Maverick Spend Is a Systems Problem Agentic AI Is a Systems Fix

Procurement has had AI for years. Spend classification dashboards, anomaly alerts, historical analytics — none of these is new. And yet maverick spend — purchasing that bypasses approved contracts, preferred vendors, or established procurement workflows — remains one of the most persistent and underreported cost leaks in enterprise operations.

The Hackett Group’s research on maverick spend found that organizations can lose up to 16% of negotiated savings to off-contract purchasing. For a mid-market enterprise with $50 million in annual procurement spend, that translates to roughly $8 million in unoptimized expenditure — much of it invisible until a quarterly review surfaces the damage. (Source: The Hackett Group, User Experience and Maverick Spend Study, 2019)

That figure has not moved materially in half a decade, despite widespread investment in procurement technology. The reason is structural: most of the AI deployed in procurement to date watches and reports. It does not act.

This is where the distinction between conventional AI tooling and agentic AI becomes operationally significant — and why procurement leaders evaluating technology investments need to understand the difference.

What “Agentic AI” Actually Means — and What It Is Not

The term “agentic AI” has entered the procurement vocabulary rapidly, and not always precisely. Gartner has identified agentic reasoning, multimodality, and AI agents as the three GenAI advancements that will reshape how procurement operates. They define AI agents in this context as autonomous systems that can perform tasks and make decisions on behalf of human operators — freeing procurement professionals to focus on strategic initiatives, complex problem-solving, and edge cases. (Source: Gartner, “Three Key Advancements in Generative AI That Will Shape the Future of Procurement,” November 2024)

This is not the same as a chatbot that answers questions, a dashboard that flags anomalies, or a rules engine that routes approvals. Those are tools that assist. Agentic AI systems execute — they reason through multi-step tasks, act across systems, and complete procurement workflows with minimal human intervention at each step.

The distinction matters for maverick spend because the problem is not a lack of information. Procurement teams generally know where off-contract purchasing happens. The problem is that by the time they know, the transaction is done. Agentic AI changes the sequence: the system intervenes at the point of transaction, not after payment.

Gartner forecasts that by 2028, 90% of B2B purchases will be intermediated by AI agents, channeling more than $15 trillion in spending through automated exchanges — a projection that signals agentic AI is not a peripheral upgrade to procurement, but a fundamental reshaping of how enterprise buying happens

Why Maverick Spend Persists — Especially in Mid-Market Organizations

Before examining what agentic AI changes, it is worth understanding why the problem is structurally resistant to conventional solutions.

Mid-market procurement teams — typically organizations between 500 and 2,000 employees — operate with lean staffing. Two to five procurement professionals managing hundreds of active vendors and thousands of annual transactions is common. In this environment, process exceptions are not the exception; they are the operating norm.

The Hackett Group’s research reports that 75% of procurement professionals cite a lack of self-service or guided buying tools as a leading cause of maverick spend. When the official procurement channel is harder to use than going around it — slow approvals, outdated catalogues, poor search interfaces — employees reliably default to ad hoc purchasing. (Source: The Hackett Group, 2019)

The visibility problem compounds this. Most mid-market ERP systems capture what was purchased, but do not flag in real time whether that purchase was on-contract, off-contract, or from an unapproved supplier. By the time a compliance gap surfaces in a spend report, the transaction has already settled.

More recent data reinforces the scale of the challenge. The Hackett Group’s 2025 research on Digital World-Class procurement organizations found that top performers experience 59% less savings loss due to maverick buying compared to their peer group — a gap achieved through better tools, policy enforcement, and organizational structure. The implication is clear: the gap between leaders and laggards in maverick spend control is widening, not narrowing. (Source: The Hackett Group, Digital World-Class Procurement Metrics & Insights, 2024)

Meanwhile, Ardent Partners’ CPO Rising 2025 report — based on surveys of 326 procurement leaders — found that 50% of procurement teams already use AI, with that figure expected to reach 80% by year-end. Yet adoption remains materially uneven across organization sizes, with mid-market firms consistently trailing larger counterparts. (Source: Ardent Partners, Procurement Metrics That Matter in 2025)

Tail Spend: Where Maverick Behavior Concentrates

Any serious discussion of maverick spend that omits tail spend is incomplete.

Tail spend — the 80% of suppliers that typically represent roughly 20% of total procurement expenditure — is where maverick purchasing concentrates most heavily. These are the low-value, high-frequency transactions that no one has the bandwidth to manage: office supplies ordered from a personal Amazon account, IT software licenses purchased on a department credit card, one-off professional services engaged without a sourcing event.

Individually, each transaction is small. Collectively, tail spend represents a significant share of total procurement volume and an outsized share of compliance failures. The Hackett Group’s 2025 Tail Spend Management Study found that only 4% of companies actively manage most of their tail spend, while 64% of procurement leaders expressed dissatisfaction with their current approach. (Source: The Hackett Group, 2025 Tail Spend Management Study)

This is where agentic AI makes its strongest operational case. Human teams cannot economically police thousands of low-value transactions. Autonomous agents can classify spend in real time, match requests against contracted suppliers, and route exceptions for approval before a purchase order is generated. The economics of agentic AI favor precisely the transaction volume and frequency profile that defines tail spend.

Platforms like Zycus have built dedicated agentic AI capabilities for tail spend management, deploying autonomous negotiation agents that receive tail spend requests, apply supplier rules and pricing thresholds, and finalize orders without manual intervention — while maintaining full auditability.

Intake as the Control Point

If tail spend is where maverick behavior concentrates, intake is where it either starts or gets stopped.

Intake — the process by which an employee’s purchasing need enters the procurement system — is the single highest-leverage intervention point for maverick spend reduction. If the first interaction an employee has with procurement is an intelligent experience that understands their request, routes them to a contracted supplier, and generates a compliant purchase order, the friction that causes maverick behavior is eliminated at the source.

This is not a new insight, but it has historically been difficult to execute. Traditional intake processes required employees to navigate forms, identify the correct category, find an approved supplier, and wait for manual routing and approval. Each step introduced friction; each point of friction increased the probability of an employee going around the system entirely.

Agentic AI fundamentally changes intake by making the compliant path the path of least resistance. Conversational interfaces — embedded where employees already work, such as Microsoft Teams — capture demand in natural language, auto-classify the request, enforce policy guardrails, and route to the appropriate channel. No forms. No category lookup. No wait.

The use cases for AI-powered intake extend beyond simple request capture: intelligent triage determines whether a request should route to a preferred catalogue, trigger a sourcing event, or escalate for approval — decisions that previously required procurement team intervention for every transaction.

The Agentic Lifecycle: From Intake to Outcomes

The structural advantage of agentic AI procurement platforms is that they operate across the transaction lifecycle, not at a single point. Understanding this full sequence is essential for evaluating how these systems address maverick spend.

Intake captures demand and prevents leakage at the source. An employee requests what they need. The system understands the request, checks it against policy, and routes it — all before a purchase order exists. Maverick spend that never enters the system is maverick spend that never happens.

Real-time spend classification replaces retroactive categorization. Gartner’s Predicts 2025 report on procurement noted that 49% of procurement leaders cite data accuracy and reliability as significant challenges, and projects that by 2027, 85% of procurement organizations will still be working to improve data quality to exploit AI efficiencies. Agentic systems address this by applying consistent classification logic — such as UNSPSC or custom category structures — at the point of transaction, not months later in a quarterly cleanse. (Source: Gartner, “Predicts 2025: Procurement Addresses Data Challenges and Embraces Rapid Change,” January 2025)

Contract compliance is enforced autonomously at PO creation. Rather than discovering contract deviations in an audit, agentic platforms cross-reference purchase requests against the active contract repository in real time. Pricing deviations, unapproved vendors, and terms mismatches are surfaced — and blocked or escalated — before the transaction proceeds.

Outcome measurement closes the loop. Savings are not just negotiated; they are realized and tracked. The distance between a sourcing event and actual spend-through-contract is measured continuously, giving procurement leaders a savings realization view rather than a savings projection.

This intake-to-outcomes approach — where autonomous agents handle classification, contract matching, compliance enforcement, and savings realization across a unified platform — represents the architectural shift from retrospective reporting to autonomous real-time intervention.

The Mid-Market Adoption Reality

Despite the operational case, mid-market adoption of AI procurement platforms has lagged. The barriers cited most frequently are implementation complexity, ERP integration concerns, and total cost of ownership uncertainty.

However, the market has responded. Forrester’s Supplier Value Management Platforms Wave, Q3 2024 — which evaluated nine leading source-to-pay providers across 33 criteria — noted that leading platforms are offering customizable configurations suitable to a wide range of enterprise maturity levels, signaling a deliberate move toward mid-market accessibility. (Source: Forrester, Q3 2024)

Several procurement vendors have introduced mid-market deployment tracks: pre-configured for common ERP environments such as SAP, Oracle NetSuite, and Microsoft Dynamics, with implementation timelines measured in weeks rather than quarters. Cloud-native deployment models have replaced the upfront capital commitments that previously restricted advanced procurement platforms to large enterprises.

For mid-market organizations specifically, the modular deployment approach — starting with a single high-impact use case like intake management or tail spend automation and expanding into full source-to-pay automation — offers a practical entry path that does not require a wholesale platform replacement.

Where Agentic AI Platforms Fall Short

A balanced assessment requires acknowledging limitations.

Agentic AI platforms are only as effective as the data they operate on. Organizations with fragmented ERP environments, inconsistent historical spend data, or poorly maintained contract repositories will see diminished returns from autonomous classification and compliance enforcement. Gartner’s research projects that by 2027, only 30% of organizations will have sufficient data quality to fully exploit advanced AI capabilities — a sobering figure for organizations rushing to deploy. (Source: Gartner, Predicts 2025: Revisit ERP Strategies)

Change management is a frequently underestimated barrier. Intelligent intake only reduces maverick spend if employees use it, and adoption rates in the first 6–12 months post-deployment can be low without active internal promotion. The technology can be flawless; if the organizational change effort is absent, results will disappoint.

Gartner has also cautioned that over 40% of agentic AI projects may be canceled by end of 2027 due to escalating costs, unclear business value, or inadequate risk controls. The recommendation is clear: pursue agentic AI where it delivers specific, measurable value — not as a broad technology initiative. (Source: Gartner, June 2025)

Mid-market organizations should build realistic data readiness assessments and change management budgets into implementation plans from the outset.

What Procurement Leaders Should Evaluate

For CPOs and procurement heads assessing agentic AI platforms, the evaluation criteria extend beyond feature checklists:

Real-time intervention vs. retrospective reporting. Does the platform act at the point of transaction, or does it generate dashboards after payment? This is the fundamental distinction between conventional AI and agentic AI in procurement.

Intake depth. How naturally does the compliant purchasing path integrate into employee workflows? Is it embedded where employees already work — in Teams, Slack, email — or does it require navigating a separate procurement portal?

Tail spend capability. Can the platform autonomously handle high-volume, low-value transactions — including supplier matching, negotiation, and PO generation — without manual intervention for each transaction?

Contract repository intelligence. Does the system actively monitor contract utilization, flag underperforming agreements, and enforce pricing compliance at the point of purchase?

ERP integration realism. What is the realistic integration timeline with your existing financial system of record? Weeks or months?

Time-to-value. Can spend classification, intake automation, and compliance enforcement be activated without requiring full platform deployment?

The platforms earning the strongest mid-market traction are those that combine genuine agentic capability — autonomous execution, not just AI-assisted dashboards — with deployment architectures designed for organizations that do not have dedicated procurement IT teams.

Conclusion

Maverick spend is not a behavioral problem that training and awareness campaigns can solve in isolation. It is a systems problem — rooted in workflow friction, poor real-time visibility, and the absence of autonomous controls at the point of transaction.

Traditional AI in procurement addressed the visibility dimension: better dashboards, more accurate spend reports, smarter anomaly flags. What it did not address was the execution gap — the distance between identifying a problem and preventing it from occurring. Agentic AI closes that gap by making compliant purchasing the path of least resistance, intercepting deviations before they become sunk costs, and autonomously managing the long tail of transactions that no human team can economically police.

For mid-market enterprises with the operational scale to absorb maverick losses but not the headcount to manually prevent them, the case for agentic AI-powered procurement is increasingly difficult to set aside.