Every vendor pitch deck says "autonomous" now. Autonomous outreach. Autonomous campaigns. Autonomous everything. Almost none of them mean it the way you think.
Here's the tell: most of these platforms are excellent at starting work and mediocre at finishing it. They'll draft the email, spin up the ad set, generate the report. Then they stop right at the point where a human has to step in and clean up, approve, or fix something. That's not autonomy. That's a very expensive first draft machine.
If you're evaluating vendors, don't ask "what can it do?" Ask "what happens after it does that?" This checklist is built around that one question.
Start With Money: Who Actually Approves Spend
This is the fastest way to separate marketing from theater. Any platform touching budget needs a clear, tested answer to these:
- Can the platform reallocate budget across channels without a human clicking approve?
- What's the dollar threshold before it requires sign-off, and who set that number?
- If spend spikes 3x overnight due to a bug or bad signal, what stops it? Is there a hard ceiling, or does it just keep going until someone notices?
- Can you see a real audit log of every spend decision it made, with the reasoning, not just the outcome?
If the vendor's answer is vague, or if "autonomous spend" actually means "generates a recommendation for you to approve manually every time," you're not buying autonomy. You're buying a suggestion box with a subscription fee.
The Exception Question: What Happens When Things Go Sideways
Demos are always the happy path. The customer answers politely, the data is clean, the campaign performs as expected. Real marketing is not like that. Ask vendors to walk you through their exception handling, not their ideal-case workflow.
- What happens when a lead gives an ambiguous or off-script response?
- What happens when a data source goes down mid-campaign?
- Does the system know when it's confused, or does it plow ahead confidently with a wrong answer?
- Is there a defined escalation path, or does "exception handling" just mean it silently fails and someone finds out three days later?
Ask for a specific example of a failure the platform caught and routed correctly, not one it caused. If they can't produce one, that capability probably doesn't exist yet.
What Breaks Without a Human in the Loop
This is the core diligence question and most buyers never ask it directly. Every platform has a point where the automation quietly depends on a person. Find that point before you sign, not after go-live.
- If your team stopped checking the dashboard for two weeks, what would go wrong first?
- Which specific tasks does the platform claim to do "autonomously" that actually require a human to review before anything ships?
- Does the system need a person to interpret ambiguous results, or can it make a defensible decision on its own and log why?
- What's the actual ratio of tasks it completes unattended versus tasks it hands back to your team half-finished?
A platform that needs a human for every judgment call isn't autonomous. It's just automation wearing a nicer word.
Does It Close the Loop, or Just Open More Tabs
"End-to-end" gets thrown around loosely. Push vendors to show the actual last mile.
- Does it just generate content, or does it publish, monitor performance, and adjust based on results without a rebrief?
- Does it just flag a churn-risk customer, or does it also trigger the retention sequence and confirm it landed?
- Does it report performance, or does it act on the report by reallocating effort or budget?
- Can you point to one full cycle, trigger to outcome, where zero manual steps happened in between?
If a vendor can only show you the first two steps of a five-step process, that's not end-to-end. That's a head start.
Integration Reality Check
Autonomy that only works inside one walled garden isn't very autonomous in practice. Your stack is messy. The platform needs to survive contact with it.
- What happens when your CRM data is incomplete or inconsistent? Does the system flag it or work around bad data silently?
- Can it write back to your existing systems, or only read from them?
- What's the actual setup time to get from "signed contract" to "first fully autonomous cycle completed," in weeks?
- Who on their side owns fixing an integration break, and what's the SLA?
Accountability: Who Owns the Outcome
The last section of any real due-diligence conversation should be about ownership, because that's where vague platforms get exposed fastest.
- If the platform makes a bad decision that costs you money or a customer relationship, who's accountable? Vendor, you, or "the algorithm"?
- Is there a named team on their side that monitors outcomes, or is it entirely self-managing with no human oversight anywhere?
- Can they show you a customer reference where the platform ran unattended for 90+ days with measurable results, not just a pilot?
Any vendor confident in their product will answer these without flinching. Vendors who dodge, redirect to feature lists, or start talking about "roadmap" are telling you something important. The gap between what they're selling and what actually ships tends to be exactly where these questions point.
The Short Version
Before you sign anything, get answers on spend approval, exception handling, what breaks without a human, and what a complete closed loop actually looks like in their product, not their pitch. If a vendor can answer all four cleanly, with real examples, you've probably found something that finishes work instead of just starting it. If they can't, you've found a very well-designed to-do list generator.