The First 30 Days After Your Agent Goes Live

Shopify's Spring '26 edition shipped Brand Agents and Copilot Checkout on June 17, and most of the merchants who turned one on that week are now past launch day and into the part nobody put on the project plan: what do you actually do for the next month? Launch is a milestone, not a finish line. An agent that goes live and gets left alone drifts — it answers yesterday's catalog, misses today's most-asked question, and nobody notices until a shopper complains or the numbers look flat. This article gives you the operating rhythm for the 30 days after go-live, week by week, so the agent gets better every week instead of quietly stalling.

Why the first 30 days are different from steady state

A newly launched agent is operating on assumptions: the catalog data you had at launch, the conversation flows you configured before a single real shopper used them, and guesses about what people would ask. Every one of those assumptions gets tested against reality starting minute one. The first month is when you find out which assumptions held and which didn't — and the cost of finding out late is a month of shoppers hitting the same gaps.

This is also the highest-leverage month you'll get. Attention is high, the team that built the launch is still engaged, and the fixes are cheap relative to what they're worth: a missing attribute or a badly worded flow costs almost nothing to fix in week one and costs a quarter of lost conversions if it sits until someone finally looks. Treat the first 30 days as a distinct phase with its own rhythm, not an extension of launch week or the start of steady-state operations.

Launch drift

The gradual mismatch between an agent's configuration and reality that sets in when nobody actively tunes it after go-live — catalog changes go unreflected, new question patterns go unanswered, and the agent that was accurate on day one becomes gradually less accurate with each week of neglect.

The 30-day rhythm, week by week

Four weeks, four distinct jobs. Each week's work depends on the one before it — you can't tune measurement in week 3 if you skipped monitoring in week 1, and you can't make an honest handoff decision in week 4 without weeks 1 through 3 behind it.

1. 1

   Week 1 — Close monitoring and daily transcript review

   Read every conversation, every day. Volume is still low enough that this is possible, and it's the cheapest week you'll ever have to catch a broken flow, a wrong price, or a policy answer that's flat-out incorrect. Fix anything broken same-day — don't batch week-1 fixes into a future sprint.

2. 2

   Week 2 — Conversation-flow gap-filling from real questions

   By now you have a real sample of what shoppers actually ask, and it will not match what you configured for at launch. Pull the recurring questions the agent hedged on or answered poorly, and rebuild the conversation flows around them — not around the launch team's best guesses.

3. 3

   Week 3 — Measurement baseline and tuning

   Enough volume has accumulated that the numbers stop being noise. Stand up the metric stack, compare against your pre-launch baseline, and start making configuration changes based on data instead of anecdote.

4. 4

   Week 4 — Decide handoff vs. retainer

   Look at what the first three weeks actually showed: how much ongoing tuning the agent needs, whether the team has the bandwidth and skill to keep doing it, and whether the fix rate is slowing down or still steady. Decide, in writing, who owns the agent starting week 5.

Week 1: what to actually monitor

"Monitor closely" is vague enough to mean nothing, so here's the specific version. Every day in week 1, read the full transcript list — not a sample, all of it. At typical early volume this is 20 minutes to an hour of reading, and it is the single highest-value hour in the entire 30 days. You're looking for three things.

• Factual errors. The agent states a wrong price, a discontinued product as available, or a policy that no longer applies. These erode trust fast and need same-day correction.
• Hedges and declines. The agent says some version of "I'm not sure" or refuses to answer a reasonable question. Each one names a data gap — usually a missing attribute — that you can log and fix.
• Tone and behavior problems. The agent pushes checkout too early, over-recommends, or answers in a voice that doesn't match the brand. This is a configuration issue, not a data issue, and it's the fastest of the three to fix.

This maps directly onto the failure patterns worth knowing before you start reading: catalog gaps, policy gaps, and configuration gaps. If you haven't already, it's worth reading what typically goes wrong with a new Brand Agent before week 1 starts, so you recognize the pattern the first time you see it instead of the fifth.

Week 2: rebuild the conversation flow around reality

Every launch configuration is a hypothesis about what shoppers will ask. By the end of week 1 you have data, and the data rarely matches the hypothesis exactly — some anticipated questions never come up, and a handful of unanticipated ones dominate. Week 2's job is to close that gap.

Pull the recurring questions from week 1's transcripts and sort them into two piles: questions the agent answered well, and questions it hedged, declined, or answered clumsily. The second pile is your week-2 backlog. For each one, decide whether the fix is a catalog fix (the fact doesn't exist as a structured field — see catalog enrichment for AI), a content fix (the policy or brand-voice guidance the agent draws on is thin), or a flow fix (the conversation design itself steers the agent wrong).

The most common week-2 finding

In our field work, the single most common week-2 discovery is a question category nobody configured for at all — often something specific to the merchant's category that the launch team didn't think to anticipate. A gift-focused retailer discovers half their agent traffic is gift-recipient sizing questions nobody scripted for. A parts retailer discovers compatibility questions dominate and the "works with" data is thin. You cannot know this in advance; you can only find it by reading week 1's transcripts and rebuilding around what's actually there.

Week 3: establish the baseline and start measuring

By week 3, volume has accumulated enough that metrics stop being noise, and this is the week to stand up real measurement — not before. If you tried to read numbers in week 1, you'd be reading launch-day novelty traffic and mid-flight configuration changes, which tells you nothing reliable.

The five-metric stack — conversation volume, intent-to-purchase rate, conversion rate versus your site average, attributed revenue, and AOV lift — is the same stack you'll use for the life of the agent, so week 3 is where you start it properly rather than bolting it on later. If you haven't set this up yet, measuring agentic commerce ROI walks through the full stack, the attribution rules for direct versus assisted revenue, and the baseline you need before any of the numbers mean something.

Two things matter specifically in week 3, beyond just standing up the dashboard:

• Compare against your pre-launch baseline, not against an assumed target. If you didn't capture a pre-launch baseline, reconstruct what you can this week — it's late, but not too late to be useful for month two.
• Start tuning against the numbers, not just the transcripts. If intent-to-purchase rate is low, the entry point or prompt placement may be the issue. If conversion versus site average is weak, go back to the transcripts — it's almost always a catalog or configuration gap, not a fundamental problem with the agent.

Week 4: decide who owns the agent next

Week 4 has one job: make the handoff decision, in writing, before week 5 starts without a clear owner. The question is simple even though the answer isn't always: does this move to an internal team, or does it stay on a retainer with outside help?

Base the decision on what the first three weeks actually showed, not on how the launch felt. Three signals matter most.

1. 1

   How much ongoing tuning does the agent need?

   If week 2's gap-filling turned up a handful of fixes that are now largely closed, the agent may be close to steady-state and cheap to maintain. If new gaps kept surfacing through week 3 at the same rate as week 1, expect that pace to continue and staff for it.

2. 2

   Does the team have the bandwidth and the skill?

   Reading transcripts and shipping catalog or conversation-flow fixes is a specific, recurring skill — not a one-time project task. Be honest about whether the people who did week 1-3 work can keep doing it at the same quality alongside their other responsibilities.

3. 3

   Is the fix rate slowing or holding steady?

   A slowing fix rate across weeks 1 through 3 is the strongest signal that the agent is approaching a stable state an internal team can maintain. A steady or rising rate suggests the catalog or conversation design needs more structural work before handoff makes sense.

Neither answer is a failure. A merchant with a strong internal ops team and a catalog that was well-enriched before launch often can and should bring the agent fully in-house by week 5. A merchant whose team is stretched, or whose catalog still needs structural work, is usually better served keeping outside support engaged until the fix rate genuinely slows. What matters is that the decision gets made deliberately in week 4 instead of by default — the worst outcome is an agent nobody owns, quietly drifting through month two.

What good looks like at day 30

By the end of the 30-day rhythm, you should have: zero unresolved factual errors from week 1's transcript review, a conversation flow rebuilt around the real question patterns from week 2, a measurement dashboard running against a real baseline from week 3, and a written decision from week 4 on who owns the agent going forward. If any of the four is missing, that's the gap to close before calling the launch complete — not a reason to worry, just the next thing on the list.

Before you begin the 30-day clock, it's worth confirming the agent went live on solid footing in the first place — the agentic commerce readiness checklist covers the pre-launch conditions that make week 1 monitoring find fewer surprises instead of more.