Rollout guide
Three phases, ground, recommend, delegate, that take a team from first connection to agents acting on their own.
This guide is for the person leading a Meerkats rollout across a team or several businesses. The Quickstart gets one workspace to its first approved change in an afternoon; this guide gets a team from there to agents acting on their own, without anyone losing trust in the numbers along the way. It is organised as three phases of about a month each.
The principle throughout: recommendations first, approvals second, autonomy last, and only where the record earns it.
Before you start#
- One person who owns the rollout and the goal frame. They need Read & Write on every workspace.
- Admin access to each platform you’ll connect, or the people who have it.
- The first use case: frequent, data-rich, judgement-heavy, measurable. Advertising performance monitoring is the usual choice. Avoid starting with something that needs data you haven’t connected yet.
- A Slack channel for alerts and one for approvals, if you use Slack.
Phase 1: Ground#
Goal: every number in Meerkats is one your team would sign off on.
- One workspace per business
Create a workspace for each brand, client or market, with its currency and time zone. See Workspaces.
- Connect the sales source first
Shopify, or the marketplace seller account. Without it, revenue and margin stay empty and ROAS is only the platform’s number. Fill in cost per item in Shopify if you want margin.
- Connect the ad platforms
Meta, Google, Amazon, Flipkart, as the workspace uses them. Check auto-tagging and UTMs so clicks match orders.
- Reconcile once
Compare Cockpit with each platform’s own UI for the same week: same date range, same time zone, same attribution basis. Expect platform ROAS to match and corrected ROAS to be lower; that gap is the point. Note any difference you can’t explain and raise it with support before going further.
- Confirm the goal frame
Meerkats proposes targets, caps and protected entities from your data. Confirm or edit them. Mark anything that must never be touched automatically.
- Upload the rules of the business
Naming conventions, the scaling playbook, what you never discount, client instructions. See Documents. Connect the Slack channels where decisions get written down.
Don't skip reconciliation
Teams that go straight to automations and later find a number they can’t explain stop trusting all of them. One afternoon of comparing against the platforms, with the attribution bases understood, saves that.Phase 2: Recommend#
Goal: agents propose; people decide; the record builds.
- Turn on reports and alerts
A morning report by email or Slack, and alerts for the two or three conditions you check by hand today (spend over a cap, ROAS under target for three days, a product under seven days of cover with live ads). These run on their own; they change nothing.
- Add the first rules that change something, in suggest mode
Start with low-risk, reversible actions: budget reductions, bid reductions, negative keywords. Leave every rule in suggest mode so proposals wait in the Inbox.
- Decide on the evidence
Approve, edit or reject each proposal. Reject with a reason; it teaches the memory. Use Approve all low-risk once you’ve read a few and trust them.
- Review the record weekly
In Agents run history, look at predicted versus actual for each rule. Was the forecast right? Did the metric move as expected? A rule that is right most of the time is a candidate for autonomy; one that isn’t needs its threshold or window adjusted.
- Set access
Read Only for people who need to see; Read & Write for the people who decide. See People and roles.
Phase 3: Delegate#
Goal: the routine runs alone; people spend their time on judgement.
- Switch proven low-risk rules to auto
Only where the record clears the bar; Meerkats recommends the switch when it does. Only low-risk actions can run alone, caps still apply, and every run is still recorded.
- Keep medium-risk changes on approval
Budget increases, pauses and structural changes keep waiting for a person. That is by design, not a limitation.
- Add investigation and audit agents
Weekly investigations of what moved and why; a monthly audit per account. These produce recommendations for people, not actions.
- Extend to the next use case
Stock-aware ad control, competitor price tracking, CRM follow-ups. Each new connection joins the same context, so later use cases are faster to set up than the first.
- Review monthly
The Activity log and run history for each workspace: what ran, what changed, what was rolled back, what credits were used. Adjust frames, caps and cadences.
Roles in a rollout#
| Role | Does | Access |
|---|---|---|
| Rollout owner | Owns workspaces, the goal frame and the autonomy decisions | Owner |
| Operators | Build rules, approve proposals, read run history | Read & Write |
| Platform admins | Connect tools; may not use Meerkats day to day | Read & Write during setup |
| Stakeholders and clients | Read Cockpit and reports | Read Only |
| Developers | Build on the API or connect Claude over MCP | Keys scoped per workspace |
Milestones#
| By the end of | You should have |
|---|---|
| Week 1 | Every workspace connected, reconciled, with a confirmed goal frame |
| Week 2 | Morning reports and the first alerts running; documents uploaded |
| Week 4 | Three to five suggest-mode rules with a decision record |
| Week 8 | The first low-risk rule on auto; medium-risk on approval; weekly investigations |
| Week 12 | A monthly review habit, the next use case live, and a clear view of credits per workspace |
Support during rollout#
Email support on paid plans; priority support on Growth and Pro; Pro includes an onboarding call. Ask the chat agent on Cockpit about Meerkats itself as well as your data; it reads these docs.