# What did your company learn from its last ten campaigns?

Fabio Munhoz · 2026-11-20

Canonical: https://fabiomunhoz.com/blog/learning-from-last-ten-campaigns/

Language: en

A decision record can turn campaign results into knowledge the next campaign can use.

## In brief

Learning from campaigns requires recording the question, hypothesis, intervention, comparison, and limitations of the conclusion. A useful decision explains what will change and why. Preserve historical evidence while identifying recommendations that have become outdated.

Ask for the reports from recent campaigns. Your team can probably find spending, reach, replies, and opportunities. Now ask a different question: which decision changed because of those results?

Answering that question requires something a dashboard alone does not provide. The team needs to recover what it believed, what it chose to test, what it observed, and which interpretation it considered reasonable.

The ten campaigns in the title are a starting point for a conversation, not a statistical requirement. The purpose is to examine whether the work left behind usable knowledge. A team can learn from a small test or repeat many campaigns without explaining what it should do differently.

## Reports describe results; decisions need a record

A completed campaign can leave a polished presentation and no guidance for the next person. “It performed better” is an incomplete conclusion unless it explains for whom, compared with what, and under which conditions.

I would look for a record connecting the hypothesis to the outcome. It does not need to be long. It needs to preserve enough information for someone who did not participate in the campaign to understand and challenge the reasoning.

This is the cycle I introduced in [Context is king. How does it become revenue?](https://fabiomunhoz.com/blog/context-becomes-revenue/): execution puts a hypothesis to the test, and review helps update the next decision.

## A record someone else can use

The following template is a proposed working method. It can live in the system your team already uses, provided people can find it and preserve its relationship with the underlying evidence.

| Field | What to record |
| --- | --- |
| Question | Which commercial uncertainty prompted the campaign? |
| Hypothesis | What did we expect to observe, and why? |
| Intervention | What changed, for which audience, and during which period? |
| Comparison | Against which reference would we assess the outcome? |
| Observation | What happened, including unwanted effects? |
| Limitations | Which alternative explanations remain possible? |
| Decision | What should we keep, change, stop, or investigate? |
| Validity | Who owns the conclusion, and when should it be reviewed? |

Separating observation from interpretation prevents an opinion from becoming a fact simply because it was copied into a document. “We received more replies” describes an outcome. “The new positioning caused the increase” requires additional analysis.

## When a conclusion goes too far

Consider an illustrative scenario. A company changes its campaign message and sees more meetings. During the same period, it replaces the account list and starts offering a free assessment.

The result may be commercially useful. It does not isolate the effect of the message. Declaring the new copy a winner could lead the team to reuse it in situations where the other conditions do not exist.

A more useful conclusion would be that the combination of audience, message, and offer produced a result worth investigating. The team does not yet know which component explains the difference.

The next decision could be to test one change at a time under comparable conditions. Where appropriate, controlled experiments help evaluate hypotheses. Microsoft's Experimentation Platform describes hypothesis validation and impact measurement as part of product development. Applying that discipline to campaigns requires attention to differences in context and design. [Source: Microsoft Experimentation Platform](https://www.microsoft.com/en-us/research/group/experimentation-platform-exp/).

## Memory needs a way to retire outdated conclusions

Keeping every report is not enough to guide future work. A conclusion may depend on an old price, an offer that has changed, or a segment that is no longer a priority.

I would preserve the historical evidence while updating the conclusion's status: current, under review, or superseded. The new decision should reference the earlier one and explain what changed. This keeps the history without presenting every old recommendation as current guidance.

Access must also respect the permissions of the original sources. A summary does not automatically make customer or negotiation information available to every agent. Reusable learning needs to be separated from the restricted data that produced it.

## Make the learning visible in the next brief

One way to test a record's usefulness is to give it to the person preparing the next campaign. Can that person identify a choice to preserve and a condition to check before reusing the conclusion?

In the messaging example, the next brief could note that the previous test changed the audience and offer at the same time. The team would know it still needs to investigate the effect of the copy. It could also preserve the objections received, separately from the interpretation made by the person reviewing responses.

If the next campaign targets another segment, the decision should explain why the comparison is relevant or why a new investigation is needed. This prevents a recommendation from extending beyond what its evidence supports.

During review, ask for a reference to the earlier record and a description of the change made. The team may choose to keep everything the same, provided it explains why. The purpose is to make the decision traceable, including a decision to continue. Without that handoff, memory remains available for consultation, but nobody knows whether it informed the work.

## Where agents can help

Agents can support record organization, retrieve related tests, and flag differences between a new proposal and previous decisions. Their output should identify the records used so the team can check the comparison.

Interpretation still requires care. Similarity between campaigns does not guarantee that a conclusion transfers. Audience, offer, channel, and timing can change what a comparison means.

I would start the next meeting by choosing one completed campaign and writing down a single decision its evidence supports. Then I would check whether that decision appears in the next brief and changes something observable.

**Learning becomes operational when someone can show what they will do differently and why.** That is a more useful contribution to company memory than accumulating presentations nobody consults.

## Questions and answers

### What distinguishes a report from a learning record?

A report describes outcomes. A learning record also preserves the hypothesis, conditions, limits of interpretation, and decision that will guide the next round of work.

### Do more meetings prove that a message worked?

Not on their own. Changes in audience, offer, or timing may explain the difference. Examine the comparison before attributing the effect to the message.

### How can a team keep campaign memory current?

Preserve the records and mark conclusions as current, under review, or superseded. Identify an owner and explain what changed when a new decision replaces an earlier one.
