Give the build a decision sentence
A dashboard can have a dependable refresh, carefully chosen colors and a loyal audience while leaving its purpose unresolved. Ask three people what they should do after reading it. If their answers differ, adding another chart is unlikely to settle the question.
The opening Conference Insights Engine episode offers a practical review standard for Slate work: identify the decision, its owner, the evidence, the action and the way the result will be assessed. David develops this as his synthesis across conference sessions on change, Querybase Views, yield volatility and origin sources. The framework connects those subjects without treating every local example as a universal prescription.
Start with one object people already use. Finish the sentence, “This build helps us decide…” Use a choice rather than a topic: which admitted students need follow-up this week is more useful than student engagement. The sentence gives the next design conversation a boundary.
In the recording · 17:52 ↗Separate maintaining the object from owning the choice
The person who maintains a query may have no authority to decide how its results should guide staff work. The episode separates technical ownership from decision ownership for exactly this reason. A working report can outlast the practice it was meant to support.
Write down both roles. Then ask who can change the action rule, who resolves an ambiguous case and where the reasoning will be kept. For a recurring vendor review, that reasoning might include the original use case and the basis for the last renewal. A successor should be able to reconstruct the choice without finding the person who made it.
In the recording · 18:58 ↗Choose evidence that answers the actual question
The origin-source discussion provides a useful boundary: the initial hand raise and the full recruitment journey are different analytical questions. A single field cannot reliably explain first entry, subsequent influence and the contribution of every campaign. Define what the field represents before making an investment decision from it.
The same discipline applies to a counselor list. Recent event attendance, application stage and a missing deposit may support a review queue. They do not establish why someone has not deposited. Keep the observable evidence separate from the explanation staff still need to learn.
For Querybase Views, the episode asks whether repeated calculations are expensive, whether a defined population needs guardrails and whether scheduled lag is acceptable. These are design questions to resolve locally. A faster result is useful only if its scope and freshness still fit the decision.
In the recording · 19:54 ↗Make the response visible
A score or dashboard becomes easier to use when the next step is explicit. The episode illustrates this with an admitted student who attended a high-intent event but has not deposited: a possible response is counselor follow-up within three business days. That is an example to assess against local practice, not a demonstrated outcome or an automatic rule for every student.
Try writing the response in four parts: when this evidence appears, this role reviews it, within this period, using this process. Include exceptions. A student who has already received help should not get duplicate outreach merely because yesterday’s list still contains the record.
Apply the same test to maintenance work. An unmapped source belongs in a governance queue with an owner. A contract nearing renewal needs a review window. Naming the queue without naming who acts still leaves the work unfinished.
In the recording · 21:18 ↗Measure use without inventing impact
Decide what improvement would look like before the next review. A reporting change might reduce manual preparation or make a weekly meeting more useful. A view might reduce query errors. Source cleanup might reduce unmapped values and clarify conversion reporting. Those are concrete results worth checking.
The recording also cautions against treating every improvement as causal. A deposit increase following a campaign does not, by itself, prove the campaign caused it. Record what changed, who was included, what comparison is available and what remains uncertain. Operational usefulness and causal evidence answer different questions; both deserve accurate descriptions.
In the recording · 22:38 ↗Leave the next owner a reviewable tool
Keep the first pass small enough to finish. Choose one build, write its decision sentence, name the owners, define an action rule and put a review date on the calendar. Preserve the explanation alongside the object so the next person can understand more than its configuration.
At the review, compare intended use with actual use. Was the evidence timely? Did someone act? Did the action help resolve the question? If the answers expose a gap, revise the definition, the workflow or the evidence before adding features. The useful outcome is a tool whose purpose can be explained, questioned and improved.
In the recording · 25:24 ↗