AI is moving quickly. No longer confined to experiments or proofs of concept, autonomous systems – often referred to as agentic AI – are beginning to take hold in real business workflows. These systems go beyond analyzing data; they reason, act, and provide actionable insights at scale.
When Unilever launched Horizon3, its global AI lab in Toronto, in 2023, the alignment with Vector Institute’s mission was immediately apparent. Embedded within Unilever’s global Operations Data and Analytics team, Horizon3 operates through a hybrid model that brings together Unilever practitioners, academic researchers, start-ups, and industry leaders to collaborate on applied AI research—precisely the kind of ecosystem-driven approach Vector is designed to support.
A recent collaboration between Vector Institute and Horizon3 provides a practical view of what rapid experimentation with agentic AI looks like in enterprise settings. During Vector’s Agentic AI Bootcamp, an intensive, hands-on experience focused on developing autonomous agentic systems, Unilever explored how autonomous agents could streamline logistics analysis, reduce manual effort, and enable faster, data-informed strategic decisions.
Reimagining logistics intelligence
Supply chain and procurement workflows are notoriously complex. Analysts routinely spend hours navigating dashboards, applying filters, and compiling reports before delivering actionable insights. For global operations, this time cost quickly becomes significant.
Unilever approached this challenge with a Market Insights agent – an AI system designed to turn manual, labor-intensive workflows into a conversational, natural-language interface. Users can ask questions like, “How do our freight costs compare to benchmarks for North American shipments this quarter?” and receive structured answers in seconds, whether as tables, charts, or Excel files.
“Agentic AI has the potential to support our initiatives today while opening new ways to structure work tomorrow. The bootcamp allowed us to explore these possibilities in a practical, hands-on way.”
This approach is part of a broader trend to make analytical insights more accessible and actionable without replacing human judgment. By turning complex, time-consuming tasks into instant, interactive insights, Unilever buyers are able to sharpen negotiations and achieve significant productivity gains.
Striking a balance: Flexibility vs. reliability
Working closely with Unilever teams, Vector guided participants through tool integration and multi-agent architecture design; performance evaluation benchmarks; and modular system design for scalability and reliability. During the bootcamp, a recurring theme emerged: the trade-off between autonomy and reliability. Early tests showed that allowing the agent to generate complex SQL queries independently led to errors. The team responded by implementing predefined, validated SQL templates – sacrificing some flexibility in exchange for accuracy and trustworthiness.
“Rather than just a library or just a tool, it was a concept of design. Modularity increases the reliability.”
This mirrors a wider industry shift: in enterprise AI adoption, trust and repeatability often matter more than maximum flexibility, particularly in operational workflows. By carefully balancing autonomy with safeguards, organizations can unlock agentic AI’s potential while maintaining governance and consistency.
Democratizing logistics market analysis
Enterprise logistics analysis is often slow, labour-intensive, and inaccessible to anyone without deep domain expertise. The collaboration identified an opportunity to change this fundamentally.
The Market Insights agent enables buyers and analysts to:
- Compare actual logistics costs to benchmarks and “should-cost” models in real time
- Generate instant tables, charts, and Excel reports from natural-language queries
- Access insights without needing to understand underlying data structures or KPI derivations
By replacing dashboard navigation and repetitive reporting with conversational queries, the agent drastically reduces manual effort and improves accessibility. This is in line with a broader industry trend where companies like McKinsey andOracle, and startups like TinyFish are deploying autonomous agents to generate insights, optimize operations, and monitor competitive intelligence in real time. Unilever’s solution delivers actionable intelligence in minutes rather than days.
“We can talk about impact in two ways: first, the immediate value agentic AI brings to our current initiatives, and second, the long-term potential it unlocks for transforming how we operate.”
Leveraging custom tools to empower agents
Agentic AI becomes exponentially more powerful when equipped with specialized tools. Vector collaborated with Unilever to create a toolkit enabling the agent to:
- Retrieve KPI data from large datasets via natural language using a get_kpi_data function
- Execute complex queries using predefined KPI templates for reliability
- Produce structured outputs in multiple formats (tables, charts, Excel files)
- Conduct Pareto analysis and financial impact simulations
- Clean and standardize text inputs for precise query execution
This approach mitigates errors while maintaining flexibility, enabling the agent to function reliably across diverse logistics scenarios.
The Market Insights agent is built on a custom Agent Development Kit (ADK) designed for rapid, consistent agent development. Key aspects include:
- Integration with libraries such as litellm, LangChain, and LangGraph
- Support for single- and multi-agent systems for scalability
- Modular design: tasks are broken into smaller units linked to SQL templates
- Powered by Gemini 2.5 Pro, improving comprehension and generation
Through guided experimentation, the team discovered that breaking monolithic instructions into smaller, modular tasks significantly improved accuracy and completeness. A structured “chain-of-thought” process in the system prompts guides reasoning while ensuring reliability.
“Watching Unilever’s team transform a complex, manual procurement workflow into an intelligent, conversational system in just eight weeks demonstrates the tremendous potential when industry leaders commit to hands-on experimentation with agentic AI.”
Testing, evaluation, and data security
Agentic AI systems are dynamic and non-deterministic, making robust evaluation essential. Vector guided Unilever through a combination of automated LLM evaluation and manual human review, measuring:
- Correctness: Are outputs accurate relative to source data?
- Completeness: Does the agent return all requested KPIs and analysis?
- Professionalism: Are text outputs clear, concise, and business-ready?
- Figure feneration: Are tables and charts accurate and relevant?
Data privacy was addressed using fully synthetic, anonymized datasets that replicated real logistics data structures, enabling safe experimentation without exposing proprietary information.
Partnership outcomes
By equipping Unilever’s team with specialized toolkit knowledge and Vector’s pre-defined SQL template methodology, the collaboration successfully demonstrated how to build highly reliable agentic solutions. Unilever’s Market Insights agent transforms time-consuming, manual workflows into instant, conversational experiences, reducing analytical effort from hours to minutes while providing accurate, on-demand strategic insights.
Key outcomes include:
- A practical blueprint for other enterprises adopting agentic AI
- Immediate operational efficiency gains
- Enhanced decision-making through accessible, structured insights
- A validated methodology for balancing flexibility and reliability
Moving forward, Unilever is planning to pursue a user-driven improvement approach, helping establish seamless feedback loops with their business users to guide systematic expansion of SQL template libraries. The teams are also exploring enhanced visualization capabilities, moving beyond static charts toward dynamic, interactive dashboard outputs for richer user experiences.
This ongoing collaboration demonstrates how sustained partnerships between research institutes and enterprises can accelerate AI adoption while producing real-world impact.
“This case study exemplifies our commitment to bridging cutting-edge AI research with enterprise outcomes. By collaborating with Unilever, we are helping Canadian businesses harness agentic AI while contributing insights to the global AI community.”
Conclusion
The Vector-Unilever collaboration offers a window into the future of enterprise AI: autonomous, reliable, and practical systems that transform workflows without replacing human judgment. By combining experimentation, structured methodology, and human expertise, agentic AI is no longer an abstract concept—it is actively reshaping how work gets done, how decisions are made, and how organizations unlock value at scale.
Turn AI possibilities into competitive reality
Explore how Vector Institute can support your organization’s AI adoption journey.
