AI Agents That Cut Response Times by 60%
Client
A global education services company operating in more than 30 countries, serving thousands of business and university clients across a growing portfolio of educational products, certifications and programmes.
About
Industry
Education and professional certification services
Company size
10,000+
Client Market
Global
Location
USA
Services Offered
- AI Development
- AI Agents
Challenge
Product knowledge trapped inside growing documentation
As the product catalogue and international footprint expanded, customer success teams spent longer finding answers than delivering the advisory work their clients actually needed.
With an expanding portfolio of educational products, certifications and programmes spanning multiple countries and languages, customer success teams had to search extensive documentation and product catalogues to answer client inquiries. That manual searching created a bottleneck: slow responses to sales inquiries meant deals were lost to faster competitors, and each new market or language required significant staff ramp-up time before anyone could answer questions confidently.
The company had already tried hiring more customer success staff and standing up offshore support centres, but neither solved the fundamental inefficiency: knowledge was trapped in complex documentation rather than reaching the people who needed it. With the client base and product complexity both growing, the business could not simply hire its way out of the problem.
Their customer success teams knew the products and the clients well; what they lacked was a way to surface the right answer from thousands of pages of documentation in seconds.
Solution
A purpose-built assistant to resolve custom querries
A working proof of concept delivered in two weeks
Zartis’s AI consultants partnered with the customer success and product teams from day one, building the solution from scratch as an embedded innovation team rather than following a standard requirements-to-build handoff. The two-week proof of concept demonstrated technical feasibility and business value before further investment was committed.
An embedded team built the assistant around real workflows
The team worked to understand existing documentation, customer inquiry patterns and day-to-day customer success workflows before writing any code, rather than plugging a generic model into existing content. That grounding shaped every subsequent technical decision, from model selection to how information was structured.
Architecture designed for multiple languages and markets from day one
The technical architecture was built to support additional geographies and languages without a rebuild, and to flex as the client’s product portfolio continued to evolve. That scalability is what later enabled deployment into new markets in weeks rather than months.
The Zartis Approach
What we delivered
Custom-Built LLM Architecture
Off-the-shelf chatbots could not balance the accuracy, speed and cost this use case demanded, or scale cleanly across the geographies and languages the client needed to support from day one.
Problem: Generic chatbot platforms could not meet the accuracy, cost and scale requirements.
Solution: Selected and configured the optimal LLM for the use case, with architecture built to support multiple markets.
- Selected the LLM balancing accuracy, speed and cost
- Designed architecture to support multiple geographies and languages
- Built in flexibility for an evolving product portfolio
Restructuring documentation against hallucinations
Plugging AI into existing documentation as-is would have carried the same inaccuracies into every answer, so the underlying information had to be reorganised before the assistant could be trusted.
Problem: Existing documentation was not structured for reliable AI retrieval.
Solution: Reorganised and tagged information so the model could access and serve it reliably.
- Restructured how information was organised and tagged
- Reduced inaccurate responses through better information access
- Built feedback loops from real customer success usage
Actionable Analytics Layer
A virtual assistant that only answers questions leaves insight on the table, so inquiry-pattern analysis was added to turn usage data into a strategic asset for leadership.
Problem: Inquiry data was not being captured or used to inform strategy.
Solution: Built intelligence to analyse inquiry types, content gaps and volume trends.
- Analysed inquiry types, content gaps and volume trends
- Gave leadership data to guide documentation priorities
- Turned the assistant into a source of business intelligence
Business Impact
Response times cut by 60&
New markets launched in weeks, not months
Teams freed for advisory, not searching
Reliable answers built trust in the tool
Proactive selling replaced reactive searching
In Numbers
What we delivered
60%
Faster handling of customer success inquiries, freeing teams for higher-value advisory work
10X
Faster deployment of the virtual assistant into new languages and international markets
2 Weeks
To deliver a working proof of concept that proved technical feasibility and value
