How TaxBox automated 70% of courier tickets on sensitive tax questions
~70%
Of courier tickets handled by the system
37%
Fully automated resolutions
31% more resolved with an agent in the loop
2 weeks
To first results
From signed agreement to a live workflow
Accounting for Ukraine's couriers, built into their workflow.
TaxBox automates sole proprietorship (FOP) accounting in Ukraine - taxes calculated, filings submitted, every status visible to the user. In partnership with Glovo, TaxBox registers couriers as sole proprietors, verifies their tax regime, and handles payments and filings on their behalf.
The audience makes this case unusual: not customers asking about orders, but couriers asking about their own money, tax status, and the risk of losing their Glovo account.
TaxBox × Glovo
Repetitive questions, irreversible consequences.
The difficulty is not the variety of questions. It is their nature.
Glovo's courier base is large and constantly turning over, so ticket volume into TaxBox support stays consistently high - and mostly repetitive: who the sole proprietorship is registered to, who can see the money, how taxes are debited, whether a courier can leave the service.
These are financially sensitive topics where an imprecise answer erodes trust. Scaling support with people is inefficient - and the team had no systematic view of what couriers actually asked most often.
An AI workflow that lives inside the helpdesk.
Obriy AI deployed an AI workflow for TaxBox on the SURE platform, integrated directly into the client's helpdesk. The system receives a courier ticket, categorises it, looks for an answer in the knowledge base, and either responds itself or escalates to a human agent - performing every action inside the helpdesk.
Categorisation
The system receives the courier request and identifies the topic: registration, taxes, filings, access and security, changes to an existing setup, leaving the service.
Knowledge lookup
The agent queries the TaxBox knowledge base and checks whether it holds information sufficient to answer this topic.
Answer or escalation
If the knowledge is there, the courier gets an automated answer. If not, the ticket is escalated to a human agent.
Actions in the helpdesk
The agent performs every operation inside the client's helpdesk: replying in the ticket, changing flags, moving the ticket into the human queue.
A knowledge base your team owns.
A living knowledge base in Google Docs
The agent's knowledge source is a tabbed Google Doc. To add new knowledge or update existing information, the team simply edits that document - no deployments, no development tickets, no requests to Obriy AI. The source is dynamic, and the TaxBox team maintains it themselves in a tool they already use every day.
Reporting that compounds automation.
Obriy AI gives the client an up-to-date picture of what couriers are asking and exactly where the agent ran out of knowledge. The TaxBox team promptly extends the Google Doc, and the automation rate climbs with every iteration.
Report → extend the knowledge base → higher automation. This loop drove most of the performance gains.
TaxBox
Nearly 70% of courier tickets are handled by the system. Everything else is escalated automatically, so no ticket goes unhandled.
What changed for the business
- Couriers get answers to sensitive tax and registration questions immediately, without waiting in a queue.
- Human agents are freed from repeating the same explanations and work only on tickets that genuinely need a person.
- The knowledge base evolves under the TaxBox team's own control: edit the Google Doc, and the agent answers accordingly.
- Topic reports give the client a precise view of the gaps, so automation grows deliberately rather than by guesswork.
- Platform
- SURE by Obriy AI
- Architecture
- AI ticket-handling workflow: categorisation, knowledge lookup, escalation
- Integration
- Direct integration into the client helpdesk: replies, flag changes, ticket tagging
- Knowledge source
- Tabbed Google Doc - a living knowledge base maintained by the TaxBox team
- Analytics
- Regular reports on ticket topics and knowledge gaps
- Time to first results
- Two weeks from signing; current performance reached in two months
- Deployment
- Fully delivered by the Obriy AI team
Ready to size up automation on your own tickets?
See how Routines → keep every step of the process consistent - no matter who asks.