← Research library

A Q1 retrospective template for marketing programs that use AI

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 started using AI and agents in earnest this year, and the first quarter is now closed. For many, Q1 produced a familiar mix: some real wins, a few abandoned experiments, and a lot of anecdotes.

Anecdotes make poor inputs for planning. A structured retrospective turns that mix into decisions for Q2, with an owner and a measurement plan for each one. This post sets out the format we use, which takes about 90 minutes with the people who did the work.

Before the session

A retrospective is only as good as what people bring into the room. Send a short request a week ahead, so the session is spent deciding rather than remembering.

  • Workflow list: every AI-assisted workflow each person ran during the quarter, including abandoned ones.
  • Evidence: any numbers collected, such as time logs, edit rates, or campaign results.
  • Examples: two or three outputs per workflow, one good and one poor where possible.
  • Friction notes: a sentence on where each workflow cost time.

Invite the people who used the workflows, not only those who chose them. Practitioners know where the time went. Appoint one person to run the session and keep time and one note-taker who records decisions as they are made.

Part 1: Inventory (15 minutes)

Start by listing every AI-assisted workflow the team ran during the quarter, including the ones that were quietly dropped. For each, capture four facts in a shared document.

  • What the workflow was supposed to do
  • Who owned it
  • Roughly how often it ran
  • Whether it is still running today

The abandoned workflows are often the most instructive. Someone stopped using each of them for a reason, and that reason is data. Ask about it directly, without blame, since the answer often reveals a design problem rather than a people problem.

Part 2: Evidence (30 minutes)

For each workflow still running, ask what evidence exists that it helps. Sort the answers into three tiers. The tiers describe the evidence, not the workflow, so a valuable workflow can still sit in a low tier.

  1. Measured: there is a number, such as time saved per task, edit rate, or conversion change, collected consistently.
  2. Observed: people believe it helps and can describe specific examples, but nothing was measured.
  3. Assumed: it is still running mostly because nobody turned it off.

Be strict about the first tier. A number counts as measured only if it was collected the same way across the quarter and compared with something, such as the previous process or a baseline sample. A single impressive week is an observation.

Expect many workflows to land in the observed or assumed tiers. That is normal in a first quarter, and it is exactly what Q2 should fix. If you want one measure to add quickly, start with human edit rate, since it needs only drafts and finals saved side by side.

Part 3: Friction (20 minutes)

Next, ask the people who used the workflows where they lost time. Ask for specific moments rather than general impressions. Answers tend to cluster around a few themes.

  • Fixing outputs that missed brand voice or facts
  • Copying data between tools by hand
  • Waiting for approvals
  • Re-explaining context the system should have known

Friction points are design inputs. Each one points to a specific investment: better context, better tools, better review flows, or a clearer brief. Re-explained context, in particular, usually means the brief format needs work, which we cover in writing briefs agents can actually follow.

Estimate the cost of each friction point roughly, in hours per week. A rough figure is enough to rank them, and ranking is what Part 4 needs.

Part 4: Decisions (25 minutes)

For each workflow, make exactly one of four decisions. Write the decision, the reason, and the evidence tier beside each workflow in the shared document.

  • Scale: measured evidence of value, ready for more volume or autonomy.
  • Instrument: promising but unmeasured, so add measurement before investing further.
  • Redesign: valuable goal, but friction is eating the benefit.
  • Retire: no evidence and no clear path to evidence.

Assign an owner and a Q2 checkpoint date to every decision. For "instrument" decisions, also name the measure and when it will be first reviewed. For content workflows, a shared scoring method such as our starter rubric makes the Q2 checkpoint comparable across reviewers.

A worked example

Consider a hypothetical team that ran 10 AI-assisted workflows in Q1. Suppose two were abandoned, two have measured evidence, four are observed, and two are assumed.

The decisions might look like this. The two measured workflows scale. Three of the observed ones get instrumented, and the fourth gets redesigned because reviewers spend most of their time fixing product facts. Both assumed workflows retire, and the two abandoned ones are documented and closed.

The team leaves with eight decisions, eight owners, and five measurement plans. That is a far better starting point for Q2 than 10 workflows running on goodwill. It also makes the next step clearer, which we discuss in Q2 planning for agentic marketing teams.

A note on candor

Retrospectives fail when they become celebrations. The point is not to prove that AI is working. The point is to find out where it is working, where it is not, and why.

A team that retires two workflows and doubles down on one has learned more than a team that keeps all 10 running on assumption. Make retiring a workflow a normal, respected outcome, so people bring forward the weak ones honestly.

What we are studying next

The four decisions work well for a first quarter, when most workflows are small experiments. We are studying how the format should change as programs mature, when workflows connect to each other and retiring one can affect several others.

How to apply this: run the four-part retrospective before Q2 planning is final, with evidence gathered a week ahead. Make sure every workflow leaves the room with a decision, an owner, a checkpoint date, and a measurement plan.

Related research

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.