Merlin Agentic Sourcing starts before the request is raised.
Most agentic sourcing tools start with a request, a form, or an event. MAS starts with a business problem and decides what to do about it.
Three things are being called "agentic sourcing." They're not all doing the same work.
Intake-first tools have added AI to a request form. Point-vendor agents reason over their own slice of data. Autonomous negotiation has been extended past tail spend. This page draws the line between MAS and each of them - on the dimensions a buyer can actually evaluate.
The Full Comparison
Six dimensions. Four comparators. One phrase per cell.
| Dimension | Keelvar Agentic sourcing point solution | Legacy S2P Suites Legacy multi-module S2P suites | Legacy ERP Procurement Procurement inside ERP suites | Point Solutions Intake-only / niche tools | Merlin Agentic Sourcing Strategic sourcing advisor |
|---|---|---|---|---|---|
| Starting point | A sourcing event the user wants to run | A trigger inside the vendor's workflow | An intake request inside the ERP | A requisition someone has written | A business problem in plain language |
| Data surface | Bid data, supplier responses, scenario inputs | One suite's data slice; uneven across modules | ERP master data; procurement is a sub-system | Form fields, routing rules, intake metadata | Full S2P fabric + benchmarks + external market + MCP |
| Scope | Sourcing events at volume and complexity | Whatever module the customer has licensed | Procurement at parity with finance, HR, supply chain | All requests, no analytical depth | Strategic categories. High-value spend |
| Action set | Run the event better -faster, optimised, automated | Task-level AI features (summarise, draft, suggest) | Bid analysis, supplier summary - inside the event | Route, approve, escalate | Run, renegotiate, consolidate, diversify, hold |
| Human role | Sets the rules, approves the award | Approves task-level AI outputs | Approves at ERP workflow gates | Approves at intake gates | Approves the strategy, then each consequential decision |
| Delivery model | Workflow. Real-time | Workflow. Real-time | Workflow. Real-time | Workflow. Real-time | Workflow. Real-time. Minutes |
← Swipe to compare all columns →
Four comparisons. Four different kinds of work.
Where MAS lines up against each comparator - and where it diverges.
Keelvar automates the event. MAS decides whether to run one.
Keelvar is the closest live peer in the market - an agentic sourcing platform that builds the event, engages suppliers, runs the negotiation, analyses bids, and recommends an award. MAS reasons one step earlier: over the full procurement data fabric and the market beyond it, to recommend whether an event is the right action at all - or whether the answer is renegotiate, consolidate, diversify, or hold.
Legacy suites added AI to discrete tasks. MAS reasons across the workflow.
Established S2P suites have layered generative AI onto specific tasks - drafting an RFP, summarising a contract, suggesting questions. Each task is faster; the workflow is unchanged. The human still initiates and progresses every step. MAS takes a business problem and reasons across the full S2P data fabric to recommend the right action from the full strategic action set.
ERP procurement is a module. MAS is procurement-native.
Procurement inside an ERP suite competes with finance, supply chain, and HR for engineering bandwidth. Agentic features here are task-specific copilots - bid analysis, supplier response summary - layered into a sourcing event that has already been decided on. MAS is built ground-up for procurement, with an agent that reasons over the full data fabric and decides what work needs doing before any event exists.
Intake tools digitise the request. MAS starts before the request is raised.
Intake-only tools automate the form, the routing, the approval. The starting point is a requisition someone has already drafted. MAS takes a business problem in plain language and recommends whether to source at all, what to source, and how. The request is the output of the conversation, not the input to it.
ANA gave procurement an agentic agent on the tail. MAS gives procurement an agentic agent on the categories that matter most.
Merlin ANA is the autonomous negotiation agent built for tail spend - triggered by a specific purchase request, optimised for volume, speed, and cost reduction at scale. No pre-analysis, because the spend type doesn't require it. Merlin Agentic Sourcing is built for strategic and high-value spend - triggered by a business problem or risk event, not a purchase request. Pre-analysis is the work.
A category manager has two situations on the same day. (1) Marketing needs 200 laptops next month- established supplier pool, run the negotiation, award. That's ANA. (2) A bundled IT services contract auto-renews in 90 days with a 196% escalation flagged. Not a purchase request, a business problem, requires a strategy before any event. That's MAS. Same category manager, same week. Different agents because different work.
Five questions for any vendor.
Each maps to one row of the matrix above. The answers place the vendor in the frame more reliably than any product demo.
What triggers your agent? Walk me through the first interaction.
If the answer is "the user fills in this intake form" or "the user kicks off a workflow," the agent is positioned post-decision. MAS's answer is: the user describes a business problem in plain language and the agent decides what comes next.
What data does the agent actually reason over? Show me the data sources panel.
Single-data-surface agents will name one or two sources. Full-fabric agents will name S2P + benchmarks + external market intelligence + adjacent systems. The recommendation is only as good as what the agent can see.
Show me an example where the agent's recommendation was "do not source."
Most agentic-sourcing tools cannot produce this output - their action set is "run an event, faster." MAS's recommendation set includes hold, renegotiate, consolidate, and diversify on equal footing with run-an-event.
Where in the workflow is human approval explicit, and what happens if I disagree with the agent?
Governed autonomy is the difference between agentic AI and rogue AI. The agent should act decisively inside guardrails and route to the human at every consequential decision - with the user able to modify or override at every checkpoint.
How long from the moment I describe the problem to the moment I have a ranked recommendation?
Workflow-class agents answer in minutes. Project-class engagements answer in weeks. If the answer is "days," the agent is doing project work without the project rigour.
Three patterns we see across competitive evaluations.
How vendor positioning gets ahead of vendor capability when a category is forming.
The intake-tool-rebranded-as-agentic pattern
Workflow-routing tools have added an LLM to the request form. The agent rephrases the user's description into a structured request and routes it.
The GenAI-feature-bolted-onto-sourcing pattern
Established sourcing platforms have added generative AI to discrete tasks - drafting an RFP, summarising a contract, suggesting questions. Each task is faster. The workflow is unchanged.
The autonomous-negotiation-extended pattern
Vendors who have shipped autonomous negotiation on tail spend (Zycus included, with ANA) have a strong claim to agentic procurement. Some have positioned that same engine as a strategic-sourcing solution.
Agentic sourcing is not an AI feature added to a request form. It is not an agent reasoning over one slice of data. It is not autonomous negotiation extended past its scope.
Agentic sourcing is an agent that takes a business problem in plain language, reasons over the full procurement data fabric and the market beyond it, and recommends the right action from the full strategic action set - with the human approving the strategy and every consequential decision inside it.
MAS gives procurement an agentic agent on the categories that matter most.
Ready to see MAS on your categories?
Bring a real category - a renewal, a tariff exposure, a should-cost question. We'll show you what MAS recommends, in minutes, in your workflow.
