Somewhere in your inbox this week sits a pitch for an AI platform. It promises to handle creator discovery, or campaign reporting, or outreach. The pricing is on page one. What happens when you leave is nowhere.
Before signing anything (a platform contract, a consulting agreement, anything), there is one question worth putting to the other side of the table:
If we stop paying, what still works?
Most managing directors never ask it. Yet the answer tells you more about a vendor than any feature list, because there are two different ways to build an agency's AI stack (the set of tools, accounts and files your automation runs on), and they look identical in a demo.
Model one: renting
The default model works like this. A software company hosts a platform. Your team logs in, uploads data, builds workflows in their interface, and pays a monthly fee per seat or per campaign.
Three things quietly accumulate inside that platform. Your data: creator lists, rates, performance history, client details. Your configurations: the workflows, templates and report formats your team spent weeks tuning. And your context: every note, rating and lesson about which creators deliver and which clients need what.
Cancel, and all three stop existing for you. Some platforms offer an export, usually a spreadsheet dump no other system can read meaningfully. The workflows and the context are simply gone. You start again from zero somewhere else.
There is a second cost hidden in the model. Most of these platforms run on the same underlying AI providers (OpenAI, Anthropic, Google) that everyone else uses. When you pay per seat, you are paying the raw usage cost plus their margin on it: a markup on electricity, effectively, with the meter in their name.
To be clear: renting is fine for commodity software. Nobody should run their own accounting software on their own servers. Discovery databases like Modash or Lefty do a job and do it well. The problem starts when the thing you rent becomes your operational memory.
Model two: owning
The alternative looks like this.
The AI account is in your agency's name. You pay the provider directly, at cost: no intermediary margin. The system runs on infrastructure you control: your accounts, your logins, your billing. The knowledge layer (briefs, creator notes, rate history, tone-of-voice guides) lives in plain text files in a folder you own. Any team member can open them. Any future system can read them. And the connections between the AI and your tools use MCP (an open standard that lets an AI read and act on other software) rather than proprietary plumbing that dies with the vendor relationship.
Under this model, the question has a clean answer. Stop paying the consultant, and the system keeps running. The files stay readable. The accounts stay yours. What you lose is the person improving it, not the thing itself.
If we stop paying, what still works? Ask it before the demo, not after the renewal notice.
The part most worth owning
Run automated workflows for a few months and here is what they produce. The scripts are replaceable; a competent builder can recreate them. What is not replaceable is the accumulated context: which creators actually delivered on time, what each client considers "on brand", what a fair rate looked like in that niche last quarter, why a certain campaign structure failed.
We have written before about why in-house AI experiments tend to stall (Why agency AI projects plateau within weeks), and the short version is: without a structured memory, every AI session starts from a blank page. Fixing that means building what we call an AIOS, an AI operating system: one knowledge layer both your team and your AI work from. It is the single most valuable asset the whole exercise creates. Which is exactly why it should never live inside someone else's platform.
Six questions before you sign anything
Put these to any AI vendor or consultant, in writing if you like:
- Who owns the AI account? If usage runs through their account, you have no cost visibility and no continuity. It should be in your name.
- Where does client data live? "In our secure cloud" is not an answer. Which provider, which region, and can you delete it on demand?
- What happens on exit? Ask for the specific list of what keeps working the day after the contract ends. Vague answers here are the whole warning.
- Is the knowledge layer portable? Notes, briefs and history should be plain files you can open without their software. A proprietary database is a lock, however friendly the interface.
- Are usage costs marked up? If AI usage is bundled into a licence fee, you cannot see the margin. Direct billing from the provider means you pay what it costs.
- Can our team change things without you? A good setup is documented well enough that non-technical staff can make everyday tweaks. If every change needs a ticket, you have hired a landlord.
A vendor with a solid offering answers these plainly. Discomfort with the list is information too.
The trade-off
Owning asks slightly more of you. Someone at the agency has to open the AI provider account, add a payment card, and hold the passwords: an afternoon of admin, done once. You will receive two or three small invoices instead of one bundled one. And you carry a little more responsibility: the system is yours, the way your website or your email domain is yours.
That trade is worth making. Here is what rides on it. As we laid out in The 10-hours-per-creator problem, the number we hear consistently from senior people at Benelux PR firms is roughly ten hours of manual coordination per creator per campaign: around 1,500 hours a year for an agency running five 30-creator campaigns. Automate one step of that workflow well (a 70–90% reduction on that step) and hundreds of those 1,500 hours come back every year. Whoever owns the system that removes those hours holds real leverage over your margin. That should be you.
"If we stop paying, what still works?" takes ten seconds to say. The answer sorts an inbox of AI pitches faster than any feature comparison, and it separates the partners building an asset for your agency from the partners building a dependency on themselves.