Mid-year review: a scorecard for your agentic marketing program
AI-assisted research. This note was prepared with the assistance of AI tools. It may contain errors and is not professional advice. See our Legal Disclosures.
Marketing teams spent the first half of the year moving agents from pilots into real workflows. June is when many organizations take stock, and for teams that invested in agents, leadership's question is predictable: is it working?
A good answer covers more than time saved. It shows value, quality, risk, and whether the program is ready to grow. This post sets out the scorecard we use for mid-year reviews, along with the rating rules that keep it honest.
Why a scorecard instead of a slide of wins
The usual mid-year update is a slide of highlights: a campaign that went out faster, a quote from a happy colleague, a chart that points up. Each item may be true, but together they answer a different question. They show that something went well, not whether the program is sound.
A scorecard forces coverage. Every section gets a rating, including the ones that would never appear on a highlights slide. That is where its value lies, since the weak sections are the ones leadership most needs to hear about before approving the second half.
How to rate each section
The scorecard has five sections, each with a few questions and one of three ratings: strong, developing, or weak. Ratings depend on the kind of evidence behind them, not on how confident the team feels. We suggest these rules for every section.
- Strong: measured evidence, collected consistently, covers most of the section's questions.
- Developing: specific observations or partial measurement, with a plan to close the gaps.
- Weak: mostly belief, or a known gap with no plan yet.
To reduce self-grading bias, have two people rate each section independently, ideally one from the agent team and one from outside it. Where their ratings differ, the evidence decides. If neither can point to a measure, the lower rating stands.
The five sections
The sections are value, quality, operations, risk, and readiness. We group quality and operations below because their questions overlap, but each still gets its own rating. Answer every question in writing, with a link to the evidence where it exists.
Value delivered
Value is the section leadership reads first. These questions test whether the claimed value is measured or assumed.
- Can you show measured time saved on specific workflows, not estimates?
- Has any agent contributed to an outcome metric, such as pipeline, conversion, or retention, with a credible comparison?
- Are people choosing to use the agents when they could skip them?
Strong programs have measured evidence for at least a few workflows. Developing programs have consistent observations. Weak programs have mostly enthusiasm. A credible comparison means a control group or holdout, as described in incrementality testing for agent-driven campaigns, rather than a before-and-after chart.
Quality and operations
Quality and operations sit together because both describe whether the program runs well day to day. These questions cover both.
- Do you score agent output with a consistent, calibrated rubric?
- Is human edit rate tracked, and is it falling for your main agents?
- Have quality problems been traced to causes and fixed?
- Is someone responsible for each agent's day-to-day performance?
- Are costs tracked per agent and per task, with a regular review rhythm?
Edit rate is often the quickest of these to get started on. Our post on measuring human edit rate describes a version that needs only drafts and finals saved side by side. Ownership is the operational question we would check first. We expect an agent without a named owner to degrade without anyone noticing.
Risk and governance
Risk is the easiest section to leave out of a mid-year update, and the one we expect to matter most in the second half. These questions test whether the program can grow safely.
- Is every agent's autonomy level documented, with the evidence that justified it?
- Are important rules enforced in code rather than only in instructions?
- Can you trace any customer-facing output back to the run that produced it?
- Have there been incidents, and were they reviewed and addressed?
A program can deliver value and still be weak here. That combination is the most dangerous, because success encourages expansion before the foundations are ready. Moving rules from instructions into checks, as described in guardrails as code, is usually the first fix. The second is a documented position for each agent on the autonomy ladder.
Readiness for the second half
The final section turns the review toward planning. It asks what the evidence supports doing next.
- Which agents are ready to take on more volume or autonomy, based on evidence?
- Which workflows are next, and why?
- What capabilities are missing, such as data quality, evaluation, or staffing?
Readiness ratings should follow from the other four sections. An agent with strong quality evidence and weak governance is not ready for more autonomy, however popular it is.
A worked example
Consider a hypothetical program with three agents: one for lifecycle email, one for paid social copy, and one for lead research. The team rates value as developing, since edit time is measured for email but not for the others. Quality is strong for email and weak for lead research, which has no rubric.
Suppose risk is rated weak, because two compliance rules live only in instructions. The honest summary is that the email agent is ready to scale and the other two need instrumentation. The program as a whole needs a governance fix before any agent gains autonomy.
That summary is less exciting than a highlights slide. It is also far easier to fund, because it says what the money will buy.
Presenting the results
Leaders tend to respond well to candor. A scorecard showing strong value in two areas, developing quality practices, and a clear plan for governance gaps is more credible than one claiming success everywhere.
Pair each weak rating with one specific action, an owner, and a date in the second half. Then show the evidence for each strong rating in an appendix, so anyone who wants to check it can.
From scorecard to plan
The scorecard feeds directly into second-half planning. Strong areas are candidates for scaling, and weak areas need investment before expansion. We describe how to turn ratings into a plan in second-half planning.
We are now studying how ratings should change as programs mature. A rating of strong in June may mean something different by December, once evaluation and governance practices are in place across every agent.
How to apply this: have two people rate your program on the five sections independently, using evidence rules agreed in advance. Bring the ratings, the evidence behind them, and one action per weak area to your mid-year review.
Related research
Cost per task: budgeting model spend for marketing agents
Model costs are easy to ignore during a pilot and hard to ignore at scale. Measuring cost per completed task gives finance and marketing a shared language.
Attribution in an agentic funnel: credit decisions, not just channels
When agents research, write, target, and follow up, the question shifts from which channel worked to which decisions worked. Attribution has to follow.
Shadow mode: run agents alongside humans before going live
The safest way to learn how an agent will perform on real work is to let it do the work in parallel, without its output reaching anyone. That is shadow mode.
Put this research to work
Marketix turns these patterns into software: agents, automated workflows, and a hub that keeps people in control. Sign up for early access.
