Most automation pitches fail in the same four places, and you can catch all four in the first call. The automation pitch red flags below show up before you’ve signed anything, not after: a number with no denominator, a demo that never fails, no plan for leaving, and “AI-powered” doing the job a feature list should be doing. None of these require technical expertise to spot. They require you to ask one follow-up question and watch whether the answer gets specific or gets vague.
The automation pitch red flags that show up before the contract
If a vendor cannot answer “what happens when this is wrong” with a concrete mechanism, they have not built the failure path yet, and they are selling you the demo instead of the system. The demo is always the happy path. It runs on clean input, in a quiet room, with someone standing by to intervene if the model does something strange. None of that is what Tuesday looks like six weeks after go-live. The questions that separate a working system from a demo wearing a system’s clothes are mostly about what breaks, not what works, and I’ve written out the full set of automation vendor questions worth asking before money changes hands.
No number, or a number nobody can audit
The ROI figure in most pitches is gross, not net, and that is the tell. It counts the hours saved and ignores the hours spent reviewing output, fixing exceptions, and re-running the parts that failed silently. Unquantified ROI claims sound precise because they use a specific number, but precision and honesty are different things. A real automation ROI math exercise includes the maintenance hours and the exception rate as line items, not footnotes, and when you run it that way the payback period usually moves, sometimes by a lot. I go through a worked version of this in the honest maths of automation ROI, and the gap between the pitched number and the audited one is the whole point of that post.
No error handling, just a happy path
Ask what happens on the fifth retry, and if the answer is a shrug, that is the system’s actual design document. Error handling in agentic workflows is not a nice-to-have bolted on after launch. It is the difference between a system that degrades gracefully and one that returns confident nonsense to a customer at 2am because nobody specified what “I don’t know” should look like. A pitch that spends twenty minutes on capability and zero minutes on failure modes is telling you where the engineering effort actually went. The fallback order that should exist, retry, repair, degrade, escalate, is covered in AI agent error handling, and if a vendor can’t map their system onto that sequence, they haven’t built one.
AI-powered as the whole feature
“AI-powered” is a sourcing decision, not a feature. A pitch that leads with it instead of with what the thing does is pointing at the part that matters least to you. The model underneath is commodity. What you’re buying is the judgment about when to call it, what to do when it’s wrong, and how the output gets checked before it reaches a customer or a CRM record. A pitch that can’t describe that judgment in plain terms, and instead repeats “AI-powered” as though it were a capability, is usually a workflow with an API call added late. The ten-minute test for telling the two apart is worth running before the sales deck gets another slide: see agent or workflow.
No exit plan
If nobody can tell you how your data comes out, you don’t own the system, you rent it, and you’ve agreed to rent it indefinitely. An exit plan for automation should exist on day one, written down, covering data export format, how long historical runs stay accessible, and what it costs to migrate to a different vendor or back in-house. Most pitches skip this because asking about it implies the relationship might end, which is awkward in a sales conversation. Ask anyway. The vendors who have thought about your exit answer in one sentence. The ones who haven’t start talking about “partnership.”
What this adds up to
None of these four red flags, alone, should kill a deal. A vendor who can’t give you a precise ROI number but has a credible error-handling story might still be the right call for a process that’s genuinely yours and genuinely hard to staff, and that’s a build, buy, or commission question, not a pitch-quality question, deserving its own evaluation. But two or more of these together, in the same pitch, means the system was built to close the sale, not to survive contact with your actual process. Ask the uncomfortable follow-up before the contract, not after the first outage.