Procurement SaaS isn’t dead. Even though in February this year, investors behaved as if SaaS more generally was going that way.
A wave of AI agent launches came first. Then, on 3 February, software stocks lost around $285 billion in market value in a single day. Since then, the agentic AI vs. SaaS debate has reached every procurement event I attend.
But, the people leading that debate on stage and all of the cool new startups in this space are not a representative segment of the wider procurement profession’s demographics.
Procurement is one of the most conservative functions in any business. The leaders you meet at those events, deploying AI-native tech and hiring entrepreneurially-minded team members, are the top 5-10% of the profession.
Most conversations I have outside of this bubble, especially in the mid-market, are about chasing approvals over email and reporting savings on a spreadsheet.
But AI does change one thing quickly: the speed of building and piloting. And that speed could also help procurement SaaS vendors, much as it could be seen as a threat too.
Procurement moves slower than the hype
Plenty of large companies are still signing multi-year contracts for legacy S2P (source-to-pay) suites. They budget for long implementations and pay systems integrators to configure them.
The reasons make sense to the people signing. The suite already connects to their ERP system. Infosec and internal audit teams know how it works. And nobody gets fired for choosing the perceived safe option.
Is it really the safe option though, in such a time of rapid change?
AI will speed up the decline of legacy vendors with clunky interfaces and closed architecture. They were losing ground to newer tools anyway. AI shortens this timeline.
The challengers face a different problem.
AI coding tools let procuretech companies build and ship prototype features faster than ever. But enterprise procurement teams simply can’t adopt all of these new product features at that pace.
Every new feature still needs security sign-off, user training and a change to an existing process. The bottleneck has moved from the vendor to the buyer.
Gartner predicts that by 2030, AI agents will replace or absorb 35% of point-product SaaS tools. If we flip that, then it assumes that 65% of the current SaaS tools will still survive this “revolution”.
So, what we’ll likely see is that AI won’t replace traditional procurement software as a category, because the two now depend on each other. An agent needs a system to act on. The system needs agents to stay relevant.
Where is might kill off some of the legacy players is if they follow a closed-off strategy and refuse to open their APIs to AI-native tools. This new breed of agentic tool can complement the legacy suite by leveraging some of the unstructured data that is stored in these systems.
AI-native procurement SaaS already exists
Intake and orchestration software is the clearest example.
Zip, Omnea, ORO Labs, Levelpath, Vertice and Pivot are good examples. They use AI to read free-text requests, route approvals and chase stakeholders. Their founders built them with AI at the core. Everyone still calls them SaaS though, which kind of makes things confusing to those looking from the outside in.
The same shift is happening across the market. The Hackett Group published its 2026 Procurement Agenda and Key Issues Study in March. 69% of organizations access AI through features embedded in existing procurement platforms.
Gartner expects 33% of enterprise software applications to include agentic AI by 2028, up from less than 1% in 2024.
The major players in the stack like SAP, ServiceNow and Salesforce, whose products sit close to or at the “system of record” point, already ship their own agents (Joule, Now Assist and Agentforce).
The line between “an agent” and “an application” already looks blurry. It will likely get blurrier.
Then there’s the debate around “what is considered AI-native anyway?”, which we could spend a whole new article looking at.
The limits of building your own agents
No-code and low-code make simple automations cheap
3 years ago, an AI-powered workflow needed a developer, a budget line and a slot in the IT backlog. Today, a tech-savvy procurement analyst can build one in n8n, Make or Zapier in an afternoon.
They can also build their own workflows inside intake and orchestration tools, using the platforms’ no-code builders. The major players in this space all have a feature that enables build-your-own no-code workflows.
Think of an agent that reads supplier emails, extracts order confirmations and updates a tracker. That is useful. It’s also narrow use case, and it easily sits on top of and complements your core systems. The key point here is that it doesn’t replace or compete with them.
Vibe coding meets the CISO
“Vibe coding” is a term that means describing an app or an agent in plain English, and letting an AI model write the code. It is brilliant for building prototypes.
Production is another matter. Veracode’s 2026 GenAI Code Security Report found that 44% of AI code generation tasks introduced a risky security vulnerability.
Procurement systems hold supplier bank details and contract pricing. Your CISO (chief information security officer) won’t sign off on an unreviewed app that touches that data. Nor should they.
Who will build and maintain the agents?
Agents handle simple, rules-based workflows well. Building a reliable one takes more than a login to an AI platform. You need to understand your approval matrix, your ERP data model and every exception your team handles by instinct.
How many procurement teams have that expertise today?
Gartner surveyed 101 CPOs (chief procurement officers) in early 2026. Only 36% were very confident in their ability to redesign the function for AI.
Maintenance is the hidden cost. Agents break when a system updates or a policy changes. Gartner also predicts that organizations will cancel over 40% of agentic AI projects by the end of 2027. It cites escalating costs, unclear business value and inadequate risk controls.
| Variable | Build your own agents | Buy agentic SaaS |
|---|---|---|
| Upfront cost | Low, using consumption-based no-code tools | Usually an annual subscription, possibly with usage-based element |
| Security sign-off | Every workflow potentially needs an infosec review | Vendor holds security certifications such as ISO27001 or SOC-2 |
| Maintenance | Your team fixes every break and deals with every user issue | Vendor ships updates and fixes; has customer success team |
| Skills needed | Procurement and technical expertise all in one team | Procurement expertise, market knowledge, general vendor management |
| Best for | Narrow, repetitive, internal user tasks; currently done in email or Excel | Core processes that need audit trails |
Agentic AI vs SaaS: systems of record still rule
Large enterprises run on a legacy tech stack that acts as their “system of record”. That means the ERP, plus some key layer one platforms such as Coupa, SAP Ariba, Salesforce and ServiceNow. These hold the official version of every supplier and payment. Possibly contracts and SLAs too.
Agents don’t replace a system of record. They read from it and write back to it, making sense of data dispersed through clunky systems with restricted access rights and poor UI/UX.
And this is where agentic AI orchestration in procurement gets interesting.
An agent can pick up a request in Slack and check the contract in a CLM tool. It can then raise the PO in the ERP or Source-to-Pay suite. Nobody copies data between screens.
The stumbling block is access.
Orchestration only works if each system has open APIs (application programming interfaces), which let software talk to other software. Many older ERP and S2P installations have limited or expensive API access.
If the agent can’t get in, it can’t act.
| Layer | What it does | Examples |
|---|---|---|
| System of record | Stores the official data on suppliers, contracts, POs and payments | SAP S/4HANA, Oracle, Coupa, SAP Ariba, Salesforce |
| AI-native SaaS | Runs specific procurement processes with agents built in | Pactum, Keelvar, Lio, Rivio, Fairmarkit, Archlet |
| Orchestration layer | Connects systems and routes work between them | Zip, Omnea, ORO Labs, Levelpath, Pivot, Vertice |
| DIY agents | Automate narrow tasks for one team | n8n, Make and Zapier workflows |
Front-line procurement gets the biggest boost
Category managers and sourcing managers gain the most from AI agents in procurement right now. Agents draft RFx (request for information, proposal or quotation) documents, summarise supplier proposals and prepare negotiation briefs.
That matters because workloads keep growing. The Hackett Group expects procurement workloads to rise 8% in 2026, while headcount and operating budgets decline. Agents let each category manager cover more spend with the same hours, allowing them to dedicate their time to the important tasks.
Admin work, research and market intelligence can increasingly be delegated to AI.
Some of this runs inside dedicated tools. Pactum’s agents run negotiations with long-tail suppliers that a human team would never get round to. Keelvar, Fairmarkit and Archlet automate sourcing events and bid analysis.
Other teams just use a general AI assistant, such as Claude or ChatGPT, to research a supplier before a negotiation.
None of that removes the need for existing procurement technology. The agent makes the category manager faster inside those tools.
Gartner’s Fareen Mehrzai adds a warning. Without redesigning roles and processes, she says, generative AI productivity gains remain confined to the individual level.
Redesigning roles starts with the human-in-the-loop question. When should an agent decide the outcome, and when should a person? The answer depends on the value and risk of each decision:
- Agent decides: low-value, repeatable tasks with clear rules, such as catalogue reorders or comparing quotes for tail spend.
- Agent recommends, human decides: supplier selection, contract renewals and anything above an approval threshold.
- Human leads: strategic supplier negotiations, disputes and decisions that carry legal or reputational risk.
Your delegation of authority policy is a good starting point. If a junior buyer can approve a decision today, an agent is a candidate to make it too.
Tools already exist to do that: agents negotiate within limits that a human sets.
Agentic AI can’t fix bad data
This is the most important point in the whole “procurement SaaS isn’t dead” debate.
An agent is only as good as the data it acts on. Duplicate supplier records and miscoded spend lead to bad decisions, made faster.
In February 2025, Gartner predicted that organizations will abandon 60% of AI projects unsupported by AI-ready data. In the same research, 63% of organizations lacked AI-ready data management practices or weren’t sure they had them.
There is some good news. LLMs (large language models) handle unstructured data far better than older machine learning models. They can read a PDF contract or a messy free-text request and pull out usable fields. That unlocks data your team could never use before.
It won’t autonomously clean up a vendor master with 4 versions of the same company Someone still has to own the data. Good SaaS tools help by enforcing structure at the point of entry.
Conclusion
The agentic AI vs SaaS debate assumes one side has to win.
What I see is convergence. And I’m not the only one. Joel Collin-Demers reached a similar conclusion when looking specifically at the intake and orchestration vs. source-to-pay debate.
SaaS vendors are building native agents into their products. Agentic point solutions are cropping up left, right and centre. Procurement teams are building their own small agents that solve specific bottlenecks.
Within a few years, we will all stop caring which is which.
A “swarm of agents” is a different story. Building and maintaining dozens of connected agents in-house needs people who understand both code and procurement.
Very few teams have them today. This will change over time, but we’re still a long way away.
The scarce resource is the procurement professional who can build as well as buy. Could they become the hottest property in the procurement hiring market?
I’d bet on it.
