Measure AI gains across the marketing workflow
An AI tool can produce email variants or retrieve account intelligence in seconds while the organization’s revenue workflow stays unchanged. The marketer finishes sooner, but the value depends on what happens next: who receives the output, whether they can use it, and whether it improves work shared across departments. So marketing leaders need to measure changes across that workflow before counting faster individual production as an operational gain.
The gap between tactical experimentation and enterprise adoption framed the November 2026 MarTech Conference session, “Marketing automation is dead. Long live marketing automation!” Frans Riemersma, CEO of MartechTribe, moderated the discussion about turning individual gains into coordinated operations with clear responsibilities and measurable results. As the leader of a marketing-technology business, Riemersma has a commercial interest in organizations continuing to invest in this field. The participants brought their own business and operating interests to their recommendations.
Shared workflows are the unit of improvement
Some of those individual gains come from tools already inside marketing platforms. Milton Hwang, principal at Mission MarTech LLC, and Morgan Lehmann, VP of global demand generation at Iterable, described embedded agents: AI tools that perform tasks within an existing platform, bringing account research into automation hubs and improving delivery timing. Hwang has a commercial interest in demand for marketing-technology expertise, while Iterable benefits when organizations invest in marketing automation. Their examples show how assistance within an existing task can reduce the wait for information or action.
Yet a shorter wait establishes only the first part of the benefit. Hwang used the example of saving 20 minutes on an ad-hoc activity to distinguish personal efficiency from improvement across the full pipeline. The wider effect depends on how the task connects to later decisions and actions, so he recommended designing the workflow before counting faster completion as an operational gain.
Hwang’s prospect-dossier example shows what that design requires. An automated system can assemble research about a potential customer, but sales colleagues need to know where to find the dossier and what to do with it. Generating the document completes one activity; an agreed handoff makes the result usable. Without that agreement, marketing can record a completed task while the intended recipient keeps working without the information.
The same handoff problem extends to customer success and other shared processes. The panel called for explicit agreement with sales and customer success about the outputs they receive and how to use them. Coordinating the processes behind those exchanges gives departments a common understanding of where work goes and who acts next. Otherwise, teams can consider their own activities complete before the shared work has achieved its purpose.
Individual experiments can make coordination harder when colleagues independently solve similar problems. Eric Dates, director of growth marketing at Luxury Presence, described organizations paying multiple times for the same underlying solution because their AI experiments were uncoordinated. As a growth-marketing leader, he has an operating interest in making marketing work and spending more effective. He recommended a centralized, collaborative working environment where colleagues can reuse existing work before repeating the effort or buying another tool for the same task.
Reusing that work also requires access to the tools involved. Lehmann advocated equitable access so teams can build common practices, a recommendation that also aligns with Iterable’s commercial interest in broader automation use. Employees without access may struggle to adopt a colleague’s successful experiment, while sales or customer success may later receive an unfamiliar output with no clear location or instructions. Access therefore needs to come with shared choices about tools and explicit responsibilities for using their results.
A project in mind?
Schedule a 30-minute meeting with us.
Senior experts helping you move faster across product, engineering, cloud & AI.
Match oversight to the work automation performs
Once teams share an automated workflow, they also share the consequences of its actions. Moni Oloyede, founder of MO MarTech, described background activities such as data hygiene, which keeps records accurate and usable, and pipeline maintenance as carrying comparatively manageable risk. Her business has a commercial interest in organizations seeking marketing-technology expertise. She argued that autonomous customer-facing execution requires stronger safeguards because the system acts directly in an external interaction.
The consequences of an action should therefore determine its oversight. Oloyede emphasized that the operator remains accountable for the result regardless of the model’s sophistication. A more capable system can perform more work, but responsibility stays with the person operating it. Workflow design must therefore specify which actions the system may take and which safeguards apply.
Permission to act also creates a spending responsibility as automation runs more often. The panel identified the need to monitor API token usage: consumption measured in units of text processed through a model’s application programming interface, the connection software uses to exchange requests and results. These variable consumption costs sit alongside conventional software-seat costs, with seats measuring access and consumption measuring activity. A workflow used repeatedly across teams can therefore change spending even when the number of licensed users stays the same.
Managing that activity also means watching the platform’s operating conditions. Current defaults, vendor policies, experimental capabilities and data-retention rules all affect how automation behaves and how information is handled. Buying the platform starts an ongoing responsibility to manage those conditions. Execution risk, operator accountability, consumption spending and data controls each need a clear owner, even when they concern the same platform.
Buyer attention and trust limit useful automation
Even a coordinated, well-governed workflow can produce more material than buyers want to receive. Lehmann warned that faster production can tempt teams to fill inboxes with additional messages. Another asset is easy to count, but a buyer’s time and interest determine whether it has value. Measuring success by output volume can therefore reward activity that exceeds the audience’s capacity to absorb it.
Because that capacity is limited, Lehmann recommended prioritizing useful information and communicating when buyers are receptive. Her warning places a practical limit on automation use even as Iterable benefits commercially from adoption. Leaders still need to decide whether another message serves the recipient. Faster production expands their choices without making every additional contact worthwhile.
A buyer’s willingness to commit places a further limit on what automation can accomplish. Oloyede described a buying environment that increasingly includes autonomous research agents, or AI systems that gather information on a buyer’s behalf. She also emphasized that people still seek verified peer validation and authentic human reassurance before signing a contract. Automated research can help gather information, while marketing operations must preserve the human interactions that help buyers reach a decision they trust.
Test one workflow before expanding
These operating and buyer constraints give leaders a practical boundary for adoption: test a specific change before extending it. The panel set out four steps covering problem selection, platform review, controlled testing and departmental coordination. Each addresses a different way a useful task-level improvement can fail to improve the work around it.
First, select one narrowly defined workflow and identify a single recurring bottleneck within it. Establish a cycle-time baseline, the time the work currently takes, before introducing the tool, then measure the change in the handoff to the next participant. Measuring both keeps the test tied to a recurring problem and shows whether faster completion helps the shared process move forward.
Second, inspect the platform’s current settings and the vendor’s governance policies. The panel specifically recommended checking whether experimental capabilities are enabled and requiring clear data-retention rules before allowing them to operate. This review establishes the conditions under which the chosen workflow will run. Even when a capability is available, leaders remain responsible for deciding whether its behavior and data handling fit the task.
Third, test automated optimizations with hold-out control groups: groups kept outside the change so their results can be compared with those receiving it. The panel’s example was predictive send timing, in which automation selects when to deliver a message. Run that controlled experiment before extending the optimization throughout the customer lifecycle. Comparing the groups gives teams a basis for judging whether the timing change improves performance.
Fourth, agree directly with sales and customer success on where automated outputs will be stored and how each team should apply them. As Hwang’s prospect dossier illustrates, the recipient needs to be able to find the output and understand the next action. Agree on those responsibilities as part of the workflow so that producing the information leads to someone using it.
Key highlights
- Measure shared workflows: Marketing operations owners need cycle-time baselines and clear handoffs to sales and customer success. Agree on where automated outputs will live, who will use them and what action follows.
- Assign oversight and cost ownership: Platform owners need safeguards matched to each automated action, with stronger controls for customer-facing execution. Assign responsibility for API consumption, platform settings, vendor policies and data retention.
- Protect buyer attention and trust: Marketing teams need to judge automated messages by their usefulness and timing. Preserve access to peer validation and human reassurance as buyers move toward a purchase.
- Test a narrow workflow first: Marketing operations owners can start with one recurring bottleneck, review platform controls and compare results against a hold-out group. Measure the handoff to the next department before expanding the automation.
A project in mind?
Schedule a 30-minute meeting with us.
Senior experts helping you move faster across product, engineering, cloud & AI.


