This post was written with Lior Heber and Yarden Ron of Skai.
Skai (formerly Kenshoo) is an AI-driven omnichannel advertising and analytics platform designed for brands and agencies to plan, launch, optimize, and measure paid media across search, social, retail media marketplaces and other “walled-garden” channels from a single interface. By unifying data from over 100 publishers and retail networks, Skai applies real-time analytics, predictive modeling, and incremental testing to surface budget and bidding recommendations, connect media spend to sales outcomes, and reduce channel silos, giving marketers full-funnel visibility and higher return on ad spend at scale.
Skai recognized that our customers were spending days (sometimes weeks) manually preparing reports, struggling to query complex datasets, and lacking intuitive visualization tools. Traditional analytics platforms required technical expertise, leaving many users overwhelmed by untapped data potential. But through the partnership with AWS and adoption of Amazon Bedrock Agents AI assistants that can autonomously perform complex, multi-step tasks by orchestrating calls to APIs, we’ve redefined what’s possible. Now, customers can analyze their data in natural language, generate reports in minutes instead of days, and visualize insights through natural language conversation.
In this post, we share how Skai used Amazon Bedrock Agents to improve data access and analysis and improve customer insights.
Challenges with data analytics
Before adopting Amazon Bedrock Agents, Skai’s customers accessed their data through tables, charts, and predefined business questions. Campaign manager teams, looking to do deep research on their data, would spend around 1.5 days a week preparing static reports, while individual users struggled to connect the dots between their massive amount of data points. Critical business questions, like where should a client spend their time optimizing campaigns, and how, remained hidden in unstructured knowledge and siloed data points.
We identified three systematic challenges:
- Time-consuming report generation – Grids display flat and grouped data at specific entity levels, like campaigns, ads, products, and keywords. However, gaining a comprehensive understanding by connecting these different entities and determining relevant time frames is time-consuming. Users must manipulate raw data to construct a complete narrative. Summarization – Analyzing extracted raw data posed significant challenges in understanding, identifying key patterns, summarizing complex datasets, and drawing insightful conclusions. Users lacked intuitive tools to dynamically explore data dimensions, hindering their ability to gain a holistic view and extract crucial insights for informed decisions. Recommendations – Presenting data-driven recommendations to stakeholders with varying understanding requires deep data analysis, anticipating perspectives, and clear, persuasive communication to demonstrate ROI and facilitate informed decisions.
How Celeste powered transformation
To address the challenges of time-consuming report generation, the difficulty in summarizing complex data, and the need for data-driven recommendations, Skai used AWS to build Celeste, a generative AI agent. With AI agents, users can ask questions in natural language, and the agent automatically collects data from multiple sources, synthesizes it into a cohesive narrative with actionable insights, and provides data-oriented recommendations.
The Skai Platform absorbs an enormous amount of data about product searches across many retailers and traditional search engines. Sorting through this data can be time-consuming, but the capabilities in Celeste can make this type of exploratory research much easier.
Skai’s solution leverages Amazon Bedrock Agents to create an AI-driven analytics assistant that transforms how users interact with complex advertising data. The system processes natural language queries like ‘Compare ad group performance across low-performing campaigns in Q1,’ eliminating the need for a database specialist. Agent automatically joins Skai’s datasets from profiles, campaigns, ads, products, keywords, and search terms across multiple advertising publishers. Beyond simple data retrieval, the assistant generates comprehensive insights and case studies while providing actionable recommendations on campaign activity, complete with detailed analytical approaches and ready-to-present stakeholder materials.
For example, consider the following question: “I’m launching a new home security product and want to activate 3 new Sponsored Product campaigns and 2 new Sponsored Brand campaigns on Amazon. What high-performing keywords and their match types are already running in other campaigns that would be good to include in these new activations?”
When asked this question with real client data, Celeste answered quickly, finding a combination of branded and generic category terms that the manufacturer might consider for this new product launch. With just a few follow-up questions, Celeste was able to provide estimated CPCs, budgets, and a high-level testing plan for these hypothetical campaigns, complete with negative keywords to reduce unnecessary conflict with their existing campaigns.
This is a great example of an exploratory question that requires summary analysis, identification of trends and insights, and recommendations. Skai data directly supports these kinds of analyses, and the capabilities within Celeste give the agent the intelligence to provide smart recommendations. Amazon Bedrock makes this possible because it gives Celeste access to strong foundation models (FMs) without exposing clients to the risk of having those models’ vendors use sensitive questions for purposes outside of supporting the client directly. Celeste reduces 75% on average the time needed to build client case studies, transforming a process that often took weeks into one requiring only minutes.
Accelerating time-to-value through managed AI using Amazon Bedrock
One critical element of Skai’s success story was our deliberate choice of Amazon Bedrock as the foundational AI service. Unlike alternatives requiring extensive infrastructure setup and model management, Amazon Bedrock provided a frictionless path from concept to production.
The journey began with a simple question: How can we use generative AI to provide our clients a new and improved experience without building AI infrastructure from scratch? With Amazon Bedrock, Skai could experiment within hours and deliver a working proof of concept in days. The team could test multiple FMs (Anthropic’s Claude, Meta’s Llama, and Amazon Nova) without managing separate environments and iterate rapidly through Amazon Bedrock Agents.
One developer noted, “We went from whiteboard to a working prototype in a single sprint. With traditional approaches, we’d still be configuring infrastructure.”
With Amazon Bedrock Agents, Skai could prioritize customer value and rapid iteration over infrastructure complexity. The managed service minimized DevOps overhead for model deployment and scaling while alleviating the need for specialized ML expertise in FM tuning. This helped the team concentrate on data integration and customer-specific analytics patterns, using cost-effective on-demand models at scale while making sure client data remained private and secure.With Amazon Bedrock Agents, domain experts can focus exclusively on what matters most: translating customer data challenges into actionable insights.
Benefits of Amazon Bedrock Agents
The introduction of Amazon Bedrock Agents dramatically simplified Skai’s architecture while reducing the need to build custom code. Built-in action groups replaced thousands of lines of custom integration code that would have required weeks of development time. The platform’s native memory and session management capabilities meant the team could focus on business logic rather than infrastructure concerns. Declarative API definitions reduced integration time from weeks to hours. Additionally, the integrated code interpreter simplified math problem management and facilitated accuracy and scale issues.
As a solution provider serving many customers, security and compliance were non-negotiable. Amazon Bedrock addressed these security requirements by inheriting AWS’s comprehensive compliance certifications including HIPAA, SOC2, and ISO27001. Commitment to not retaining data for model training proved critical for protecting sensitive customer information, while its seamless integration with existing AWS Identity and Access Management (IAM) policies and VPC configurations simplified deployment.
During every client demonstration of Celeste, initial inquiries consistently centered on privacy, security, and the protection of proprietary data. With an AWS infrastructure, Skai confidently assured clients that their data would not be used to train any models, effectively distinguishing Skai from its competitors.With pay-as-you-go model, Skai scaled economically without AI infrastructure investment. The team avoided costly upfront commitments to GPU clusters or specialized instances, instead leveraging automatic scaling based on actual usage patterns. This approach provided granular cost attribution to specific agents, allowing Skai to understand and optimize spending at a detailed level. The flexibility to select the most appropriate model for each specific task further optimized both performance and costs, ensuring resources aligned precisely with business needs.
AWS Enterprise Support as a strategic partner in AI innovation
Working with cutting-edge generative AI agents presents unique challenges that extend far beyond traditional technical support needs. When building Celeste, Skai encountered complex scenarios where solutions didn’t emerge as expected, from managing 200,000-token conversations to optimizing latency in multi-step agent workflows. AWS Enterprise Support proved invaluable as a strategic partner rather than just a support service.
AWS Enterprise Support provided dedicated Technical Account Management (TAM) and Solutions Architect (SA) services that went well beyond reactive problem-solving. Our TAM and SA became an extension of our engineering team, offering the following:
- Regular architectural reviews to optimize our Amazon Bedrock Agents implementation Proactive monitoring recommendations that helped us identify potential bottlenecks before they impacted customer experience Direct access to AWS service teams when we needed deep technical expertise on the advanced features of Amazon Bedrock Agents Strategic guidance and optimization as we scaled from prototype to production
When complex issues arose, such as our initial 90-second (or more) latency challenges or session management complexities, Enterprise Support provided immediate escalation paths and expert consultation.
This comprehensive support framework was instrumental in achieving our aggressive KPIs and time-to-market goals. The combination of proactive guidance, rapid issue resolution, and strategic partnership helped us achieve the following:
- Reduce proof of concept to production timeline by 50% Maintain 99.9% uptime during critical customer demonstrations Scale confidently, knowing we had enterprise-grade support backing our innovation
The value of Enterprise Support provided the confidence and partnership necessary to build our product roadmap on emerging AI technologies, knowing AWS was fully committed to the success of Celeste.
Solution overview
The following diagram illustrates the solution architecture.
Our Amazon Bedrock Agent operates on several core components.
First, a custom layer comprises the following:
- Customer Experience UI (CX UI) – The frontend interface that users interact with to submit questions and view responses Chat Manager – Orchestrates the conversation flow, manages session state, and handles the communication between the UI and the processing layer Chat Executor – Receives processed requests from Chat Manager, interfaces with Amazon Bedrock Agent and handles the business logic for determining when and how to invoke the agent, and manages the overall conversation workflow and short memory
Second, we used the following in conjunction with Amazon Bedrock:
- Amazon Bedrock agent – An orchestrator that receives queries from Chat Executor, determines which tools to invoke based on the query, and manages the tool invocation process. Anthropic’s Claude 3.5 Sonnet V2 – The FM that generates natural language responses. The model generates queries for the API and processes the structured data returned by tools. It creates coherent, contextual answers for users.
Finally, the data layer consists of the following:
- Tool API – A custom API that receives tool invocation requests from the Amazon Bedrock agent and queries the customer data Customer data – The data storage containing sensitive customer information that remains isolated from Amazon Bedrock
The solution also includes the following key security measures:
- Data isolation is enforced between the Tool API and Amazon Bedrock agent Raw customer data is never shared Skai can maintain data privacy and compliance requirements
Overcoming critical challenges
Implementing the solution brought with it a few key challenges.
Firstly, early prototypes suffered from 90-second (or more) response times when chaining multiple agents and APIs. By adopting a custom orchestrator and streaming, we reduced median latency by 30%, as illustrated in the following table.
| Approach | Average Latency (seconds) | P90 | P99 |
| Baseline | 136 | 194 | 215 |
| Optimized Workflow | 44 | 102 | 102 |
Secondly, customers frequently analyzed multi-year datasets, exceeding Anthropic Claude’s 50,000-token window. Our solution uses dynamic session chunking to split conversations while retaining key context, and employs Retrieval Augmented Generation (RAG)-based memory retrieval.
Lastly, we implemented the following measures for error handling at scale:
- Real-time tracing using WatchDog with Amazon CloudWatch Logs Insights to monitor more than 230 agent metrics A retry mechanism, in which failed API calls with 500 error: “BEDROCK_MODEL_INVOCATION_SERVICE_UNAVAILABLE” are automatically retried Amazon CloudWatch monitoring and alerting
Business results
Since deploying with AWS, Skai’s platform has achieved significant results, as shown in the following table.
| Metric | Improvement |
| Report Generation Time | 50% Faster |
| Case Study Generation Time | 75% Faster |
| QBR Composition Time | 80% Faster |
| Report to Recommendation Time | 90% Faster |
While the metrics above demonstrate measurable improvements, the true business impact becomes clear through customer feedback. The core challenges Skai addressed—time-consuming report generation, complex data analysis, and the need for actionable recommendations, have been resolved in ways that fundamentally changed how users work with advertising data.
Customer testimonials
“It’s made my life easier. It’s made my team’s life easier. It’s made my clients’ lives easier and better. So we all work in jobs where there’s millions and millions of data points to scour through every day, and being able to do that as efficiently as possible and as fluidly as possible with Celeste AI is always a welcome addition to Skai.” – Aram Howard, Amazon Advertising Executive, Data Analyst | Channel Bakers
“Celeste is saving hours of time. It’s like having another set of eyes to give suggestions. I’m so stoked to see where this could take us.” – Erick Rudloph, Director of Search Marketing, Xcite Media Group
“It truly feels like having a data scientist right next to me to answer questions, even with recommendations for starting an optimization or looking at an account’s performance.” – Director of Search Marketing at Media Agency
Looking ahead: The future of Celeste
We’re expanding Celeste’s capabilities in the following areas:
- Personalizing the user experience, retaining memories and preferences across multiple sessions. Ingestion of custom data assets, so the client can bring their own data into Celeste and seamlessly connect it to Celeste’s existing data and knowledge. New tools for seamless team integration. These tools will allow Celeste to generate client presentations, build data dashboards, and provide timely notifications.
Conclusion
With Amazon Bedrock Agents, Skai transformed raw data into strategic assets, helping customers make faster, smarter decisions without technical bottlenecks. By combining a robust AWS AI/ML infrastructure with our domain expertise, we’ve created a blueprint other organizations can follow to democratize data analytics.
What truly set our journey apart was the ease with which Amazon Bedrock helped us transition from concept to production. Rather than building complex AI infrastructure, we used a fully managed service that let us focus on our core strengths: understanding customer data challenges and delivering insights at scale. The decision to use Amazon Bedrock resulted in considerable business acceleration, helping us deliver value in weeks rather than quarters while maintaining production grade security, performance, and reliability.
Skai’s journey with Amazon Bedrock continues—follow our series for updates on multi-agent systems and other generative innovations.
About the authors
Lior Heber is the Al Lead Architect at Skai, where he has spent over a decade shaping the company’s technology with a focus on innovation, developer experience, and intelligent Ul design. With a strong background in software architecture and Al-driven solutions, Lior has led transformative projects that push the boundaries of how teams build and deliver products. Beyond his work in tech, he co-founded Colorful Family, a project creating children’s books for diverse families. Lior combines technical expertise with creativity, always looking for ways to bridge technology and human experience.
Yarden Ron is a Software Development Team Lead at Skai, bringing over four years of leadership and engineering experience to the AI-powered commerce media platform. He recently spearheaded the launch of Celeste AI – a GenAI agent designed to revolutionize how marketers engage with their platforms by making insights faster, smarter, and more intuitive. Based in Israel, Yarden blends technical acumen with collaborative drive, leading teams that turn innovative ideas into impactful products.
Tomer Berkovich is a Technical Account Manager at AWS with a specialty focus on Generative AI and Machine Learning. He brings over two decades of technology, engineering, and architecture experience to help organizations navigate their AI/ML journey on AWS. When he isn’t working, he enjoys spending time with his family, exploring emerging technologies, and powerlifting while chasing new personal records.
Dov Amir is a Senior Solutions Architect at AWS, bringing over 20 years of experience in Software, cloud and architecture. In his current role, Dov helps customers accelerate cloud adoption and application modernization by leveraging cloud-native technologies and generative AI.
Gili Nachum is a Principal AI/ML Specialist Solutions Architect who works as part of the EMEA Amazon Machine Learning team. Gili is passionate about the challenges of training deep learning models, and how machine learning is changing the world as we know it. In his spare time, Gili enjoys playing table tennis.
