Something interesting is happening in procurement teams that have adopted AI agents. The technology is working — savings are being captured, tail spend is getting managed, category strategies are being built faster. But the bigger transformation isn't in the technology. It's in how people work.
Procurement professionals who spent their careers building expertise in specific categories are discovering they can now cover three times as many categories with AI support. Generalists who were hired for their communication skills are producing category analyses that rival what specialists took weeks to build. The teams that are thriving aren't the ones running fully autonomous AI — they're the ones that have figured out exactly when and how humans should stay in the loop.
This is human in the loop automation applied to procurement. And getting it right is less about the technology than most people think.
Key takeaways:
- The shift to agentic AI in procurement isn't about automation replacing people — it's about building hybrid teams where humans handle judgment, relationships, and strategy while AI agents handle scale, speed, and synthesis.
- McKinsey identifies human-agent teaming as one of four foundational shifts in the "rewired procurement model," noting that procurement staff will need new capabilities including prompt engineering, scenario evaluation, and change management.
- Suplari, rated 4.8/5 stars on Gartner Peer Insights, is built for this hybrid model — with embedded prompt libraries, contextual AI suggestions, and a conversational interface that helps procurement teams learn to work with AI agents without requiring technical expertise.
- The organizations that invest in adoption and upskilling now — not just technology — will be the ones that capture the compounding value of human-agent teaming over the next three to five years.
What is human in the loop automation?
Human in the loop (HITL) is an approach to automation where humans remain actively involved at critical decision points rather than handing full control to a system. In a human in the loop model, AI handles the heavy lifting — data processing, pattern recognition, initial recommendations — but a human reviews, validates, and approves before action is taken.
This is different from traditional automation, which replaced discrete, mechanical tasks (matching invoices to POs, routing approvals, generating reports). In those workflows, the human still owned the decision-making process — software just made certain steps faster. The workflow stayed the same.
It's also different from fully autonomous AI, where the system acts independently end-to-end without human checkpoints.
Human in the loop automation sits in the middle: the AI reasons, recommends, and drafts — but the human provides judgment, context, and final authority. In procurement, this distinction matters enormously because the stakes of getting a decision wrong (a bad supplier, a missed compliance issue, an overpriced contract) are high and the context that matters most (organizational politics, supplier relationship history, strategic priorities) often isn't captured in data.
Why procurement needs a human in the loop approach
According to Deloitte, only 21% of companies report having a mature governance model for autonomous agents, despite 74% planning deployment within two years. The report warns that “agents are scaling faster than the guardrails.” Meanwhile, the Deloitte study finds AI agents “don’t eliminate the value of humans”—instead giving workers “force multipliers where they can be more effective.”
Procurement has been through several waves of automation — from e-procurement platforms in the early 2000s to robotic process automation (RPA) in the 2010s. Each wave promised to free up professionals for "more strategic work." Each wave delivered efficiency gains. And each wave left the fundamental way people worked largely unchanged.
AI agents are different. They don't just execute steps — they reason, recommend, and act across complex, multi-step processes. An agent can analyze a supplier's performance history, cross-reference it against market benchmarks, identify pricing drift, draft a renegotiation strategy, and prepare the communication — all before a procurement professional has finished their morning coffee.
This creates a different dynamic. The human is no longer the sole thinker who delegates mechanical tasks to software. Instead, the human and the agent are collaborating — with different strengths, different roles, and a genuine need for coordination.
McKinsey's research on the "rewired procurement model" frames this clearly:
"In a rewired procurement function, humans and AI agents will work side by side. Procurement staff will guide and coach their digital counterparts, while agents will take over most repetitive transactional work, freeing up people to focus on strategic decision-making, orchestration, and oversight."
The key phrase is "guide and coach." This isn't a manager-to-tool relationship. It's closer to a senior professional working with a very capable but junior colleague — one that needs direction, benefits from feedback, and improves with every interaction. That's a human in the loop model.
The three tiers: when humans step in and when they don't
Early adopters of human in the loop automation in procurement are finding the work naturally breaks into three tiers, each with a distinct balance of human judgment and AI capability.
Tier 1: Agent-led, human-approved
The AI agent does the heavy lifting. It scans continuously, flags issues, and proposes actions. The human reviews a queue of recommendations and makes go/no-go decisions.
What the agent does: Monitors data, identifies anomalies, drafts actions.
What the human does: Reviews, approves, handles exceptions.
Where this applies: Tail spend monitoring, spend categorization, data quality maintenance, contract compliance checking.
This tier delivers the highest time savings — up to a 90% reduction in analyst time on routine analysis. The human in the loop here acts as a quality gate, catching the edge cases where the AI's recommendation doesn't account for context it can't see.
Tier 2: Human-led, agent-assisted
The professional sets direction, defines objectives, and manages relationships. The agent provides intelligence, benchmarks, scenario modeling, and first drafts.
What the human does: Sets strategy, manages stakeholders, refines output.
What the agent does: Market intel, benchmarks, scenario analysis, first drafts.
Where this applies: Category strategy development, negotiation preparation, supplier risk assessment, scenario modeling.
This is where the human in the loop approach creates the most leverage. Professionals can manage three times more categories than before — not because they're working harder, but because the agent handles the research and analysis while the human focuses on judgment and direction.
Tier 3: Human-only, better informed
Work that requires relationship depth, political awareness, and ethical reasoning stays fully human. But the human now has more time and better data because the AI handles Tiers 1 and 2.
What the human does: Full ownership of relationships, judgment, and organizational alignment.
How AI helps: By freeing up bandwidth. When agents handle routine monitoring and analytical heavy lifting, humans can spend twice as much time on the strategic work that actually moves the needle.
Where this applies: Strategic supplier relationships, executive negotiation, cross-functional stakeholder alignment, ethical judgment calls.
The key insight across all three tiers: human in the loop automation in procurement isn't about humans double-checking every AI output. It's about designing the right level of human involvement for each type of work — from light oversight at Tier 1 to full ownership at Tier 3.
Human in the loop vs. human on the loop: what's the difference?
These two terms are often confused, but they describe meaningfully different levels of human involvement.
Human in the loop means a human is directly involved in the decision process. The AI proposes, the human decides. Nothing happens without human approval. This is Tier 1 and Tier 2 in the framework above — the human is an active participant in the workflow.
Human on the loop means the AI acts autonomously, but a human monitors the system and can intervene if something goes wrong. The human isn't reviewing every decision — they're watching dashboards, checking outputs periodically, and stepping in only when the system flags an exception or they spot a problem.
In procurement, the distinction matters for risk management:
- Human in the loop is appropriate for high-value decisions: strategic sourcing, contract negotiations, supplier selection, anything involving significant spend or risk.
- Human on the loop may be sufficient for lower-risk, high-volume tasks: routine spend classification, standard invoice matching, automated reorder triggers for well-established categories.
Most procurement teams implementing AI agents today are operating primarily in a human in the loop model — and for good reason. The consequences of an unchecked bad decision (a compliance violation, a damaged supplier relationship, an unfavorable contract term) are too high to remove human judgment from the loop entirely. As trust in AI systems builds and the models improve, some workflows will naturally migrate from human in the loop to human on the loop. But that transition should be deliberate, not accidental.
What the human in the loop actually does in procurement
If the work is changing, the skills procurement professionals bring to the loop need to change too. This doesn't mean existing expertise becomes worthless — far from it. But the weighting shifts.
Judgment and contextual reasoning becomes the primary differentiator. The ability to evaluate an AI agent's recommendation against business context the agent may not have — organizational politics, supplier relationship history, strategic priorities not captured in data. When the analytical work is handled by AI, human judgment is what separates good decisions from bad ones.
AI direction and prompt craft emerges as a new core skill. This isn't prompt engineering in the technical sense — it's more like knowing how to brief a capable analyst. What context do they need? What are the constraints? What does a good answer look like? Jeff Gerber, CEO and co-founder of Suplari, describes the learning curve as surprisingly short:
"Usually, it's like one — 'Well, how about this?' — and then that person is off on a tear. They have all the knowledge. They just need to be able to get started."
Cross-functional communication becomes a core competency, not a nice-to-have. As AI frees up bandwidth, procurement professionals are increasingly expected to engage with finance, operations, sales, and the C-suite. The ability to translate procurement intelligence into business language — and to sell procurement's value proposition internally — is what the freed-up time should be spent on.
Change management is the skill most teams underinvest in. Someone on the team needs to drive adoption, manage resistance, build the prompt libraries, and continuously refine how humans and agents work together. McKinsey's research supports this, noting that the rewired procurement model requires "reskilling for strategy, exception management, and insight interpretation."
Meanwhile, some traditionally valued skills matter less (not zero, but less). Manual data wrangling — the hours spent cleaning spreadsheets and building pivot tables — is absorbed by AI. Deep single-category specialization is still valuable, but as Gerber explains, you no longer need 15 years in a category to be effective:
"It's actually very quick for people to get up to speed in a category because they can work with an agent that has all that knowledge that maybe you lost or you didn't have."
The blank cursor problem: why adoption stalls
Here's the uncomfortable truth about human in the loop automation: the technology is ahead of the adoption.
McKinsey identifies "embedding change, not just tools" as a critical success factor, noting that "successful transformations pair technology with operating model redesign, new KPIs, and strong change leadership." Yet most procurement organizations are still focused on selecting and implementing AI tools rather than redesigning how their teams work with them.
The biggest barrier isn't technical capability. It's what Gerber calls the blank cursor problem:
"Especially if you go to folks that are maybe used to doing category management or sourcing in a particular way and they're presented with just a co-pilot, you kind of get the 'I don't even know what to ask it' response."
This is a human in the loop design problem, not a technology problem. If your humans don't know how to participate in the loop effectively, the loop breaks down.
Suplari's approach is instructive. Rather than dropping users into an open-ended chat interface, they embed contextual prompts throughout the product — when you're looking at a supplier, the system suggests relevant questions and actions specific to that supplier. When you're reviewing a category, it surfaces opportunities the AI has identified and offers starting points for deeper exploration.
"Just by navigating in our user interface, we're showing you suggestions and interesting ways that you might want to interact with AI. Those starting points — that's really the thing you need."
They also built a shared prompt library — a repository where team members save effective conversations, queries, and workflows. When someone leaves the team, their best prompts stay behind. When a new person joins, they inherit not just the data but the questions their predecessor found most valuable.
"Someone can come in to a new role and as long as their predecessor had been using it effectively, they can pick it up and they're off to the races already with what to ask it."
This solves the knowledge transfer problem that has plagued procurement for decades — and it makes the human in the loop model sustainable across team changes, not dependent on any individual.
How to measure a human in the loop procurement model
Traditional procurement KPIs — cost savings, spend under management, contract compliance — still matter. But they don't capture the value of a well-designed human in the loop model.
A team that covers twice as many categories, responds to market shifts in days rather than weeks, and builds strategies incorporating scenarios they'd never have had time to model — that team creates value traditional metrics miss.
McKinsey proposes "Procurement ROI" as a unifying metric: total value created divided by total cost to achieve it. Value includes realized savings, leakage avoided, working capital benefits, risk reduction, and revenue enablement.
For human in the loop teams specifically, consider tracking:
Category coverage per analyst — how many categories can each person effectively manage with AI support? This measures whether the loop is actually creating leverage.
Time-to-insight — how quickly can the team go from question to actionable recommendation? A well-functioning human in the loop model should compress this dramatically.
Savings identification to realization ratio — are the opportunities AI surfaces actually being captured? This tells you whether the human in the loop is effective at converting AI recommendations into real outcomes.
Adoption depth — not just logins, but meaningful interactions. Prompts saved, recommendations acted on, strategies built with agent assistance. This measures whether the humans are truly participating in the loop or just rubber-stamping outputs.
How to build your human in the loop procurement workflow
For procurement leaders ready to move from piloting AI tools to a genuine human in the loop operating model, here's what the organizations doing it well have in common.
Start with the work, not the tool. Map your team's actual time allocation. Where do people spend hours on analysis an agent could do in minutes? Where are the judgment-heavy tasks that should stay human? This audit tells you where to place the human in the loop — and where to let the AI run with lighter oversight.
Pick your first HITL workflow. Choose one workflow — tail spend monitoring, category strategy development, contract compliance — and redesign it with explicit human checkpoints. Define who does what, what the approval gates look like, and how you'll measure success.
Invest in prompt literacy. Your team doesn't need to write Python or understand transformer architectures. They need to know how to frame a good question, evaluate an AI's response critically, and iterate when the first answer isn't right. Budget for this training — it pays back faster than almost any other investment.
Build the prompt library from day one. Every effective interaction with an AI agent is an asset. Capture it. Share it. Make it discoverable. This is how institutional knowledge gets built in a human in the loop world.
Redesign roles gradually. Don't restructure your org chart on day one. Let the model reveal what's needed. As people free up time from analytical work, direct them toward the strategic and relational work you've never had bandwidth for.
Create feedback loops. AI agents improve with use and feedback. Build processes for your team to flag when the agent gets it wrong, celebrate when it gets it right, and continuously refine the collaboration. McKinsey emphasizes: "AI systems improve with use. Treat every cycle as a learning opportunity."
The bottom line
Human in the loop automation in procurement isn't a compromise between full automation and doing everything manually. It's a deliberate design choice that recognizes where AI excels (speed, scale, synthesis) and where humans are irreplaceable (judgment, relationships, context). The organizations that get this balance right — that invest in training their people to be effective in the loop, not just in buying better tools — are the ones capturing compounding value.
The AI agents are ready. The question is whether your team knows when to step in, when to step back, and how to make the loop work. To assess how ready your organization is for AI-powered procurement, take Suplari's AI readiness assessment.
This article draws on insights from the Procurement Software Podcast, hosted by James Meads, featuring Jeff Gerber, CEO and Co-founder of Suplari. Listen to the full conversation on Youtube.
