A decision log — one maintained row per decision, with what changed, what was expected, what happened, and who decided — is what keeps a conversion answer in the building after the person who found it leaves. Each row also gets a review date, since a result describes a buyer at a moment and buyers keep changing. A subscription program run this way has held a 48% engaged-subscriber-to-buyer conversion rate across 11,600 subscribers for over four years.
Pick any structural feature of your commerce experience and ask why it is built that way. The required account step, the three-tier shipping display, the order in which the subscription options appear, and the wording on the club signup. In most programs, the answer comes back as a person’s name and an approximate year, and the reasoning itself is unavailable because it was never written down anywhere except in the head of somebody who now works elsewhere.
The consequence is a specific and expensive kind of waste. A team spends a year answering questions; the answers live in a slide deck and a thread, and three years later, a new Director inherits a site full of decisions with no rationale attached. Everything looks arbitrary because, functionally, it is. The safe move is to re-test: ask the same questions again with the same traffic and the same resolution limits, and pay twice for one answer.
You are almost certainly the third or fourth person to hold your seat. Consider what you were handed when you arrived, then consider what you are currently on track to hand over.
The Decision Log
The log is a single maintained record of what has been settled about how your DTC program works. It is deliberately unglamorous: a table, one row per decision, owned by the person accountable for the number. Three components decide whether it becomes an asset or another abandoned document.
Component 1: One row per decision, with the reasoning attached
Each row carries six fields: what changed, on which surface, when it shipped, what outcome was expected, what was observed, and who made the call.
The expected outcome field is the one that does the work and the one that teams leave out. Recorded before the result arrives, it makes the row honest and turns the log into a calibration record: over a year, you learn whether your team systematically overestimates copy changes and underestimates friction removal, which is worth more than any individual finding in the table.
The last field matters for a reason that is unrelated to blame. A decision with a name attached is a decision somebody can be asked about while they are still here, and it separates the choices that were argued through from the ones that happened because a template defaulted that way.
Component 2: Give every shipped change a review date
A result is a description of a buyer at a moment. Your buyers keep changing, and the broader category is shifting beneath them: US wine volume has fallen by roughly 19% since 2019, with younger cohorts drinking less (Silicon Valley Bank, State of the US Wine Industry 2026). A conversion answer from four years ago was measured on a population that no longer exists in the same proportions.
So each row gets a review date at the time of shipping, sized to how fast the thing it touches moves. Pricing and offer decisions age fast. Structural checkout decisions age slowly. Copy and creative sit between them. At the review date, the row gets one of three verdicts: still holds, needs a re-test, or retire the change.
That single field is what prevents the log from becoming an archive of stale claims, the failure mode of every version of this that has ever been abandoned.
Component 3: Log the judgment calls, labeled as judgment calls
Most of your site was never tested and never will be, due to the resolution limit in Monday’s email. Those decisions still need rows.
A judgment call row records what changed, why it was believed to be right, and explicitly that no measurement was taken. The labeling is the entire point. An unlabeled judgment call is indistinguishable from a finding a year later, and a log that mixes the two is worse than no log, because it launders opinion into evidence and your successor cannot tell which is which.
There is a cultural effect here worth naming. When judgment calls are written down as judgment calls, a team no longer needs every decision validated, which is precisely what frees it to ship the backlog items nobody could ever measure.
Keeping It Alive
Every abandoned version of this document died the same way, so the failure mode is worth designing against directly.
It has to live where the work happens rather than in a separate system that somebody has to remember to open. A tab in the file your team already uses for planning beats a purpose-built tool nobody logs into. It has to be small enough to fill in during the meeting where the decision is made, which is the argument for six fields rather than fifteen. And it needs exactly one owner, because a record everyone is responsible for maintaining is a record nobody maintains.
The review date is what gives it a heartbeat. Put those dates on the same calendar you already use for planning, so the log surfaces itself a few times a quarter instead of waiting to be consulted. A row that resurfaces on its own gets a verdict; a row filed in a folder does not.
What the Log Is Actually For
The internal use is compounding: each answered question stays answered, and the program stops paying repeatedly for the same finding.
The external use is the one that affects your standing. A maintained log is the artifact that lets you walk into an ownership meeting and answer why any part of the experience is the way it is, with a date and a reason. The dependable version of that conversation is what earns a DTC Director the authority to make the next set of decisions without relitigating the last set.
There is a third use that shows up in your own week. A large share of the interruptions a Director absorbs are requests to re-explain a decision to somebody who was not in the room when it was made: a new hire, a consultant, an owner who read something over the weekend. Those conversations are unavoidable and not quick, because each one rebuilds the reasoning from memory. A row with a date, a reason, and a result answers most of them in a link, and the ones it does not answer are the genuinely open questions worth your time.
It also lets you inherit from outside your own program. Findings from real sources belong in the log as rows in their own right: members who can edit their packages show a 20.7% higher average order value and roughly 50% lower churn across 1.4 million memberships and 17,000 clubs (Commerce7 Data Drop, December 2025). That is a settled question you did not have to spend a quarter answering, and it should sit in your log with its citation, its review date, and a note on whether your program has acted on it.
Durability is the real return, and it is the pattern behind a subscription program we operate: 11,600 subscribers, a 48% engaged-subscriber-to-buyer conversion rate, sustained across more than four years at roughly a 5% response rate. Those are our own results rather than an industry benchmark. What keeps them stable over that many years is not a campaign; it is that the same questions do not get relitigated every time the calendar turns over, because the answers are written down.
This Quarter’s Action
Start the log backward. Take the ten most-questioned features of your current DTC experience, and write a row for each: what it is, when it was decided as best anyone can establish, whether any evidence exists, and a review date.
Most rows will say “no evidence, reason unknown,” and that is the useful output. You have just produced a prioritized list of the assumptions your revenue currently rests on, which is a better test backlog than any brainstorm could generate.
P.S. One row is worth adding today regardless of whether you build the rest: the decisions you have already made this year that nobody outside your team knows about. DTC accounts for roughly 68% of revenue for premium and mid-tier California wineries (Silicon Valley Bank, 2026), which means the choices in that table are not marketing details; they reflect the shape of the majority of the business. Monday’s email brings together the three systems from this week and shows what they look like as a single operating loop.
