You can’t scroll very far on social media these days without seeing some bold claim that “AI will run your marketing for you.” It’s a catchy promise. It’s also a really quick way to burn budget.
We’re not anti-AI. We use automation every day across programmatic, search, and social. Some of the best performance jumps we’ve seen have come from smarter machine-driven bidding and pacing.
But here’s the pattern we see over and over when we’re inside real accounts. AI is incredible at buying impressions. It’s terrible at understanding your actual business.
The teams that win aren’t the ones who choose “AI or humans.” They’re the ones who understand where automation should be in charge, and where human judgment has to step in and say, “Nope, not like that.”
This is how we think about that balance.
What AI is genuinely good at
When you strip away the hype and look at what today’s platforms are actually doing, AI is genuinely great at a handful of very specific jobs.
It’s excellent at reacting to thousands of micro-signals at once: time of day, device, inventory type, audience combinations, bid landscapes that change minute by minute. A human trader simply can’t keep up with that level of constant recalculation, and honestly, they shouldn’t try. Let the machine sweat those details.
It’s great at pacing. If a campaign is behind at 10:17 a.m., an algorithm can quietly nudge bids and budgets long before anyone on the team has finished their first coffee. It doesn’t get distracted. It doesn’t have back-to-back meetings.
And when you feed it clean, well-structured options, AI is pretty good at basic creative rotation too. Give it three versions of a headline and a clear goal, and it will quickly figure out which one pulls more clicks or conversions.
In short, AI is a monster at the math. That’s where you want it. The problem starts when we ask it to do work it was never built to do.
Where automation quietly steers you into a ditch
Left to its own devices, an algorithm will cheerfully optimize to the wrong thing and tell you it’s doing a great job.
If the only signal it’s given is cheap, upper-funnel engagement, that’s what it will chase: clicks, views, low-cost traffic. Your dashboards might look fantastic while your sales team quietly wonders why none of these “perfect” campaigns are showing up as revenue.
AI also has zero context for your actual world. It doesn’t know your product margins. It doesn’t understand that your hero SKU is out of stock in half the country. It doesn’t care that you have a seasonal moment coming in three weeks and you cannot afford to waste this month’s budget on vanity metrics.
And while a machine can report that “Creative B has a 22% higher CTR,” it can’t tell you whether that creative is attracting the right people, whether it fits your brand, or whether the story it’s reinforcing is the one you want to tell.
One more problem, black-box bias. If you never look under the hood, algorithms often fall in love with certain sites, audience segments, or formats that look efficient but don’t line up with your real business goals. On paper, everything is “green.” In your P&L, not so much.
That’s why human craft isn’t a “nice to have.” It’s the thing that keeps the AI pointed at the right target.
A better framework. Not man vs machine, but a relay race
Inside AOD, we don’t treat AI like an opponent we’re supposed to outsmart. We also don’t hand it the keys and hope for the best.
We think of media ops as a relay race. The machine has legs we could never have. Humans have judgment the machine will never have. The game is to design the hand-offs so each one does the work it’s built for.
There are areas where we absolutely want automation in front:
- Managing bids and bid multipliers as conditions change
- Keeping campaigns on pace throughout the day
- Spotting weird anomalies fast (sudden CPC spikes, delivery crashes, odd conversion patterns)
- Rotating through creative variations at a volume no person could track manually
And there are areas we guard fiercely as human territory:
- Translating the messy business ask (“We need profitable subscribers, not just leads”) into the right KPI hierarchy and events
- Designing measurement: what we track, how we attribute, which signals we feed the machine first
- Shaping the creative and the story behind it
- Interpreting the “why” behind performance, not just the “what”
- Knowing when to break the glass and override the algorithm, even when the platform insists everything looks fine
When those lanes are clear, AI stops being a threat and turns into a force multiplier.
What this looks like in our day-to-day
Here’s how that balance actually plays out in our work, minus the buzzwords.
We start with the messy reality – your business model, your margins, your sales cycle, your offline activity, your creative assets, your internal politics. All the stuff that never fits cleanly in a campaign brief.
From there, our team designs the architecture of the account. That means the structure of campaigns and ad sets, the naming conventions (the boring part that saves everyone later), the UTM strategy, the way we’ll bucket audiences, the guardrails we’ll use for brand safety and reach.
Only after those foundations are in place do we flip the “smart” switches on: bid strategies, budget automation, dynamic exploration, creative rotation. The AI gets to explore, but inside the lines we drew on purpose.
Then the rhythm kicks in.
On a daily basis, we’re watching: Are we on pace? Are there placements we don’t want to be on? Is anything breaking? That’s quick-reaction work; small, tactical adjustments that keep things safe and sane.
Weekly, we zoom out and ask different questions: What’s actually driving performance? Which audiences are finding real value? Which messages are starting to fatigue? Are we pulling in the right customers, or just the cheapest to reach? So on and so forth.
Those answers don’t just live in a spreadsheet. We push them back to the people who can use them – brand leads, creative teams, agency partners. “This hook is doing the heavy lifting.” “This audience is expensive but worth it.” “This format seems to be where we’re winning new buyers, not just repeat visitors.”
That feedback loop is where the system gets smarter. The machine can learn faster, because the inputs keep getting cleaner and the human guardrails get more precise.
Questions worth asking any “AI-powered” partner
If you’re working with an internal team, an agency, or a specialist partner like us, you don’t need to be a data scientist to check if the balance is right.
A few simple questions go a long way:
- What exactly is the AI optimizing toward, and who decided that was the right north star?
- When was the last time you overrode the algorithm, and why?
- Who owns naming conventions and UTMs, and how do you keep them consistent?
- How do you review placements and audiences for brand and business fit, not just cost?
- How do learnings from one channel (CTV, for example) show up in another (like search or social)?
If the answers are vague or hand-wavy, that’s a red flag. You might be looking at a very expensive black box.
So where does this leave us?
AI isn’t coming for your job. It’s coming for your busywork.
The teams that come out ahead aren’t the ones that “automate everything.” They’re the ones that:
- Let automation handle the parts of media buying that are pure math
- Keep humans focused on strategy, context, creative, intelligent optimization, and the meaning behind the metrics
- Build feedback loops that connect platforms, people, and performance instead of treating each one as a separate universe
That’s how we run media at AlwaysOn Digital.
We’re happy to let AI assist with running campaigns. Our job is to make sure our experienced human hands are on the campaign levers, and winning the audience to create business outcomes that matter for our clients.
