Account based marketing automation: what to automate, and what not to
Every ABM platform on the market automates roughly the same four things: finding accounts that match your profile, personalizing content at scale, tracking engagement and intent, and reporting on results. We cover what that looks like in practice in our article on ABM software. This article is about a different question, one that matters more the further automation goes: which parts of an ABM programme should be automated, and which parts shouldn’t, regardless of whether the technology to automate them exists.
That question has gotten more pressing in the last year, not less. As AI has gotten better at writing convincing emails and holding a conversation, the temptation has shifted from “automate the busywork” to “automate the judgment calls too.” We think that’s the wrong direction, and not just as a matter of preference.
What the rest of the market promises
Look at how ABM and marketing automation vendors describe their own products, and a pattern shows up fast. Pardot advertises a platform you can use to “fully automate your campaigns, including lead gen.” RollWorks pitches automated email sequences as a way to increase your connection rate. HubSpot and Demandbase both build automated workflows and predictive account prioritization directly into their core pitch, deciding, not just surfacing, which accounts deserve attention next. And the newest category, AI sales agents marketed under names like “the AI BDR you hire,” promises to run prospecting, personalization, and meeting-booking with no person required at any step.
The broader content on this topic follows the same instinct from a different angle. A whole genre of ABM guides frames automation and account based marketing as two separate ingredients you combine, automation for scale and tracking, ABM for personalization, use both. That framing isn’t wrong, but it treats the amount of automation as a dial you turn up until you run out of budget, never a boundary you deliberately don’t cross. One guide even names the actual failure point without meaning to: the middle of the classic three-tier model is where most teams stall, too many accounts to write custom emails by hand, too few to justify automating them. That’s not an argument for more automation. It’s a symptom of a model that assumes you either write everything by hand or hand it all to a machine, with nothing sensible in between, which is exactly the gap 1:cluster is built to close.
What’s consistently missing from all of it is a line. Not one of these products publicly says “don’t automate this part.” The default message in this category is more automation, less human effort, better results, with no caveat attached. That absence is exactly why the correction described further down in this article landed as a surprise to a market that had been told, uniformly, that the answer was always more.
The four things worth automating
Some automation is close to free, in the sense that it removes real busywork without touching a decision that matters. Finding companies that resemble your best customers by comparing websites and filtering on company data is a search problem, not a judgment call, automating it just saves a researcher days of manual list-building. Pulling ad performance data every night instead of exporting it by hand is the same kind of automation, it doesn’t decide anything, it just removes typing. Logging a touchpoint the moment someone scans a QR code or clicks an ad is automation doing exactly what it should: capturing something that happened, accurately, without anyone needing to remember to write it down.
The pattern across all of these: automation is safe when it’s collecting or surfacing information, and starts getting risky the moment it’s deciding what happens next.
Where the line actually sits
Inside Sqrl’s own software, there’s a consistent principle running through every stage of the process: the AI does the heavy lifting, a person makes the final call. Not as a philosophical stance bolted on afterward, it’s a design decision made at three specific points where it would have been technically easy to go further.
Finding target accounts. An AI agent compares your best customers’ websites against a much larger set of companies and returns a list of fifty to two hundred look-alikes, filtered against company registry and LinkedIn data to remove the obviously too-small or too-large. What it doesn’t do is decide which of those companies belong on your actual Target Account List. That judgment, Ideal fit, Potential fit, Influencer, Low value, or Irrelevant, requires knowledge the AI simply doesn’t have access to, like the fact that a company on the list burned your team on a deal three years ago.
Assigning buying groups. Once an account shows real intent, the software imports everyone at that company from LinkedIn and an AI agent reads every profile to sort people into buying groups. That’s a real narrowing step, sometimes three hundred people down to twenty, but the final trim, twenty down to a core group of five or ten worth engaging directly, is a human call. The reasoning behind who actually matters in a specific deal doesn’t reduce cleanly to a profile-matching algorithm.
Promoting an account to focus. The move from target account to focus account, the point where nurturing actually starts, stays deliberately manual. It could be automated on signal thresholds alone. It isn’t, because the decision benefits from things a threshold can’t capture: a good conversation at a trade show that never generated a digital touchpoint, a sense that a company is not actually in-market this quarter despite ticking the engagement boxes. Full detail on how that decision gets made is in our article on sales and marketing alignment.
One click, not an automatic send
The same principle shows up in how outreach itself works. When a step in a campaign is assigned to sales, whether that’s a LinkedIn connection request, a personal letter, or a phone call, the software gets you to the right person’s profile or contact details in one click. It doesn’t send anything on its own. The salesperson still clicks connect, still writes the message, still makes the call. That’s a deliberate line, not a missing feature. The claim is “one click to the right person,” never “automatic invites sent.”
This isn’t just a philosophical preference for the human touch, though it is partly that. It’s also a practical one. LinkedIn and email providers increasingly penalize accounts that send at automated volume, and a message a recipient can tell was written and sent by a machine tends to get worse results than a shorter, worse-written message that’s obviously from a person. In a model built around twenty to a hundred and fifty accounts, you’re never sending at the volume that would make automated sending worthwhile in the first place. The accounts are too few, and each one is worth too much, to treat any of them as a number in a sequence.
The market already ran this experiment
This isn’t a hypothetical concern. Over the past year, a wave of AI sales agents promised to replace human outreach entirely, prospecting, personalizing, and booking meetings with no person in the loop. The results coming back are consistent: fully autonomous outreach gets more replies, but converts a meaningfully smaller share of those replies into real opportunities than a human-run process does, and multiple independent reviews of the category now conclude that hybrid models, AI doing the groundwork with a person still deciding and sending, outperform full autonomy. The industry is quietly re-learning a version of the same principle Sqrl built in from the start.
What we don’t bother automating, on purpose
Not every automation decision is about risk. Some of it is just proportion. A CRM integration that would push qualified leads straight into a client’s sales system is technically straightforward to build, but for a programme measured in a handful of contacts and a handful of new clients a year, not hundreds, manual handoff isn’t a burden that justifies the engineering. At enterprise scale, where a single account list might run into the thousands, automating that handoff is close to mandatory, the volume makes manual work impossible. At Sqrl’s scale, entering a new contact into a CRM by hand takes a few minutes a week, not a bottleneck worth solving in advance of it ever becoming one.
The same logic applies to touchpoint logging. Yes, a colleague occasionally forgets to log a good conversation at a trade show. That’s a real, acknowledged gap, not a reason to automate around human input entirely, because the review process built around that input, not the logging itself, is where the actual value sits.
The short version
Automate the parts of account based marketing that are genuinely mechanical: finding candidate accounts, pulling performance data, logging what happened. Leave the parts that require judgment, who counts as a good account, who actually matters inside a buying committee, when an account is really ready, and whether this specific message should go out right now, in the hands of the person who has the context an algorithm doesn’t. That’s not a compromise made for a small team without the budget for more automation. It’s a design choice, and increasingly, it looks like the correct one.
