Renewal season is the one moment in the year when a software budget becomes visible. Every per-seat contract lands in the same four weeks with a price increase attached, and every one of them is signed by somebody who has other work to do.
That is the window where the agentic shift is either captured or missed. Not on a strategy away-day, and not in a vendor demo. In the renewal quote sitting in an inbox right now, where a seat is either renewed for another twelve months or cancelled and replaced with a workflow an agent executes.
The renewal quote arrives, and nothing gets decided
Nobody chose the current stack. It accumulated, one urgent purchase at a time.
The renewal email lands in the same week as eleven others. A number goes up by six to nine percent, a finance lead approves it because the tool is still open in somebody's browser, and the seat count stays frozen at whatever it was during the year the company hired fastest.
That is how a software budget turns into a historical document rather than a decision. One project board per team. Two document suites because two departments each picked their own. A data extraction tool bought for a migration that finished in 2024. A reporting layer that four people have open and one person actually updates.
Agentic AI has made that accumulation newly expensive. An agent does not occupy a seat. It needs an API, a consumption budget and a workflow worth handing over. Every seat you keep for work an agent can already do is rent paid on an empty chair, and the rent is not small.
What an autopilot renewal costs a 40-person company
The waste is never in the price per seat. It is in the seats that stopped doing work two years ago.
Published benchmarks put business software spending at 4,800 to 5,600 US dollars per employee per year, with duplicate tools and unused seats as the reliable warning sign. Take a mid-range figure of 4,500 euro per employee for the arithmetic below, which sits inside that range at recent exchange rates.
A 40-person service company at 4,500 euro per employee is running a 180,000 euro annual software budget. If 46 percent of those applications are underused, roughly 82,800 euro a year is paying for licences nobody opens. A conservative first pass recovers 15 to 25 percent of the total base, which is 27,000 to 45,000 euro a year. Renew on autopilot and that money does not disappear from the P&L. It simply keeps buying nothing.
The point is not that software is waste. It is that this specific money is no longer buying capability. It is buying continuity, because cancelling a tool means somebody has to check whether the workflow around it still matters. That check takes an afternoon. The renewal takes a signature.
Why this renewal season is genuinely different
The pricing model underneath business software is changing from per-human to per-unit-of-work, and 2026 is the year the two models started competing in the same procurement.
In February 2026, roughly 300 billion dollars of market value left software and data companies in a single trading day, and average forward earnings multiples for software compressed from about 39 times to about 21 times. The reason traders gave was straightforward: if an agent can do the work, the seat disappears, and a licence count that only grows with headcount stops being a growth story.
The pricing has already moved. Autonomous agents are now sold per conversation or per unit of work, while the incumbent productivity suites still charge per user per month. Those two line items can sit in the same renewal pack, which means the comparison is finally visible to whoever signs it.
For an EU SME the practical translation is simple. The question at renewal is no longer which plan tier you need. It is which of these recurring charges is really a headcount-shaped subscription to work that a machine now performs.
The seat audit: four passes before you sign anything
Four passes, one afternoon, and a decision per line item. This is the version we run with clients before the first agent is built.
- Inventory the withdrawal, not the invoice. Pull twelve months of card and invoice data, not the procurement list. Shadow spending lives on team cards, and the tools that hurt are the ones nobody remembers approving.
- Score real usage against the seat count. Most vendors expose active user data. Compare monthly active users with licensed seats. Any tool sitting below half utilisation is a cancellation candidate this cycle, before any AI is involved.
- Ask what the tool absorbs. For each surviving tool, name the actual job it does. Invoice data capture is a job. A status board is a job. A shared drive with a naming convention is a job. Jobs are what agents take over, and the tool that existed only to hold the job goes with it.
- Fund the replacement from the recovery. Do not present the agent layer as new spending. Present it as a transfer: the seats cancelled in step two pay for the workflows automated in step three. That single framing turns an AI budget request into a reallocation, which is a far easier signature to get.
The decision matrix: what to cancel, what to defend
The failure mode is not cancelling too little. It is cancelling the system that holds your numbers. Sort by category before you sort by cost.
| Category | Typical tools | Verdict at renewal |
|---|---|---|
| Systems of record | Accounting ledger, invoicing, payroll, tax reporting, e-signature | Keep. Agents read and write into these. They never become them. |
| Per-seat collaboration | Document suites, note apps, chat, project boards with overlapping features | Consolidate to one. Then cut the duplicate seats outright. |
| Workflow and status tools | Ticket queues, approval chains, hand-off trackers | Replace. This is the layer an orchestration agent owns, and it is where the visible savings sit. |
| Data entry and document handling | Invoice capture, form processing, extraction, re-typing tools | Replace. An agent writes straight into the ledger. The tool in the middle was a human workaround. |
| Reporting and dashboards | BI seats, metric boards, scheduled exports | Replace. A scheduled agent report costs a fraction of a seat nobody logs into. |
| Compliance and audit tooling | Anything your auditor, bank or regulator reads directly | Keep, and extend. Verify your agent layer produces logs the auditor accepts. |
The EU-only risks the American coverage skips
Most writing on the agentic shift is written for US enterprises and investors. Three obligations apply to you specifically.
Article 50 transparency. From 2 August 2026, AI systems that interact with people must make that clear. If a customer-facing agent answers your support line or your enquiries, the person on the other side has to know they are not talking to a colleague. This is an interface requirement, not a legal review, but it has to be designed before launch rather than discovered during an audit.
Data residency and processor terms. When an agent touches customer records, you need a lawful basis, a data processing agreement with the vendor, and a retention rule you can point at. Self-hosted or EU-region deployments are the cleanest answer, and they remove the conversation about where your client data sleeps. Our EU AI Act guide for SMEs covers where the line sits between what you must do and what is merely sensible.
Staff AI literacy. If your team uses AI systems at work, there is an expectation that they understand the limits. A thirty-minute internal briefing on what the agent will not do is cheaper than discovering the answer in production.
The anti-hype filter
The same analyst houses forecasting the seat purge are forecasting its failure rate, and both numbers are true at the same time.
Gartner expects over 40 percent of agentic AI projects to be cancelled by the end of 2027, on cost escalation, unclear value and missing risk controls. The same research estimates that only around 130 of thousands of self-declared agentic vendors are genuinely agentic. Everything else is a chatbot wearing a workflow diagram.
A real agentic deployment can show you three things: a working loop in systems like yours, the failure behaviour when the model is unavailable or wrong, and who carries liability when the agent acts. A vendor that can only show slides is not selling you an agent layer, and cancelling seats to fund that slide deck is how the 40 percent failure rate gets produced.
This is why the audit comes before the build, and why the first agent should always replace a workflow you have already proven manually. If a human with a checklist can do the job reliably, an agent can eventually do it at a tenth of the cost. If nobody can do it reliably today, automation only makes the mess faster.
What the agent layer costs instead
Published pricing, because the fastest way to lose a renewal-season conversation is to make the buyer book a discovery call to hear a number.
| Tier | Price | What it covers | Best for |
|---|---|---|---|
| Automation Blueprint | Free | Full workflow and licence audit, usage scoring across your live stack, and an ROI map of the top three opportunities | Any SME heading into renewal season with an unread quote |
| Growth Infrastructure Setup | From 3,000 euro setup + 2,000 euro per month | Deploys the top three highest-ROI automations, including licence replacement where it applies | Companies that want the first recoveries inside one quarter |
| Scale | From 5,000 euro setup + 4,000 euro per month | Continuous build capacity for teams rewriting several workflows and retiring several vendors at once | Companies replacing a large part of the seat-based stack |
Every engagement starts with the free Automation Blueprint, so the ROI is mapped before you commit anything. If we cannot find a path to lower operating costs, you keep the map and pay nothing. The wider engine behind this, including what the automation layer looks like once the seats are gone, is covered in the one-person company operating system and in our 2026 cost benchmarks.
Where to start this week
Five actions, none of which require buying anything or waiting for a strategy document.
- List every subscription renewing in the next 90 days, with its annual value and its monthly active user count. That single spreadsheet is the whole project.
- Mark the top three by cost, then ask who would notice if each disappeared. “Nobody would notice” is a cancellation, not a negotiation.
- For every tool you keep, write the job it does in one sentence. If the sentence describes moving information from one place to another, it is an automation candidate.
- Check your customer-facing AI for Article 50 disclosure before 2 August 2026 obligations are behind you.
- Add up the cancellations and treat that number as the agent budget. A reallocation gets approved; a new line item gets deferred.
The renewal quote in your inbox is a decision whether you make it one or not. The companies that come out of this cycle with lower operating costs will not be the ones that bought the most agents. They will be the ones that stopped paying for chairs. To see how the audit runs on your own stack, start with the free Automation Blueprint or look at how an engagement works.
