H2 planning: turning first-half experiments into second-half systems
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.
Agentic marketing spent the first half of the year in experiment mode for many teams, and the result is often a scattered portfolio. Some experiments are promising, some were abandoned, and a few are quietly running without much oversight.
Second-half planning is the moment to decide which of these become real systems, with owners, measurement, and governance, and which should stop. This post sets out a five-step method for making that call, along with what to measure as the second half unfolds.
Why experiments do not become systems on their own
An experiment is built to answer a question. A system is built to run every week without its creator watching. The two need different things, and the gap between them is mostly unglamorous work: documentation, monitoring, review routines, and a named person who cares when quality slips.
Without a deliberate decision, experiments drift. The ones that work get more use without gaining the structure that use requires. The ones that do not work keep consuming attention because no one formally ends them. Planning is the right moment to break that drift, because it is when the team already expects to make choices.
Step 1: sort the portfolio
Start from your mid-year review. If you used our mid-year scorecard, you already have ratings. Sort every agent and AI-assisted workflow into four groups.
- Systematize: measured value and stable performance, ready to become a standard part of operations.
- Strengthen: promising, but missing measurement, documentation, or governance.
- Rethink: a valuable goal, but the current approach is not working.
- Stop: no clear value and no clear path to it.
The stop group is the hardest to fill honestly. People grow attached to their experiments, and stopping one can feel like admitting a mistake. We recommend writing a short note for each stopped experiment: what it tried, what it showed, and what might be reused. That note turns a stop into a result.
The rethink group needs a different kind of attention. These are cases where the goal still matters but the design is wrong. The scope may be too broad, the data unreliable, or the review step too late.
Give each one a short redesign brief and a decision date. By then it either re-enters the portfolio as a new bet or moves to stop, so rethink never becomes a polite word for stalled.
Step 2: define what a system means
An experiment becomes a system when it meets a written standard. Our standard has seven parts, and it doubles as a checklist for the strengthen group.
- Ownership: a named owner responsible for its performance.
- Documentation: a recorded goal, tools, memory, and guardrails, as set out in the anatomy of a marketing agent.
- Versioned instructions: every change recorded with a note, as we describe in versioning agent instructions.
- Quality measurement: ongoing rubric scores or edit rate.
- Cost tracking: cost per task, reviewed monthly.
- Autonomy and escalation: a documented autonomy level and clear rules for when the agent stops and asks.
- Operating rhythm: a place in the team's regular reviews.
Any agent in the systematize group that misses one of these moves to strengthen until the gap is closed. That rule keeps the label meaningful.
Step 3: choose a small number of new bets
It is tempting to launch many new experiments in the second half. We recommend no more than two or three new bets, chosen with an impact and risk matrix.
The reason is capacity. Second-half calendars include summer slowdowns and year-end pushes, which leave less room for experimentation than a blank planning document suggests. Each new bet also competes for the same reviewers and owners that the strengthen group needs. A team that takes on five new experiments while strengthening four existing agents will likely finish neither.
Step 4: plan around the calendar
The second half has a distinct shape, though it varies by business. For many teams it looks roughly like this.
- July and August: lower volume in some businesses, which makes these months good for improving systems and running careful experiments.
- September: planning for Q4 and the following year.
- October to December: the holiday rush for many consumer businesses and budget closeout for many B2B ones, which is not the time to introduce risky changes.
Schedule agent launches and major changes so they are stable before your busiest periods. A useful rule is to count backward from your change freeze. If an agent needs four weeks of supervised running before it can be trusted at volume, it must launch at least four weeks before the freeze. Launch earlier if you want room for a fix.
Step 5: plan capacity for operations
Systems need care. Every agent that becomes a system needs ongoing time from someone to review logs, update instructions, and track quality. Budget that time explicitly, or the systems will slowly degrade.
Consider a hypothetical team that systematizes four agents and adds two new bets. Suppose each production agent needs two hours of operations time a week and each experiment needs five. That is 18 hours a week of work that did not appear in anyone's plan.
Spread across a small team, that load is easy to absorb if planned and easy to drop if not. The numbers will differ for every team, and the habit of adding them up is what matters.
What to measure through the second half
A plan is only useful if you can tell whether it is working. We suggest a short monthly check against the decisions made in this exercise.
For each systematized agent, track quality and cost against the levels it showed when it graduated. For each strengthen item, track whether the missing pieces are closing on schedule. For each new bet, track whether it is producing the evidence that would justify a decision by the end of Q3.
Watch the stop list too. If stopped experiments keep coming back under new names, the sorting criteria were not clear enough, or the team did not agree with them.
The question we are studying next is how the four groups should shift over a year. We expect the systematize group to grow and the rethink group to shrink. We do not yet know what a healthy balance looks like at the end of a first full year.
How to apply this: sort every agent into systematize, strengthen, rethink, or stop. Pick at most three new bets for the second half, add up the operations hours they need, and schedule major changes to be stable before your busiest weeks begin.
Related research
Q3 kickoff: three agentic experiments worth running this quarter
A calmer quarter suits careful experiments. These three teach a lot about planning, real-world quality, and AI answers without putting much at risk.
Agentic ABM: orchestrating research, outreach, and sales handoff
ABM depends on research and coordination that rarely scale. Orchestrated agents can extend it to more accounts while people keep the relationships.
Summer campaign planning with an agentic workforce
Summer thins out teams while campaigns keep running. Plan coverage, autonomy, and review ahead so proven agents carry steady work while people are away.
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.
