A marketing team at a mid-sized software company finally gets access to Marketing Cloud Next they have been waiting for. The AI can find audiences, draft content, improve journeys, and personalize experiences at scale, all while taking manual work off the team’s plate. The possibilities feel enormous, so they flip the switch.
Two weeks later, things start to unravel. One agent recommends a steep discount to customers who just upgraded at full price. Another writes subject lines that sound nothing like the company. A third adjusts campaigns using customer data it was never supposed to access.
The team scrambles. Marketing tries to pause the affected journeys. Sales wants to know which customers saw the discount. Compliance starts asking who approved the data access. No one has a clear answer.
Now everyone is looking for someone – or something – to blame. Did we choose the wrong platform? Is the AI broken? No. The software did what it was allowed to do.
The team never agreed on how the AI should operate before turning it loose. No one set clear boundaries, decided who could approve what, or defined which data the agent could use. They switched on the capabilities and trusted the details to sort themselves out.
That story is more common than you might think. Teams get caught up in what AI can do and postpone the governance conversation until the first problem lands in their lap.
That is the difference between turning on an AI feature and being ready to use it. The marketing teams getting real value from these agents answer the uncomfortable questions first: How should the agent behave? What information can it see? Which actions need a human sign-off? Who is responsible when something goes wrong? Only then do they decide how the technology should work.
AI Changes the Way Marketing Automation Works
Traditional marketing automation is fairly predictable. You map a workflow, build a segment, create the campaign, connect the systems, and set up reporting. Once the pieces are in place, the platform follows the instructions you gave it.
AI changes that relationship. An agent can spot patterns, recommend a next move, generate content, and adjust as conditions change. You are no longer programming every step in advance.
While that is exciting, it also means “set it and forget it” is no longer a safe operating model.
The campaign build is no longer the hardest part. The tougher work is deciding how much freedom the agent should have. What should cause it to act? When should a person step in? Which data is fair game? When should the agent stop and ask for help?
These are business decisions about judgment, risk, and accountability. Marketing leaders need to make them with input from the people who understand the technology, data, security, and compliance.
Skip those decisions, and the problems show up quickly: erratic campaign choices, compliance concerns, frustrated employees, and no clear answer when someone asks who owns the outcome.
Seven Decisions to Make Before Activating Any AI Agent
1. Define What You Want the AI to Do
It sounds obvious, yet this step is often skipped.
Sometimes a company enables AI because the feature is available, a competitor is talking about it, or an executive saw a polished demo. Any of those can spark a useful conversation, but none tells the team what business problem the agent should solve.
Begin with one outcome you can name. Maybe you want stronger campaign performance, faster production, better personalization, or improved retention. Choose the priority before you choose the agent’s job.
If the goal stays vague, success will stay vague too. Your team will struggle to tell whether the agent helped, stakeholders will interpret results differently, and the investment will become harder to defend.
2. Decide Which Customer Data the Agent Can Access
An AI agent needs useful data, but handing it every piece of customer information you own will not automatically produce better decisions.
Sit down with the people who understand your customers, systems, security requirements, and compliance obligations. Decide what the agent needs for the job and what remains off-limits. That conversation holds importance everywhere, but especially in regulated industries or whenever sensitive information is involved.
Then write the rules down, share them with everyone involved, and enforce them the same way every time.
3. Determine What Level of Autonomy Makes Sense
Giving an agent a job does not mean giving it complete control.
Your team may want AI to make recommendations while people stay responsible for execution. Or you may be comfortable allowing certain agents to act on their own within narrow boundaries. Either approach can work. What matters is that the level of freedom is a conscious choice.
For many organizations, the right answer is a mix. An agent might identify segments, improve send times, or suggest content variations without waiting for approval. Changing journey logic or offer strategy may still require a person to sign off.
Draw that line before launch. You do not want the first real test of the agent’s authority to be a mistake in your customer communication.
4. Identify Which Actions Require Human Approval
Speed is useful. Unchecked speed is not.
Strong marketing teams decide where AI can move quickly and where human judgment is still required. Ask plainly: Which decisions must someone review? When is approval required? Who gives it? What level of risk should cause the agent to stop?
Clear answers tend to make adoption easier, not slower. People are more willing to use an AI tool when they understand the guardrails and know a reliable review process is in place.
5. Set Up Clear Escalation Rules
Sooner or later, the agent will run into a situation it cannot handle with confidence.
Plan for that moment. Decide when the agent should hand a decision to a person – for example, when confidence falls below a set threshold, an unusual business exception appears, a compliance concern surfaces, or customer behavior breaks the expected pattern.
A good escalation path protects the customer experience and gives the agent a safe way to say, in effect, “I need help with this one.”
6. Establish Metrics That Are Meaningful to Your Business
Before launch, decide what evidence would convince you that the agent is helping.
AI initiatives habitually begin with enthusiasm and no agreed measure of success. Months later, your team has activity to report but no clear way to show whether the investment paid off or where it needs to improve.
Tie the measures to the outcome you chose in the first decision. For campaign performance, look at conversion, engagement, and revenue contribution. For production efficiency, measure time saved. For personalization, track how customers respond.
Choose the measures early, make them specific, and return to them often. Otherwise, the goalposts will move every time someone asks whether the program is working.
7. Assign Clear Ownership
This decision will make or break the whole effort.
Someone must watch performance, review outcomes, update the rules as the team learns, and decide when permissions should expand or tighten.
That can get messy because marketing AI crosses so many lines: marketing, Salesforce administration, data, architecture, security, and compliance. If ownership is fuzzy, problems sit unresolved and improvements stall. Everyone is involved, yet no one is truly accountable.
Give agent oversight a named owner from day one. That person does not have to make every decision, but they should be responsible for bringing the right people together, keeping the rules current, reviewing performance, and making sure problems do not fall between teams.
How AI Decisions Shape Your Salesforce Environment
These choices affect how your entire Salesforce environment should be designed and governed.
Marketing Cloud Next does not operate on its own. It depends on Data 360 (formerly Data Cloud), Agentforce, Flow, and Salesforce Core to move information and enforce decisions. That means every governance choice has a technical consequence: data boundaries shape Data 360 architecture, approval rules shape workflows, permission choices affect security, and escalation paths often cross clouds and teams.
Work through these decisions during discovery and solution design, and you can build around the way your organization operates. Putting them off, however, means you’ll likely discover halfway through implementation that the solution needs to be reworked – usually at the most expensive possible time.
Get the Rules Right Before You Flip the Switch
A successful implementation begins long before anyone activates the first agent. It begins when the right people sit down together and agree on the rules of the road.
Bring business stakeholders, Salesforce specialists, data leaders, security and compliance partners, and governance owners into the conversation and work through the seven decisions together. Be sure to capture everyone’s answers, resolve any disagreements, and use the results to guide your design.
This preparation gives the implementation team something concrete to build from. They know the goal, the boundaries, who owns each decision, and where approvals belong. That clarity reduces rework and keeps the team from discovering critical gaps halfway through the rollout.
What It Comes Down To
Marketing Cloud Next can change how your team engages customers. But powerful capabilities do not automatically produce a strong AI program. Teams that skip the operating decisions often learn that lesson only after something goes wrong.
To see the strongest returns, your team must decide how its agents will work before those agents ever reach a customer. Agree on the boundaries, document them, and keep refining them as you learn.
Before you enable the next agent, gather the people who will own the outcome and make these seven decisions together. That conversation can save months of cleanup later.
Is Your Org Ready for Marketing AI Agents?
If you are not sure whether your team has answered these questions, Sikich’s Marketing Cloud Next AI Readiness Assessment is a practical place to start.
It can help you identify gaps, work through the decisions with the right stakeholders, and turn the answers into an implementation plan built for your environment.
