- +340%
- Phone calls and messages answered/month
- 22
- Front-of-house hours saved per location/week
- 11 days
- Time-to-deploy first workflow
- +28%
- Weekend bookings (after 60 days)
Anonymized representative case based on patterns we see in this segment. Not tied to a specific client.
Background
The operator runs five bakery-café locations in two Iberian cities, with combined daily covers in the low four figures and a takeout business that grew faster than the front-of-house could absorb. The group was traditional in its tooling — a calendar shared across locations, a phone tree that rolled to whichever location had a free host, and a WhatsApp number that one manager monitored personally.
The team had tried two chatbot vendors in the previous three years. Both were retired within six months. Customers detected the script inside two messages and switched back to calling.
Challenge
Three problems compounded during service hours:
- Calls went unanswered. Internal audits showed roughly half of inbound calls during peak (Friday 19:00-22:00, Saturday lunch, Sunday brunch) rolled to voicemail or were dropped.
- The WhatsApp number was a bottleneck. One manager handled it, which meant nights and weekends were silent.
- Takeout intake was manual. Orders came by phone, were written on paper, retyped into the POS by a host, and routinely produced errors at the kitchen pass.
The group's COO had a clear constraint: the kitchen could not change, the wage bill could not grow, and the rollout could not require a multi-month integration project.
What Nordix BIOS did
Deployment was staged across two weeks.
Days 1-4: BIOS was connected to the group's existing reservation calendar (read-write) and POS (read-only, then read-write after a week of monitored use). A policy layer was configured: deposits required for parties of 6+, allergen handling that refuses bookings the kitchen cannot safely serve, escalation to a human for any complaint or refund request.
Days 5-8: Shadow week. The agent drafted replies on the WhatsApp number; a manager approved every send. The team collected the 10% of edge cases that mattered — a regular customer with a custom standing order, a press inquiry, a local school asking about field trips.
Days 9-11: Go-live at one location, WhatsApp first. Voice was added at day 14 once WhatsApp metrics held.
By day 21 all five locations were live on WhatsApp and voice. Takeout intake (a separate workflow) shipped at day 38. Supplier coordination shipped at day 55.
Outcomes
After 60 days of full production:
- Phone calls and messages answered/month: +340%. The agent handled everything inside business hours and most of what came in outside them. The "phone goes to voicemail" problem effectively disappeared.
- 22 front-of-house hours per location per week reclaimed. Hosts spent that time on the floor, not on the phone. The COO did not cut staff; the existing team handled measurably more covers per shift.
- Time-to-deploy first workflow: 11 days. The group's CTO described the integration as "the first IT project in this company that finished on the day it was supposed to."
- Weekend bookings: +28%. Most of the gain came from late-night and Sunday-morning bookings that no human had been available to take.
A secondary effect: the no-show rate dropped from a previous baseline of ~14% to ~6%, attributed to deposits on large parties (enforced by the agent) and automated reminders 24 hours before service.
Lessons
Three patterns from this deployment apply broadly:
- Pick one workflow, ship it clean. The group's previous chatbot attempts failed because they tried to do everything. BIOS started with reservations only and expanded after that one was boring.
- Hard policy beats clever language. The biggest source of agent errors during the shadow week was edge cases the kitchen could not handle (allergens, large parties at peak). The fix was not better prompts; it was hard refusal rules.
- Voice matters more than expected. WhatsApp got the headlines, but voice — handled by the same agent on the same policies — was where the older customer base reached the business. Skipping voice would have left a meaningful slice of revenue on the table.