Skip to main content
veles

// Case study · AI Booking System

Currently in beta

An Instagram inbox that runs itself, until the owner says otherwise.

How we built a custom booking AI for a 100K-follower Belgrade auto-styling brand, with one tap that gives the owner back the conversation.

ByIgor Medic and Nikola KosaninPublished Updated 5 min read
  • AI Booking System
  • Instagram Automation
  • Multi-Worker Scheduling
  • Belgrade Client

The client

Kacimbas Autostyling. Belgrade-based car styling and lights specialists. Strong Instagram presence: 100,000+ followers built on real before/after work. Bookings come almost entirely through Instagram DMs.

The problem

Volume had outpaced the owner's ability to respond. The math no longer worked:

  • Routine inquiries (pricing, hours, "do you work on BMWs") were eating hours of the day
  • VIP customers, the ones who matter most, were waiting longer for replies
  • Generic chatbot tools were off-brand and impersonal
  • Outsourcing replies meant losing the personal voice that built the following in the first place

Lead-response research has been definitive for over a decade: firms that contacted prospects within 5 minutes were 100 times more likely to make contact than those that waited 30 minutes, and companies that responded within an hour were 7 times more likely to qualify a lead than those that waited even an hour longer (Oldroyd, McElheran & Elkington, Harvard Business Review). For a brand whose entire booking funnel is an Instagram inbox, every delayed reply is a missed booking.

The owner needed something the off-the-shelf tools could not provide: automation that knows when to step back.

Our approach

We designed a custom mobile-first system around a single principle: the AI handles the routine, the owner handles the relationships.

Customer-sentiment research validates that responsibility split. Only 19% of US consumers say they find customer-service chatbots favorable, and the number drops to 7% among people over 55 (CivicScience 2024 chatbot reviews). A fully automated bot would have eroded the personal voice that built the 100,000-follower following. A hybrid model with a one-tap human takeover preserves it.

Conversation intelligence. Every Instagram message gets routed through a dozen specialized AI agents, each trained for one job: booking inquiries, cancellations, vehicle questions, pricing, returning-customer recognition. Conversations flow naturally in Serbian, using the formal "Vi" address customers expect from a professional brand.

Customer memory. The system remembers every customer, every vehicle, every previous service. When a returning client messages, the AI can open with: "Zdravo Marko! Vidim da ste bili kod nas prošlog meseca za xenon svetla na vašem BMW-u..." (roughly: "Hi Marko! I see you were here last month for xenon lights on your BMW..."). By name, by car, by service history.

Multi-worker scheduling. The shop has multiple specialists, one for xenon installations, one for window tinting, one for general work. The calendar engine assigns bookings to the right specialist based on the service requested, worker availability, and historical preferences.

The Take Over button. This is the feature the off-the-shelf tools could not deliver. The owner sees every active conversation in real time. One tap moves a chat from AI to manual mode, the AI steps aside, the owner types personally. One more tap hands it back. No conversation is ever locked into automation.

What we built

  • Custom mobile workstation for the owner
  • Instagram messaging integration with Meta-compliant guardrails (24-hour messaging window enforced automatically)
  • Conversation memory and customer recognition system
  • Multi-worker calendar with service-duration awareness and conflict detection
  • Before/after photo gallery tied to each customer's vehicle history
  • Push notifications for VIP customers and edge cases
  • Full Serbian-language interface

What's next

The system is currently in beta with the owner. Once Instagram operations stabilize, the same architecture extends to:

  • WhatsApp. The dominant business-messaging channel in Serbia.
  • Viber. Popular with older demographics in the Balkans.

The conditions for that rollout are already in place. 91.2% of Serbia's population is online and 72.1% are on social media (DataReportal Digital 2025: Serbia), and across Eastern Europe and the Balkans Viber complements WhatsApp as a dominant messaging channel (Infobip 2026 messaging apps by country).

Same conversation memory, same customer profiles, same Take Over button, across every channel a customer might reach out on.

A brand with 100,000 followers doesn't have a marketing problem. It has an operations problem: how do you preserve the personal touch that built the following, while keeping up with the volume that following created? That's the kind of problem Veles solves. Automation that respects the relationship, designed around the operator, not against them.

// FAQ

Frequently asked

How long did the Kacimbas system take to build?
The initial build moved from discovery to working software in production-ready beta within a focused multi-week project. Once the architecture was settled, individual features (new AI agents, additional message types, new languages) ship in days, not weeks.
Does the same architecture work outside Instagram?
Yes. The conversation layer, customer-memory, multi-worker scheduling, and Take Over button are channel-agnostic. WhatsApp and Viber integrations are the next planned extensions, using the same data layer and the same operator app.
What languages does the AI handle?
Serbian is the primary trained language, including the formal "Vi" address and Balkan-dialect variations. The same underlying model handles English, German, and any other locale a customer might message in, with the brand voice rules applied per language.
Can the AI be turned off completely for specific customers?
The Take Over button locks individual conversations to manual mode. VIP-flagged customers can also be configured to always route directly to the owner, bypassing the AI entirely.
How does this compare to off-the-shelf chatbot tools?
Off-the-shelf chatbots are built for English-language customer support flows and feel impersonal to a 100,000-follower brand audience that expects a human voice. Kacimbas runs on a custom dozen-agent system trained on the brand's actual booking patterns, in Serbian, with a one-tap human takeover path off-the-shelf tools cannot provide.
What does a custom build like this cost?
Custom AI tooling is quoted per project against scope. We work in fixed-price phases (audit, sprint, partnership) rather than open-ended monthly retainers. Book a 30-minute discovery call to discuss your operations and get a scoped quote.

// More case studies

Related work

Want one of these built for your business?

Tell us about your operations problem. We'll come back with a real plan, not a sales deck.

book a 30-min audit

We use cookies to analyze site traffic and improve your experience. Privacy Policy