Start with what actually happened
A student returns to a portal several times without completing the next step. The record shows activity. It does not explain whether the student is exploring options, looking for a missing instruction or struggling with the page. Calling that student highly interested adds a conclusion the observation cannot establish.
In From Signal to Action, David develops a practical standard for using behavioral data: give each active signal a defined, proportionate response. This is his synthesis across four Summit sessions on personalized outreach, targeted campaigns, conversion rate optimization and yield volatility.
Begin with a behavioral label staff can explain. Portal visits with an incomplete checklist describes the record. Unmotivated student assigns a motive. Keep the observation, possible explanations and confidence level separate so a useful clue does not harden into an unsupported judgment.
In the recording · 13:40 ↗Choose the decision before choosing the score
The recording describes a source-session integration that separates application, yield and anti-melt use cases. David uses that distinction to question a single engagement score serving every stage. A student finishing an application and a deposited student navigating onboarding need different forms of help, even when both generate repeated visits.
Choose one decision your team can own, such as how to offer completion support for a missing onboarding task. Record which status makes the task relevant and how current that status must be. Define the support staff may offer without inferring personal circumstances from browsing behavior.
A number alone cannot settle these questions. If a threshold is experimental, label it that way. Make its purpose and limits visible before staff use it, and avoid treating a ranking as a validated prediction.
In the recording · 06:59 ↗Inspect the experience that produced the evidence
The conversion rate optimization discussion connects the visible student journey with the signals collected behind it. A form abandoned on a phone may reveal an awkward form. A page with several competing calls to action makes a click harder to interpret. More tracking does not repair an unclear path.
Before changing outreach, walk through the relevant page and task. Can someone find the next step? Does the form request information appropriate to that step? Are the labels and tracking definitions consistent? Record experience problems separately from assumptions about students. Sometimes the useful action is to repair the page and review the resulting behavior before adding another campaign.
In the recording · 08:21 ↗Complete the Signal To Action Matrix
The matrix brings eight decisions together: observed behavior, possible interpretation, confidence, allowed action, owner, timing and channel, suppressions, and measurement. Its value is the conversation between those fields. A plausible interpretation still needs a response the team can justify and someone responsible for carrying it out.
For an incomplete onboarding task, an allowed response might be a clear help link or an offer to answer questions. Name the office responsible, the contact window and the point at which another office takes over. Preserve the student's choice about whether to respond. Leave the signal unused for outreach if the team cannot explain why that response fits.
Write exclusions alongside the action. Check for task completion, a recent staff interaction, an existing support conversation and communication preferences. A policy that only describes when to send can produce duplicate messages even when every individual campaign looks reasonable.
In the recording · 12:41 ↗Give staff enough context to use the list
A list labeled high engagement leaves too much interpretation to its reader. Staff need to know which behaviors produced it, when the data refreshed, what the label leaves unresolved and which response is allowed. The episode treats this training as part of the build.
Use one fictional record to test the handoff. Ask the receiving staff member to explain why it appeared, what they would check first and when they would hold back. If the explanation depends on finding the original builder, improve the documentation before expanding the workflow. Apply the same test to an AI-generated suggestion: identify its evidence and require an appropriate review before it changes outreach.
In the recording · 21:15 ↗Review the action, then maintain the signal
Choose a measure connected to the help being offered. Checklist progress or completion of a requested appointment can be more informative than an email click. Record the period and population reviewed, and do not describe movement after outreach as proof that outreach caused it.
Start with one signal and a review date. As the work grows, keep a ledger with its source, definition, refresh schedule, owner, allowed uses, suppressions and retirement conditions. At review, ask whether the evidence remains reliable, the response remains appropriate and staff can still explain it. Keep, revise or retire the rule on that basis.
In the recording · 23:03 ↗