.webp)
The supplier a procurement team uses for a category today is often the one they found two or three years ago. Market scans run annually at best, and for tail spend they barely run at all. The result is that most organisations are negotiating from a supplier pool that the market has already moved past, with no practical way to close that gap at scale using manual research methods.
Agentic AI changes the economics of that search. Agents run continuously, analyse data at a scale no individual researcher can match, and deliver a qualified, risk-assessed supplier pool at the start of each sourcing event rather than as the output of weeks of research, compressing weeks of research into hours.
This article explains how agentic AI works in a supplier discovery context, what it takes to implement it reliably, and what organisations are actually seeing from it in practice.
Agentic AI refers to AI systems that act autonomously over multiple steps to complete a goal, rather than simply responding to a prompt. In a procurement context, agentic AI does not wait for a buyer to trigger a search. It monitors markets, evaluates suppliers against defined criteria, flags new candidates, and escalates findings to the procurement team when action is needed.
In this context, the AI agent functions as a continuous research operator rather than a search tool. It identifies new suppliers from trade registries, market databases, ESG ratings, certification authorities, and news sources. It evaluates them against cost benchmarks, quality requirements, compliance standards, and sustainability criteria, on an ongoing basis and not only when a sourcing event is triggered.
Agentic AI represents a distinct step forward from earlier procurement technology. These systems operate autonomously across multi-step research and evaluation tasks rather than waiting for a buyer to initiate each action. Procurement's shift from event-driven supplier searches to continuous automated monitoring is what this enables: when a sourcing need arises, the shortlist already exists.
Earlier forms of procurement automation, including robotic process automation (RPA) and rule-based systems, followed fixed scripts. RPA tools handle well-defined, repeatable tasks: invoice processing, data extraction from supplier portals, routing approvals through defined workflow steps. What they cannot do is synthesise conflicting data, reason across multiple sources, or identify a supplier that is not already in the company's database.
Agentic AI systems are built on generative AI foundations, large language models combined with machine learning, that allow them to reason across data sources, adapt to new information, and execute multi-step processes. An AI agent can identify specialist suppliers matching a rare technical specification, verify their certifications against the issuing authority's database, and surface those suppliers in a buyer's workflow, without each individual step needing to be scripted in advance.
Earlier generations of AI automation could surface insights but not act on them. They could tell buyers which suppliers were underperforming; they could not initiate a search for alternatives. Agentic AI is the first generation capable of completing a multi-step task from search to shortlist without requiring human intervention at each step.
Three qualities define agentic systems: multi-step reasoning, operation within defined governance guardrails, and the ability to synthesise findings into structured recommendations. The agent works through a chain of actions across searching, evaluating, verifying, and comparing suppliers, and produces a shortlist with scoring rationale, risk flags, and market context that gives the buyer what they need to decide confidently.
The agent does not replace buyer judgement. It gives buyers better information to act on: a ranked shortlist with scoring rationale, risk flags with supporting data, and market benchmarks at the point of negotiation. Procurement professionals retain control over consequential decisions. What changes is the quality of the data those decisions are based on.
Most procurement teams discover suppliers through existing databases, personal networks, and reactive market searches triggered by an active sourcing need. Each channel carries an inherent selection bias: databases reflect whoever registered, networks reflect the buyer's existing knowledge, and reactive searches only happen when someone has bandwidth to run them. The result is a supplier pool that is systematically narrower than the available market.
In a 2026 McKinsey survey of more than 300 global procurement leaders, 55% reported flat or shrinking budgets while every respondent said savings targets had increased. Spend managed per full-time position is now roughly 50% higher than it was five years ago. In practice, that pressure gets absorbed by compressing strategic work, and supplier discovery is one of the first activities to be cut, particularly for lower-value categories where the spend does not justify weeks of research time.
Even when a procurement team has the bandwidth to search, supplier data is rarely held in one place. Financial ratings are in one system. Certifications are in another. ESG performance is in a third, if it has been tracked at all. Historical purchase order data, delivery performance, and quality audit results are split across ERP, quality management, and procurement platforms. Fragmented tools produce fragmented information, and no single buyer can synthesise this picture at scale.
Static databases and siloed systems mean supplier decisions are taken on incomplete information, even when that information exists somewhere in the organisation. The risk of selecting the wrong suppliers or missing the right ones is therefore structural. Agentic AI addresses this by drawing from all relevant data sources simultaneously and assembling a complete supplier profile before the buyer reviews candidates.
For most procurement organisations, tail spend categories receive the least research relative to their complexity. Categories with many available suppliers, low individual spend, and high administrative overhead, from office supplies to specialist technical components, are precisely the kind of work that gets approximated rather than investigated. Buyers default to their existing suppliers or repeat their last sourcing decision because a full market scan is not justifiable for a small purchase order.
Agentic AI changes the economics of tail spend. The marginal cost of a supplier scan is negligible when agents are already monitoring the market continuously. Teams using agentic AI and automated RFQ workflows report average cost savings of 5-8% on spend that was previously accepted at the supplier's initial price. These are cost savings that no manual process could capture, because the search itself was never affordable enough to run.
What makes agentic AI different from a database search is step one: the monitoring is continuous. When a sourcing need arises, the buyer reviews a shortlist that has already been assembled and maintained over time. For most organisations, this compresses the research phase from weeks to hours.
Effective agentic AI sourcing draws from two categories of data. Internal sources include ERP data, historical purchase orders, spend analysis outputs, quality management records, existing assessments of current suppliers, and preferred supplier lists. External sources include trade registries, market databases tracking market trends and pricing movements, ESG rating services, news and regulatory feeds, credit rating agencies, and certification authorities.
The combination is what gives agentic AI its analytical range. Agents evaluating potential suppliers need internal spend data to assess whether proposed pricing is competitive against what the organisation has paid historically, and external ESG data to verify sustainability credentials against regulatory requirements. Internal data alone produces a narrow view of available suppliers and external data alone lacks the organisational context to produce accurate assessments.
Agents produce reliable outputs when the supplier data they work with is complete and consistently structured. When financial ratings, certifications, spend history, and delivery performance records are scattered across disconnected systems, agents work from incomplete information and the results procurement teams operate on become unreliable. Consolidating supplier data quality before deploying agents is what allows the procurement workflow to shift from manual gathering to predictive analytics and reliable shortlists. Teams that invest in data quality upfront reach successful implementation faster, see more reliable agent outputs from day one and enable data-driven decisions from the start.
The most immediate outcome procurement leaders report when deploying agentic AI for supplier discovery is an expansion of the active supplier pool, particularly in categories previously managed with minimal research. When the cost of running a supplier scan approaches zero, teams apply it more broadly. Categories that previously operated on two or three preferred suppliers are opened to autonomous sourcing and competitive evaluation.
Amer Sports, a global sporting goods group whose brands include Wilson, Salomon, and Arc'teryx, grew its sourcing events from four to 31 in a single quarter after deploying Mercanis. The number of active suppliers across indirect procurement categories reached over 2,800.
When procurement teams have access to a broader, more competitive supplier market, they negotiate from a stronger position. Based on Mercanis customer data, teams using agentic AI for supplier discovery report an average of 23% better supplier terms compared to those relying on traditional methods. Amer Sports reported 18% spend savings across indirect categories over 12 months.
Part of this improvement comes from access to more suppliers. Part comes from better information at the point of negotiation: an AI agent that has already compared pricing and terms across dozens of suppliers gives the buyer a clearer picture of what the market will bear, and a stronger basis for contract negotiation. The buyer enters a negotiation with market data rather than estimates, which changes the dynamic regardless of which supplier they ultimately select.
The research and initial qualification phase is where most sourcing cycle time is spent. Agentic AI compresses this phase significantly because much of the work runs in the background. Franz Morat Group, a precision engineering manufacturer, reduced the time required for RFQ price analysis from 30 minutes to 2 minutes per event after deploying Mercanis. Multiplied across hundreds of sourcing events per year, this represents a structural shift in how procurement team capacity is allocated.
Continuous market monitoring gives organisations earlier visibility of supply chain risk. Agentic AI tracks supplier financial health, market signals, public company disclosures, and certification status, surfacing a risk indicator before it becomes an operational problem. Risk monitoring agents score suppliers continuously on multiple dimensions, financial stability, geopolitical exposure, compliance status and flag elevated risk in real time. When a primary supplier encounters difficulties, pre-qualified alternative suppliers are already in place.
Across deployments, integrating AI reliably starts with data quality: it is the factor that most consistently determines how quickly buyers trust and act on agent outputs. An agent drawing on clean, current, and well-structured supplier data produces reliable risk assessments and accurate shortlists. An agent working with inconsistent master data, unmaintained supplier records, and siloed category information produces results buyers cannot act on.
Before deploying agentic AI for supplier discovery, procurement teams should complete an honest audit of their procurement data: where it lives, who maintains it, how current it is, and whether it is structured consistently enough to support automated reasoning. Teams that skip this step and move directly to agent configuration typically spend more time troubleshooting inconsistent outputs than they would have spent on the data preparation upfront.
Defining clear governance parameters for agentic AI is as important as the technical implementation. Organisations need to specify which decisions agents can take autonomously and which require human oversight before any action is taken. In practice, this means defining the qualification criteria agents apply, the thresholds for escalation, and the process for handling exceptions. Based on Mercanis implementation experience, teams that complete this governance framework before go-live reach full buyer adoption faster and escalate fewer exceptions back to manual review.
Agentic AI in procurement changes how procurement work is structured. Without deliberate change management alongside technical implementation, teams absorb the efficiency gains but the strategic benefit does not follow. Buyers need support in redirecting recovered capacity toward strategic initiatives: category strategy, supplier relationship development, and commercial decision-making rather than data collection and research.
As an AI-native procurement platform, Mercanis has integrated specialised AI agents across every procurement workflow. In supplier discovery, this typically means separate agents for market scanning, compliance checks, pricing analysis, and risk monitoring. Two agents are most directly involved in finding and qualifying new suppliers.
The Discovery Agent identifies new suppliers based on category requirements, searching across market databases, trade registries, and supplier networks to surface candidates that match the procurement team's criteria. It runs continuously, meaning by the time a sourcing event begins, buyers already have a pre-assembled longlist of pre-qualified suppliers.
The Market Intelligence Agent monitors global pricing trends and market conditions, delivering live data directly into sourcing workflows. For strategic procurement functions evaluating new suppliers, this means access to current benchmarks at the point of qualification, so buyers can assess whether a new supplier's proposed terms are competitive before investing time in a full sourcing event.
Both agents are part of Mercanis SRM, which includes a central repository holding complete 360-degree profiles of all suppliers: financial stability ratings including Dun & Bradstreet scores, certifications and audit status, ESG and sustainability scores, historical pricing data, and digital capability assessments. Mercanis enables the integration of AI agents into existing procurement systems and ERP platforms, giving organisations access to autonomous sourcing capabilities within their current system landscape.
For a full breakdown of how multi-agent orchestration works across the procurement lifecycle, see the guide to agentic procurement orchestration.
See how Mercanis AI Agents automate supplier discovery: book a demo.
Agentic AI in supplier discovery is the use of autonomous AI agents that independently search, evaluate, and qualify suppliers by analysing structured and unstructured data across internal and external sources. Unlike traditional procurement automation, agentic AI systems reason across multiple steps and operate continuously within governance parameters set by the procurement team.
AI agents identify and qualify suppliers through a continuous, multi-step process. They monitor trade registries, market databases, certification authorities, and news sources around the clock, not only when a sourcing event is triggered. Each new candidate is automatically evaluated against criteria the procurement team has defined. Agents then verify credentials directly against issuer databases and run compliance checks before a supplier reaches the shortlist.
Static databases depend on vendors submitting their own profiles and on buyers running manual searches. Agentic AI systems actively monitor the market, evaluate suppliers against objective criteria, and identify suppliers that would not surface in a conventional database search, including specialist small and medium-sized suppliers and emerging suppliers in a category. Unlike generative AI tools that respond to prompts, agentic AI operates continuously without manual intervention, evaluating financial stability, compliance, ESG performance, and historical data simultaneously across all candidate suppliers.
Procurement teams deploying agentic AI agents report three measurable outcomes: a broader and more competitive pool of suppliers, better terms from increased competitive pressure, and faster sourcing cycles. Amer Sports grew from four to 31 sourcing events per quarter and achieved 18% spend savings over 12 months. Based on Mercanis customer data, teams using agentic AI report an average of 23% better supplier terms compared to those using traditional methods.
Fabian Heinrich is the CEO and co-founder of Mercanis. Previously he co-founded and grew the procurement company Scoutbee to become a global market leader in scouting with offices in Europe and the USA and serving clients like Siemens, Audi, Unilever. With a Bachelor's degree and a Master's in Accounting and Finance from the University of St. Gallen, his career spans roles at Deloitte and Rocket Internet SE.
