Most conversations about AI in marketing end in the same unhelpful place: “use it, but keep a human involved.” It’s true, and it tells you nothing about what to do on Monday morning.
The trouble with advice that vague is that teams fill the gap in opposite directions. Some over-control, with people approving every bid change and reviewing every creative variation, which throws away the speed that automation exists to provide. Others under-control, switching the system on and trusting it, then discovering months later that nobody can explain where the budget went or why. Both teams would say they’ve kept a human involved. Neither has a clear answer to the question that matters: which decisions belong to whom?
So here’s something more usable. Think of your marketing decisions the way carmakers think about driving: not as a choice between a person and a machine, but as a set of lanes, each with a different split of who does what. Some decisions should be left to the machine. Some should be shared. A few should never leave human hands. In driving, nobody asks whether a car should be controlled by a person or a computer in general. They ask which tasks, in which conditions, with who ready to take over. Marketing deserves the same precision.
In this post we’ll set out a simple three-lane model, sort thirteen common marketing decisions into it, and give you four questions for placing any decision we haven’t covered. We’ll finish with a checklist you can run against your own team this week.
First, why the question is urgent.
Two Things Are Changing at Once
The customer is changing. AI agents are starting to shop on behalf of people: a customer asks an assistant to find and order a product within a budget, and the assistant compares prices, reviews and delivery options without the customer visiting a website or seeing an advertisement. Adobe projects traffic to US retail sites from AI services will rise 130% this holiday season. We’ve covered the mechanics in our guide to agentic commerce.
The ad platforms are changing too. Google’s Performance Max and Meta’s Advantage+ now handle audience selection, bidding, placements and creative testing. In last year’s holiday quarter, Tinuiti found Performance Max drove 62% of retailer spend on Google Shopping ads, and Advantage+ accounted for 27% of retail spend on Meta.
When a platform decides where most of your budget goes, the dashboard shows you results but rarely the reasons. That’s the gap human judgement has to fill, especially since a platform’s interests and an advertiser’s don’t always match perfectly.
The Three-Lane Model
This is our suggested way to organise the decision, not an industry standard. Sort each marketing task into one of three lanes. The lanes describe who decides, not how clever the technology is, so the same tool can sit in different lanes for different jobs.
Lane 1: Let it drive. The AI decides and acts. You set the limits and review results afterwards. This lane suits work that is repetitive, measurable and time-sensitive.
- What you control: the guardrails, such as budget caps, brand rules and the goals the system optimises toward.
- A typical example: automated bidding that adjusts within a daily budget you’ve set.
- The risk to watch: drift. A system left alone can slowly optimise toward something you didn’t intend, which is why the review step matters.
Lane 2: Co-pilot. The AI proposes, a person approves. This suits decisions that are consequential or hard to undo, but where the machine’s analysis is still valuable.
- What you control: the final yes or no, with the AI’s reasoning in front of you.
- A typical example: an assistant drafts a new campaign from a typed instruction, and a person checks the targeting, budget and copy before it goes live.
- The risk to watch: rubber-stamping. If approvals become automatic clicks, the lane quietly turns into Lane 1 without anyone deciding that it should.
Lane 3: Hands on the wheel. A person decides, and AI may inform. This is for anything involving identity, values, ethics or empathy.
- What you control: the decision itself. AI can supply data, options and drafts, but it doesn’t choose.
- A typical example: deciding whether to pause a campaign that’s performing well while the news cycle makes it look insensitive.
- The risk to watch: letting convenience creep in. It’s tempting to accept a system’s suggestion on a values question because it arrived with confident numbers attached.
Two points are worth keeping in mind as you use the model. First, the lanes are about decisions, not departments. One campaign will contain Lane 1 work (bidding), Lane 2 work (launch approval) and Lane 3 work (what the brand is willing to say). Second, a decision should only move to a lane with more machine control once there’s a record showing the system handled it reliably, which is what the regular review gives you.
Which Decision Goes in Which Lane
Here’s how thirteen common marketing decisions sort into the three lanes.
Lane 1: Let it drive
- Bidding and budget adjustments within set limits: fast, repetitive and measurable.
- Audience discovery: machines spot patterns in large data sets that people would miss.
- Creative testing: many variations can be tried and measured quickly.
- Product data and feed upkeep: accuracy at scale, which also helps AI shopping assistants find you.
- Personalised recommendations: relevance for each customer, within brand rules.
Lane 2: Co-pilot
- Launching a new campaign: Meta is testing an assistant that builds campaigns from typed instructions, with changes staged for approval, which is exactly this model.
- Large budget shifts or new targeting: consequential, and worth a second look before they go live.
- New claims in ad copy: accuracy and compliance need a person’s eye.
Lane 3: Hands on the wheel
- Brand identity, values and positioning: a system can tell you what’s getting clicks this week, but not who you want to be.
- Defining what success means: tell a system to maximise conversions and it will, whatever the cost.
- Ethical boundaries and customer data: someone has to own what’s acceptable and protect personal information.
- Verifying campaign results: platforms mark their own homework, so check with your own analytics.
- Pausing campaigns in a sensitive moment: a campaign can perform well on paper and still be the wrong thing to run.
Four Questions to Choose the Lane
When a decision isn’t on these lists, ask four things.
- Is it repetitive and measurable? If yes, it leans toward Lane 1.
- Can it be undone quickly? If a mistake is costly or hard to reverse, move it toward Lane 2 or 3.
- Does it involve people’s trust, privacy or feelings? If yes, it belongs in Lane 3.
- Could the platform’s interest differ from mine? If yes, add an independent check whatever the lane.
The Same Pattern Is Appearing Everywhere
It’s worth noticing that other industries are landing on the same shape. Walmart’s tools for AI shopping agents, announced on October 21, let shoppers approve each purchase or set a spending limit, and track each agent’s orders separately. In banking, reports on October 22 said major US banks had detected unauthorised AI agents inside customer accounts, mostly logging in with customers’ own credentials, and were moving to restrict them. In both cases the machine acts, but only inside limits a person set. That’s Lane 1 with guardrails, and it’s the model to copy.
What Good Looks Like in Practice
A well-run setup has a few visible habits. None of them is complicated, but each one is easy to skip when the results look good.
Written limits. Everyone knows which decisions the AI makes alone, which need approval and which are off-limits. Put it on one page, using the three lanes, and include the numbers: the budget cap, the maximum daily change and the claims that never appear in ad copy. Limits that live only in someone’s head disappear when that person goes on holiday.
Quality inputs. Accurate data, clean product feeds and strong creative assets make every automated system perform better. An automated system can only optimise what you give it, so out-of-stock items that keep being advertised, vague product titles or a thin set of creative options will cost you money quickly. Treat the inputs as a shared responsibility, not a one-off set-up task.
A scheduled review. Someone checks where automated budgets are going and why, on a regular cadence. Pick a day, such as the first working day of each week, and ask the same few questions each time. Where did the money go? What changed since last time? Do the platform’s numbers agree with your own analytics? A routine review turns a dashboard you glance at into one you actually understand.
An owner with a brake. One named person has the authority to pause a campaign at any time. Naming the person matters as much as granting the authority. In a sensitive moment, a team that has to ask “who decides?” has already lost time. Make sure the owner has a deputy, knows how to pause every platform you use and doesn’t need sign-off to do it.
It also helps to agree in advance what counts as a reason to pull the brake: a news event that makes an ad look insensitive, a sudden spike in spend with no matching rise in results, or a customer-data concern. Writing those triggers down means the decision is made calmly, before the pressure arrives.
This is the kind of structure we set up through Kilowott Intelligence, and Lane 1 tasks are often the easiest place to begin with automation. Start with one platform and one set of limits, see how the review habit settles in, and extend the same pattern to the rest of your marketing once the team trusts it.
A Lane Audit Checklist
- List the ten marketing decisions your AI tools currently make or influence
- Assign each one to Lane 1, 2 or 3
- Set limits for every Lane 1 decision, such as budget caps and brand rules
- Add an approval step to every Lane 2 decision
- Confirm that results are verified against your own analytics
- Name the person who can pause campaigns, and tell the team
- Review the lanes every quarter, since tools and risks change
Frequently Asked Questions
How much of marketing should be handled by AI? There’s no single answer. A useful approach is to sort decisions into lanes: let AI act alone on repetitive, measurable tasks, have it propose and a person approve for consequential ones, and keep identity, values and ethics in human hands.
Why can’t I just trust the platform’s automation? Automation can improve efficiency and reach, but it reduces visibility into where budgets go, and a platform’s goals don’t always match an advertiser’s. Independent verification keeps you in control.
What changes when customers use AI shopping agents? Your products need to be easy for AI systems to understand, compare and recommend, with accurate names, prices, stock, delivery details and return policies.
The Machine Drives, the Human Steers
AI will keep taking on more of the work of advertising, shopping and media planning, and the lanes will shift as the tools improve. What shouldn’t shift is who decides where the brand is going. Machines can optimise the journey, but people must choose the destination.
That’s worth remembering because the lanes aren’t permanent. A task that sits in Lane 2 today, such as launching a new campaign, may move to Lane 1 once a tool has earned trust over several months of reliable results. A task you assumed was safe in Lane 1 may need to move back if a platform changes how it works or its interests drift away from yours. The model isn’t a verdict on what AI can do. It’s a way of deciding, deliberately and on a regular schedule, how much you’re willing to hand over right now.
Trust, in other words, should be earned in stages. The brands that handle this well tend to start small: one platform, one set of limits, one person with the brake. They watch what happens, widen the machine’s lane where the evidence supports it and keep a firm hold on the decisions that define who they are. Handing over everything at once is risky. Refusing to hand over anything means working more slowly than competitors who’ve learned to use the tools well.
Where to Start This Week
If you only do three things after reading this, make them these.
- Draw your lanes. List the ten decisions your AI tools currently make or influence and sort each one into Let it drive, Co-pilot or Hands on the wheel.
- Name your owner. Decide who has the authority to pause campaigns, make sure they know how, and tell the team.
- Book the first review. Put a recurring slot in the calendar to check where automated budgets are going and whether the platform’s numbers match your own.
None of these needs a new tool or a large budget. They need a short conversation and a decision about who’s responsible for what, and they put you in a stronger position than switching automation on and hoping for the best.
If you’d like help drawing the lanes for your own marketing, our digital marketing team can walk through them with you, and our case studies show how we’ve approached it with other brands. You can also explore how we set up approval steps and automated reporting through Kilowott Intelligence and automation, so the machine does the heavy lifting and your team keeps the wheel.
