In a recent webinar, HP revealed that its indirect procurement team now contributes, on average, eight cents in earnings per share through real, measurable financial value that directly impacts the bottom line. A contributing factor is the team’s use of agentic AI, which has delivered savings of 10–15% on competitive tail spend purchases. No longer on the horizon, agentic AI in sourcing is providing real value to the world’s biggest companies right now.
Agentic AI’s prime-time readiness comes just as procurement leaders are contending with relentless chaos in the market caused by volatility and uncertainty. Independent research by HFS revealed that AI-driven sourcing already delivers 20% cost savings, making it one of the lowest-risk, highest-reward AI applications for enterprises.
That’s why companies like HP are using it to do things like automate price comparisons, generate RFPs and support supplier negotiations, freeing up the humans for high-value strategic and nuanced decision-making.
Agentic’s early days
Though the buzz around agentic AI is relatively recent, the technology’s origins reach back many decades. Shakey the Robot, a rudimentary agentic system developed in the 1960s, was born out of mid-1950s AI research. In 1973, The University of Edinburgh developed the “Freddy II” robotic system for recognising and assembling objects, which had a number of basic agentic features.
These early systems introduced some core ideas, like rule-based reasoning and symbolic logic, that remain central to AI research and application today. In more recent years autonomous computing evolved to power systems like the Mars Perseverance Rover and AlphaGo.
Today, generative agents, built on top of large and small, more specialised language models are able to learn and execute complex business tasks. In procurement, some examples of these include tail spend optimisation, RFP drafting and risk assessment.
What makes AI “Agentic”?
Amidst all the noise in the tech industry, may struggle to distinguish among automation, AI, and agentic AI. And in certain areas, the lines are indeed blurry.
True agentic systems, however, are defined by certain characteristics: they are goal-oriented, autonomous, adaptive, and interactive. In practice, that means they operate with a clear purpose, make independent decisions within defined constraints, adjust to evolving environments, and communicate effectively with users and other systems.
A supplier recommendation engine, for example, might suggest a list of office chair suppliers ranked based on several factors. These could include historical purchase data, category, location, pricing or quality ratings. This simple AI system uses rule-based and collaborative filtering, and machine learning-based ranking.
A more intelligent AI system allows the user to ask more complex questions. For example, they could ask: “source a list of suppliers of office chairs that uses sustainable materials, with a low CO2 footprint, under $40 per unit.” This context-aware system is able to take multiple variables and objectives (cost and sustainability) into account. To do this, it uses machine learning, real-time data feeds and embedded logic.
The next level up, when AI starts acting as a strategic sourcing advisor, is where true ‘agency’ kicks in. Here the system starts to pursue goals, model scenarios, and reason beyond the prompts it’s given. The user could ask it to, for example, “help reduce sourcing risk for Q4”. To produce the response, it might then do things like analyse historical risk exposure, model supplier diversification, and suggest tradeoffs (e.g., cost vs. resilience).
“True agentic systems, however, are defined by certain characteristics: they are goal-oriented, autonomous, adaptive, and interactive.”
The more agentic an AI system is, the more deeply it understands its environment. This means perceiving external conditions, processing contextual data, and dynamically determining the best courses of action in real time. Crucially, agentic systems actively align with organisational objectives, rather than reacting to prompts.
Memory is also critical. An AI agent must retain past interactions and experiences to refine its decision-making over time. By leveraging memory, it can continuously improve its performance, adapt to changing circumstances, and enhance its ability to serve users effectively.
Beyond reasoning and decision-making, agentic AI interacts with the world through various input and output mechanisms. This can include processing chat-based queries, analyzing sensor data, responding to time-based triggers, or accessing external services via APIs.
Negotiating high-stakes procurement tasks
Procurement tasks like scoping, supplier discovery, benchmarking, and proposal analysis are ideal use cases for agentic AI. These tasks are complex, reliant on data, and susceptible to human error.
One particularly high-stakes task is negotiation, which involves balancing price, market trends, scope, timelines, talent, and methodology. While some people excel at this, others find it dauntingly complex or confrontational. An AI agent helps ease the burden by evaluating proposals holistically and objectively. It can then generate tailored negotiation strategies and recommend concrete actions based on all relevant data.
Evolving further to support business
The next major evolution, the fully autonomous sourcing agent, is just ahead. This advanced system will operate with minimal human oversight. It will adapt in real time to shifting market conditions, make independent decisions, and clearly explain its reasoning.
What makes this next stage so compelling is its ability to support a fundamental business truth: gaining market advantage depends on making the right tradeoffs at the right time. Traditional business systems struggled to deliver timely insights because they relied on IT teams to first structure complex internal and external data.
Agentic AI can process vast amounts of raw information directly, identifying patterns, generating insights, and enabling leaders to make fast, high-quality decisions. That’s one reason HP’s indirect procurement team can so clearly link its EPS contribution to the company’s bottom line. And they’re only just getting started.