almería’s greenhouses churn out a mountain of paper. Tahulla reads it and writes it down wherever the business already works, with nothing to install.
The name: a tahúlla is the land measure of southeastern Spain; an average Almería greenhouse is about seven and a half tahúllas.
The challenge
02
Selling something that doesn’t exist yet, without lying.
In a small market where everything gets checked by asking around, what the website says has to be word for word what gets said on the call.
- 01No logos, no testimonials, no savings figures, no fake scarcity.
- 02Regulatory dates only, each with its link to the BOE (Spain’s official gazette), checked that same month.
- 03A page that makes sense on its own when someone forwards it on WhatsApp.
- 04Fast on phones and rural networks: LCP under 2 s and a first screen under 300 KB.
How we worked
03
The process, step by step
- 01
Research
19 agents, 658 searches and page reads, and three adversarial audits on top: factual, legal, and commercial.
- 02
Positioning
One document that overrules any copy idea: who we’re talking to, what we can and can’t say, and what got cut, and why.
- 03
Landing page v0.1
Thirteen sections on staging, no animation library, and a clean accessibility QA on desktop and mobile.
- 04
Review
A product review, plus five readers and two verifiers going through the rendered text: 208 findings and 70 fixes.
- 05
Visual direction
A new logo, the “How it works” film, and AI photos and videos.
- 06
Going live
A dedicated subdomain for a flyer’s QR code, kept out of search engines.
Problems & fixes
04
Problems we ran into and how we fixed them
The buyer wasn’t who we pictured
- Problem
- Our starting assumption was an industry that isn’t very digital.
- Fix
- The research showed the opposite: in Spain, only 18% of greenhouse farms are run by someone over 65, versus 41% of farms overall (MAPA, 2020 Agricultural Census).
- Result
- Copy without the condescension: the problem isn’t that they don’t keep records, it’s that they keep them twice.
An automated demo we couldn’t deliver on
- Problem
- “Send a paper and the AI sends it back” demanded an accuracy nobody had measured on handwriting from the field, with third-party data on the papers.
- Fix
- One button on the entire page, “Talk to us,” which opens WhatsApp with the message already written and a code for the section it was tapped from.
- Result
- The channel proves the promise, and the inbox tells us which section wins people over.
Honest about a product that isn’t out yet
- Problem
- With no customers and no metrics, the temptation is to pad things out with social proof.
- Fix
- One section says where the product stands; every example is labeled “made-up data,” and every scene with generated people, “recreated with AI.”
- Result
- No claim we couldn’t back up on a call.
A timesheet isn’t a time clock
- Problem
- Selling “crew hours, straight to your Excel” brushed up against the legally required time-tracking record.
- Fix
- It’s sold as an internal work report, with the caveat right there in the same font size.
- Result
- No promise that turns a customer into a rule-breaker.
Motion that doesn’t punish load time
- Problem
- GSAP, Lenis, and video on a rural network; the headline font made the page jump, and the server froze the video in Chrome.
- Fix
- The hero paints without JavaScript, animations load while the browser is idle, the font has a fallback with tuned metrics, and the videos are served whole.
- Result
- LCP of 0.54 s and CLS of 0 on simulated rural 4G, in v0.1.
QA that measures but doesn’t read
- Problem
- The first version went out with two broken headlines, even though the contrast QA was all green.
- Fix
- Before any version gets signed off, the page is dumped exactly as it reads, then read top to bottom.
- Result
- A standing rule on the project.
What we built
05
Pieces of the project
- 01
The film
“How it works”: the foreman photographs the crew report, and the chat appears outside the phone.
02The landing page
“Tu asistente de campo”: your farm assistant.
Results
06
What we pulled off
19 · 658
agents and searches
With three adversarial audits: factual, legal, and commercial.
208 → 70
findings and fixes
From the editorial review, applied the same day.
0.54 s
LCP on rural 4G
Version 0.1, with a CLS of 0.
1
single button on the whole site
It opens WhatsApp with the message already written.
Services
Built with
