Palmera Dining:
The chat staffed itself with Owl.

Palmera Dining is a popular Nairobi restaurant — Mediterranean menu, two private dining rooms, a steady catering side line. Their existing site listed a menu and a phone number. Everything else happened in DMs. By 9pm the front-of-house team was hours behind on Instagram replies, and the catering inbox was a graveyard of leads that had cooled by the time anyone got to them.

The problem

Three places where money was leaking:

• Late-night enquiries went unanswered. Customers asking about birthday packages at 11pm got nothing until the next morning, by which time they'd booked elsewhere.
• Catering leads required a human triage. "What's the minimum order?" "Do you do gluten-free?" — five basic questions before the conversation got to dates and headcount.
• The team had no view of what people were asking. The same dietary question got asked 40 times a week; nobody wrote it on the website because nobody was tracking the pattern.

The brief

"Give us back our evenings. Capture the leads we're losing after hours. And help us see what people actually want."

What we built

• Owl trained on Palmera's full menu and catering pack — every dish, every dietary tag, every package tier, every FAQ the team had been answering by hand. Owl learns the site keylessly; it doesn't need rebuilding when the menu changes. See how Owl works.
• Conversational reservations — Owl asks date, party size, dietary needs, special-occasion notes, then hands the structured lead to the front-of-house WhatsApp with the customer's number.
• Catering qualification flow — Owl handles the first five questions (date, headcount, location, dietary, budget band), captures the lead, and only escalates qualified ones to the catering manager.
• Owner dashboard — Sandra (general manager) sees the leads, the top questions of the week, and which dishes Owl is being asked about most often. She turned that signal into menu copy changes after week two.
• Trends panel — Owl surfaces what's bubbling in Nairobi food chatter so the catering team can pitch with the season, not against it.

The architecture (for the developers reading this)

Owl's keyless retrieval brain (owl-nlp.js + owl-brain.js) ingests the Palmera site and the catering PDF on deployment, sits in front of the chat widget, and falls back to the keyless brain when the optional Claude upgrade is offline. Lead capture goes into the Owl owner dashboard and pings the front-of-house WhatsApp via the WhatsApp Cloud API. Sessions are 1-hour in-memory; no PII is stored beyond the lead record itself.

The timeline

• Day 1: Owl trained on the live site + catering pack; chat widget embedded; lead capture wired to the front-of-house WhatsApp.
• Day 2: Sandra and the catering manager spent 90 minutes role-playing customer questions; we tuned the answers and added five FAQs Owl didn't have yet.
• Day 3: Soft launch — chat live for 48 hours, watched closely, three answers refined.
• Day 5: Hard launch with social announcement.

The result

After 60 days:

• 2,400 customer questions answered by Owl. No human shift on the chat.
• 312 leads captured — reservations, catering enquiries, private-dining bookings. Each one with a phone number, intent, and one-paragraph context.
• 68% conversion on the reservation page — visitors who hit the page and engaged Owl became a captured lead two-thirds of the time.
• Top-10 questions list driven onto the site as proper FAQ content. The same questions then stopped being asked, because the answer was now on the page.
• Late-night leads stopped vanishing. Owl captured them, the team picked them up at 9am the next morning, the customer was already qualified.

What this build is worth, if you need something similar

An Owl deployment on a Kenyan restaurant, retail or services site — training, embed, lead routing to WhatsApp, owner dashboard, 30 days of tuning — is typically KES 95,000–180,000 depending on the depth of content Owl learns. The recurring cost is light: Owl runs keylessly by default; the optional Claude upgrade is metered if you want it. See how Owl works.

← All case studies · Previous case study: Bridgerose Academy →