When does a good automation become "set-and-regret"?
The moment you stop watching it. An automation is only "set and forget" until the day the inputs shift, an edge case shows up, or you simply forget it's running, and by then it's been quietly doing the wrong thing for weeks. The scary part isn't dramatic failure. It's how boring and invisible the damage usually is.
Let me give you the two that make me wince most, one on the marketing side and one on the CRM side. Take whichever hits closer to home.
The scariest one: an ad with no end date
An automated audience feeding paid ads with no cap and no end date is about as expensive a mistake as marketing makes. I've seen it plenty: someone builds an enriched audience, points it at Google Ads or LinkedIn, sets a rough budget, and lets it run. No end date. The audience keeps growing. The spend keeps going.
"An ad quietly runs for three months with no end date, spends fifty grand instead of the twenty you agreed, and takes the budget away from everything else that might have worked."
Two or three months later, when it was meant to run for a fortnight, they realise. Finance shrugs because technically it's within the quarterly marketing spend, so "it's fine". But it isn't fine. You agreed to spend ten or twenty grand on that audience and you've spent fifty or sixty. The company might well afford it, that's not the scary bit. The scary bit is you've just pulled that money out of everything else they could have spent it on and actually won with, and forced it onto one thing. You'd better hope that one thing delivered, because otherwise you're the reason their budget got wasted, and that's how you lose a client.
It's worse for small teams. I've done this for clients spending one or two grand a month in total. If that quietly burns with no results because everyone forgot it was running, that was their whole budget. And when you're the sole freelancer juggling ten projects for them, forgetting is exactly what happens.
The fix is dull and non-negotiable. I put daily limits, monthly limits, and an end date on everything, usually "ends on the 31st". Then on the 31st of each month I go through and roll the live ones forward one, two or three months. There's always a hard limit on how long anything can run before I've laid eyes on it again.
The other one: automating the human out of late-stage sales
Automating your late-stage deals is a fast way to scare off the buyers you worked hardest to win. This is the CRM-side horror story. People automate proposal creation, then automate chasing those proposals, right up to the point where they're spamming a serious buyer to sign.
Picture a deal worth real money, a big monthly contract, or twenty grand a month for three years. A salesperson has spent weeks building trust and convincing them you're the right choice. And now an automation is firing "just following up, when are you signing, let's get this set up" at them on repeat. They get spooked, and they walk. I've seen close rates drop twenty to thirty percent at those late stages.
"Late-stage sales is exactly when a human should be doing more, not less. Automate the early graft. Once a salesperson has built the trust, don't hand the final stretch to a bot chasing a signature."
People will argue the drop is just because they're in front of more buyers now. I don't buy it. The sales team is doing the same quality job they always did. The close rate falls because the late stage, the part that most needs a human, got automated. Early stages, automate away. Once the trust is built, real people should be doing much more of it.
So don't send on the rep's behalf. Draft the follow-up for them to review and send if they agree it's the right moment and the right words, or prompt them with the question they should be asking. And mind the timing traps: a salesperson rings a client, hasn't logged the note yet, and the CRM decides "we haven't contacted them" and fires off an email. Now your buyer's just had a warm call and an AI-sounding chase in the same hour, and they're wondering who they're even dealing with.
Why is "set and forget" the trap?
Because nothing about your business holds still. Audiences grow, data changes, a tool updates, an edge case you never pictured turns up. An automation that was perfect on launch day drifts, and because it's automated, it drifts silently and at scale. "Set and forget" quietly becomes "set and regret" precisely because no one's looking.
What are your minimum guardrails before an automation goes live?
Test on a small segment first, always. Before anything goes live, we agree a minimum segment and run a proper QA round, often with a sped-up timeline so waits that would normally be hours or days compress to minutes. We watch it over a couple of days to confirm it's roughly behaving, then set the real timings back.
Then the rollout is staged, never straight to everyone:
- Start tiny. Go live to one to five percent of the segment and check it daily to be sure it's actually working.
- Then widen. Once it's proven, take it to about twenty percent. That's where a few gotchas usually surface.
- Then full. Only after that do you roll to a hundred percent.
The shape is: tiniest fraction, then a fifth of the audience, then everyone. There are exceptions, of course. If your audience is genuinely tiny or the segment is brand new, testing on two people tells you nothing, so you might start at twenty or fifty percent and climb from there. But the principle holds: prove it on a small group before you let it loose.
How do you design automations you can actually trust?
Put a ceiling on everything, keep a human on anything that touches a high-stakes moment, and roll out in stages so mistakes stay small. Caps and end dates on spend. Humans in the loop late-stage. A staged rollout with daily checks. And a standing habit of going back to look, because the whole danger of automation is that it runs perfectly happily while doing the wrong thing.
Building automations that stay trustworthy is a big part of our CRM and automation and analytics work. If you've got automations running that nobody's checked in a while, request a quote and we'll audit them with you.
Frequently asked questions
How often should we audit our automations?
Regularly, and on a schedule rather than when something breaks. Anything touching spend or customer comms deserves at least a monthly look, which is why we roll ad end dates forward month by month. The point is to never let something run for months without a human laying eyes on it.
What always needs a human check?
Anything high-stakes or late-stage, especially the final steps of a big deal. Automate the early, repetitive graft, but have a person review and send late-stage proposals and chases rather than firing them automatically. The trust a salesperson built is too valuable to hand to a bot.
How do we stop double-sends and spammy chases?
Caps, timing rules, and a CRM check before anything fires. Guard against the classic trap where a rep has just called a client but not logged it yet, so the system assumes no contact and emails them anyway. A human-in-the-loop step on customer-facing messages prevents nearly all of it.
What are the signs an automation has gone rogue?
Spend climbing with no matching results, an audience that's quietly ballooned, close rates dropping at a specific stage, or customers reacting oddly to messages you forgot were going out. If a metric moves and no one can explain it, check whether an automation is running longer or wider than you intended.
