IskayPet says its Amazon Bedrock assistant cut operational-answer lookup time from three to five minutes to about seven seconds across four languages.
IskayPet has deployed an employee knowledge assistant built with Amazon Bedrock Managed Knowledge Bases, according to an AWS case study published October 22, 2026. The pet-care company operates across Spain, Portugal and France, and the assistant is designed to answer staff questions about retail operations, veterinary care, onboarding and procurement in Spanish, Portuguese, French and English.
The system draws on the company’s existing Microsoft SharePoint documents rather than requiring a repository migration. IskayPet configured a separate knowledge base for each assistant, scoped to the relevant SharePoint sites, libraries or folders. That structure is intended to keep information for different business areas and countries separated while allowing new assistants to be added without redesigning the system.
When an employee asks a question, the application uses the employee’s identity during retrieval. AWS says the setup applies existing SharePoint permissions both to which assistants employees can see and to which documents can inform an answer. Retrieved passages are searched using both semantic and keyword methods, then used to generate a response with source references. The prompt also directs the model to say when the retrieved material does not contain an answer.
The team enabled advanced indexing to make content in images and scanned material within supported documents searchable. AWS Managed Knowledge Bases handles ingestion, parsing, embeddings, storage and retrieval, reducing the infrastructure IskayPet must operate itself. For each query, the assistant retrieves 25 passages and can include up to 15 in its prompt after removing duplicates and fragments that add little context.
AWS reports that, during roughly the first two months in production, the time to find an operational answer fell from three to five minutes to about seven seconds. IskayPet also reported around 95% less time spent searching for procedures and policies. The system is expected to handle roughly 100,000 monthly queries across its business areas and countries. IskayPet built the application and AWS deployment using Kiro with Claude Haiku 4.5; the company is evaluating prompt caching and batch inference as possible cost optimizations.