KEY TAKEAWAYS
It’s rarely the tool: Automation failure is almost never due to the software itself; it is consistently caused by a lack of foundational groundwork.
-
Map the workflow first:
You cannot effectively automate a process that hasn’t been documented. Skipping the "boring" step of mapping your current human workflow is the most common reason for project failure. -
"Garbage In" Principle:
The AI output quality is directly tied to input data quality. If your data is messy or untrustworthy, no level of AI sophistication can produce reliable results. -
Subject matter expertise is required:
Technical skill alone is insufficient. You must pair AI specialists with team members who understand your specific marketing workflows to build automations that actually work. -
Iteration is expected:
Expect to refine your automated processes over several rounds. If it isn’t working after a few tries, it is a sign that the upstream brief or process documentation is missing critical detail.
A client gets excited about a new AI automation. A few weeks in, the results still aren’t there. When Somebody Digital’s Cristiano Winckler looks into what actually happened, the problem is rarely the tool.
“What is not working and why do you think it’s not working, that’s precisely the starting point,” he says. “Sometimes the automation is indeed working, but the data you’re feeding that automation or process with is incorrect, and then you get the wrong output.”
One recent example involved a company that had hired a third-party agency to build automations using tools like Make.com. The rules never made it into the workflow. “They forgot one fundamental element in this whole thing, which is you need to understand the actual workflow,” Winckler says. “They started going straight to the tool, trying to build connections and integrations between the various tools they could potentially utilize. They forgot to map out their own process first.”
John Wilkes, Somebody Digital’s Director of Strategy, puts it plainer still. The groundwork is boring, and skipping it is where most automation projects go wrong. “The amount of groundwork and grunt work and human work to map out the workflows, how are we doing this, how is the team doing it, how can we do it better. When that stuff is not done correctly, that heavy lifting up front, that is when you don’t get the output you expect.”
The data problem hiding underneath the workflow problem
Even a well-mapped workflow will fail if the data feeding it can’t be trusted. Winckler is direct about the math: “The quality of the output is directly related to the quality of the input. Garbage in, garbage out, in plain terms.”
He cites a recent audit: a Google Analytics account with 220 tracked events. “They should have five. That’s confusing. How do you know which ones are working? How will AI be able to analyze that data?”
Fixing analytics is unglamorous work with no immediate reward, and it’s the piece organizations neglect most, right up until a CFO asks where next year’s budget should go and nobody can answer with any confidence.
How many tries does it actually take?
Marketing leaders under pressure to show results fast often expect an automation to work the first time. Winckler resets that expectation early with clients: “Having iterations is normal. It’s like you’re talking to an intern. He’s not going to get it right the first time, period.”
Two or three rounds to refine a process is typical, in his experience. Six or seven is a signal that something upstream is missing, an unclear brief, a missing example of the desired output, or a step that was never properly documented. “The devil’s in the detail,” he says. “You really need to do a thorough job.”
He points to a recent website migration project as an example of what a properly staged process looks like. Rather than handing AI a folder of raw data and asking for a plan, his team broke the work into stages: consolidate the data first, check for gaps, then move step by step through analysis and recommendations. “We were able to do it in three weeks,” he says, of work that would usually take two to three months.
The pressure that makes this worse
Wilkes names the dynamic that a lot of marketing leaders are living with right now: board pressure to move fast, competitors who appear to be moving faster, and a pace of delivery that leaves no room to do the groundwork properly. “It’s the speed that’s often kind of getting them in trouble,” he says.
There’s also a quieter problem sitting underneath the pressure to look like everyone’s using AI. “The reality is we know that the teams are already using it, just not the approved version,” Wilkes says. “Everyone’s using Claude or ChatGPT. They’re all doing their thing.”
Winckler doesn’t see that as purely a risk. Experimentation matters, and Somebody Digital actively encourages its own team to try new tools, provided there are clear policies on what data can and can’t be shared. What separates teams that get value from AI from teams that don’t isn’t enthusiasm. It’s whether someone understands the workflow well enough to know what’s worth automating in the first place.
He offers a blunt example from his own hiring decisions: “We hired an AI transformation specialist. He’s an AI automation guy who does not have a digital marketing background. You get that guy to create automations for us, nothing will work, because he doesn’t understand the workflows.” The fix at Somebody Digital was pairing that specialist with the people who already understood the process, not replacing them.
Where to start
Winckler’s advice for any leader watching an automation underdeliver is to pick one task and check whether the process behind it was actually written down before AI ever touched it. Keyword research, he suggests, is usually the easiest place to start. Get on a call, ask the specialist to describe how they do it without naming a single tool, and record it.
“If you’re not getting the output that you want from your workflows, ask to see the documented processes,” he says. “It’s boring work, but it’s absolutely necessary. It’s the foundation of everything.”
It is rarely the fault of the AI tool itself. Projects typically fail because teams skip the “boring” groundwork—specifically, failing to properly map out the actual workflow or using untrustworthy, low-quality data to feed the process.
Two to three rounds are considered normal. If you are reaching six or seven iterations without success, it is a signal that there is a problem upstream, such as an unclear brief, missing examples of desired output, or a step that was never properly documented.
Data follows a “garbage in, garbage out” principle. If the input data is messy or irrelevant (like an analytics account tracking hundreds of unnecessary events), no amount of AI sophistication can generate reliable output. Quality output is directly tied to the quality of the input
Begin by documenting the process before involving any software. Pick one task, have a team member describe how they currently perform it—without mentioning any specific tools—and record that workflow. That documented process serves as the essential foundation for any successful automation.
Technical expertise is not enough; you need that technical skill paired with subject matter expertise. An AI specialist without a background in your specific field (like digital marketing) will often struggle to build effective automations because they lack an understanding of the underlying workflows.
