Case Study
        

July 8, 2026

Bringing Agentic AI Into Customer Service

How TeraSky helped a leading Israeli beverage distributor turn approved internal knowledge into a secure Hebrew-language Copilot Agent.

 

Agentic AI becomes most valuable when it can operate inside the systems a business already trusts.

 

For one beverage distribution company, that meant exploring whether AI could support real customer service workflows without creating a new knowledge source, bypassing existing security controls, or asking representatives to change the tools they already used.

 

TeraSky helped the firm build a secure Copilot Agent that made approved internal knowledge more accessible, more consistent, and faster to act on in a Hebrew-language service environment.

 

The Challenge

 

A leading Israeli beverage bottler and distributor manages a high-volume consumer service operation across product information, kashrut, expiration dates, quality issues, marketing campaigns, and logistics.

 

Representatives relied on SharePoint as the authoritative source for approved answers, but finding the right information still required manual search. In a live service environment, that created delays, inconsistent phrasing, and a steep learning curve for new team members.

 

The company needed a secure way to surface approved answers in seconds, in Hebrew, while preserving its existing Microsoft 365 identity and access model.

 

The Solution

 

TeraSky designed and delivered a Microsoft Copilot Studio agent connected to the organization’s existing Consumer Service SharePoint site. The goal was to give representatives fast access to approved answers while keeping the solution inside their Microsoft environment.

 

The agent was built around three requirements:

 

First, it had to use only approved knowledge. SharePoint remained the authoritative source, and the agent retrieved read-only information from the Consumer Service site. TeraSky configured a Service Principal with Sites. Selected read-only permissions, limiting access to the specific content the agent was approved to use.

 

Second, it had to work within the company’s identity and security model. Representatives authenticate through Microsoft Entra ID SSO. When a Conditional Access conflict blocked the expected flow, TeraSky worked with the IT team to resolve it without weakening the organization’s Microsoft 365 security posture.

 

Third, it had to respond consistently in Hebrew. TeraSky used advanced prompt engineering and a custom system prompt override to make context-specific rules take priority over global model defaults. This helped ensure the agent responded in Hebrew and stayed grounded in the customer service use case.

 

The Impact

 

The POC showed clear operational value.

 

Answer retrieval time dropped from minutes to seconds. In structured testing, the agent consistently produced accurate, context-enriched responses, with most answers preserving or matching the intended meaning of the approved source content.

 

A company representative added, “Beyond answering correctly, the agent surfaced knowledge gaps in the underlying SharePoint content, giving our content team a concrete action list for ongoing improvement. That made the project especially valuable. It solved the immediate service need, but it also created something we can keep improving over time.”

 

Just as importantly, the architecture created a reusable model for governed agentic AI inside the organization. The same approach can be extended to other departments, such as Sales or HR, with minimal rework.

 

Technologies Used

 

Microsoft Copilot Studio | Microsoft Power Platform | Microsoft Entra ID | SharePoint Online | OpenAI models | Microsoft 365

For more information

Tags:
AI
Microsoft
Gen AI
Copilot
Agentic AI
Agent
Copilot Studio
Microsoft 365
Share:

Next Articles

Case Study
      

15 July, 2026

Accelerating Business Velocity with Hybrid Cloud
Read Case Study
Case Study
      

13 July, 2026

LLM-Powered Build Analytics
Read Case Study
Case Study
      

1 July, 2026

The Federate-to-Automate Shift: Modernizing 10,000+ Defense Desktops
Read Case Study
Skip to content