Blog › Agencies › Who Approves AI Changes on a Client's Store: the Agency, the Merchant, or Both?
Noibu blog

Who Approves AI Changes on a Client's Store: the Agency, the Merchant, or Both?

Who approves AI changes on a client's store: the agency, the merchant, or both?

Usually both, for different things. The agency approves the code and how a change is built, because it owns the build and the release process. The merchant approves anything that changes what shoppers see, pay or agree to. Write the split down by change type before any AI agent works on the store.

Last updated November 16, 2026

When an agency brings AI agents onto a client's store, the first practical question is not what the agents can do. It is who says yes before a change goes live. If the answer is vague, either changes stall waiting for the wrong person, or a change ships that the merchant never knew about. This playbook sets out a simple split, an approval matrix you can copy, and how to set it up.

Key takeaways

  • The agency approves how a change is built: code quality, testing and release timing.
  • The merchant approves what changes for shoppers: copy, pricing display, checkout, policies and brand.
  • High-risk changes, such as checkout and payment flows, need both before they ship.
  • Supervised agents make this workable because nothing ships until a named person approves it.

Part of: What Is an AI-Native Ecommerce Agency?

Why does approval need a clear owner?

AI agents speed up the find and build steps. A supervised agent can surface a problem, trace the cause and prepare a change in a fraction of the time a person would take. That moves the bottleneck to approval. If every change waits for a merchant who cannot read code, nothing ships. If the agency approves everything alone, the merchant loses visibility into changes on its own storefront, which is one of the first concerns merchants raise, as covered in should you worry if your ecommerce agency uses AI.

The fix is to split approval by what each side is accountable for. Agencies are accountable for engineering quality and safe releases. Merchants are accountable for their brand, prices, policies and customers. A change gets approved by whoever carries the risk it creates.

What should the agency approve?

The agency owns the build. That covers code review of every agent-built change, testing in a staging or preview environment, release timing, deployment and rollback. In practice this means agent pull requests go through the same code review the agency already runs for its developers, with the same standards. Questions about who owns the resulting code are covered in who owns the code when AI agents work on your store through an agency.

Technical changes with no visible effect for shoppers, such as resolving a JavaScript error, improving server response time or reducing script weight, can usually be approved by the agency alone, with the merchant informed in the regular report.

What should the merchant approve?

The merchant owns the storefront. Anything that changes what a shopper reads, sees, pays or agrees to needs merchant sign-off: product page copy, pricing and promotion display, shipping and returns messaging, legal and policy text, brand imagery and anything inside checkout. The merchant does not need to read the code. The merchant needs a clear before and after, the reason for the change, and the expected effect.

Approval matrix for AI agent changes on a client store
Change typeAgency approvesMerchant approvesExample
Invisible technical changeYesInformed in reportResolving a JavaScript error on a product page
Performance changeYesInformed, or approves if layout changesDeferring a script, resizing images
Product page copy and contentBuild and releaseYesRewritten product description
A/B testHow the test is builtHypothesis, variants and shipping the winnerNew add-to-cart button layout
Accessibility changeYesYes, on checkout and key journeysNew form labels and error messages
Checkout or payment flowYesYesChange to shipping step validation

How do you set up an approval workflow with a client?

Set up AI change approvals in five steps

  1. 1

    Agree the matrix

    Copy the matrix above into the statement of work and adjust it to the client's risk tolerance.

  2. 2

    Name the approvers

    One primary and one backup on each side, with the change types each can sign off.

  3. 3

    Use existing tools

    Route agent changes through the agency's current code review and the client's ticketing or chat.

  4. 4

    Set response times

    Agree how fast each side responds, and what happens during code freezes and peak weeks.

  5. 5

    Report what shipped

    Send a regular summary of approved changes, who approved them and the measured result.

The approval request itself should be short. A good request shows the page and the change side by side, says which agent found the issue and why it matters, links the pull request for the agency reviewer, and states the expected effect. That lets a merchant approve from a phone without a meeting.

Peak season needs its own rule. Many stores freeze code before Black Friday and Cyber Monday, and agencies running several clients need a shared policy for what can still ship. A version of that policy is covered in how agencies handle AI agents during a BFCM code freeze.

What happens if a change goes wrong?

Approval does not end at release. Agree in advance who can roll back a change and how quickly, and make sure the agent's measurement step reports the result to both sides. If a change causes a problem, the same split applies: the agency reverts and investigates the build, and the merchant decides on any shopper-facing response. Responsibility for problems is covered in who is liable when an AI agent's change breaks your store.

How do agencies handle approvals across many clients?

Agencies running agents on several stores keep one matrix template and adjust it per client, rather than inventing a process each time. Named approvers, response times and freeze windows differ by client; the categories stay the same. That keeps reviewers consistent across the book of business. The wider operating model is in how an agency can run AI agents across many client stores, and the questions merchants should ask their agency are in what should you ask your ecommerce agency about AI.

Where does approval happen with Noibu's agents?

Every Noibu agent runs the same loop: find, build, your approval, ship, measure. Nothing ships until a person approves it, which is what makes a split approval model workable. The approval step looks slightly different by agent, according to Noibu's AI agents page:

  • Bug Resolution agent: identifies top fixable issues, traces the root cause to the codebase and creates a pull request for review. The agency reviews and merges it.
  • Performance agent: measures Core Web Vitals from real shopper traffic and stages a change for review.
  • CRO agent: finds underperforming product pages and rewrites them; reviewers can make inline edits and publish to Shopify in one click, which is where merchant sign-off fits.
  • ADA agent: audits high-traffic pages against WCAG and stages changes for review, then re-audits after release.
  • A/B Testing and ROAS agents: report test winners and spend recommendations back for a decision before anything changes.

Agencies and merchants use these agents in the Noibu console and in AI tools such as Claude through Noibu MCP, so approvers can review changes where they already work.

Frequently asked questions

Who should approve AI agent changes on a client's store?

Usually both parties, for different things. The agency reviews the code and how a change is built. The merchant approves changes that affect what shoppers see, pay or agree to, such as copy, pricing display, checkout and accessibility changes to key journeys.

Can AI agents push changes to a store without approval?

Noibu's agents do not. Every agent runs a loop of find, build, your approval, ship and measure, so nothing ships until a person approves it. The approval step is where the agency and merchant decide who signs off.

Does the merchant need to review code pull requests?

Not usually. Most merchants are not set up to review code. The agency reviews and merges pull requests, and the merchant approves the outcome: what changes on the page and why.

Who approves A/B tests run by AI agents?

The merchant normally approves the hypothesis and what each variant shows shoppers, and the agency approves how the test is built. The merchant then approves shipping the winning variant.

What should an agency put in writing about AI approvals?

A short approval matrix by change type, named approvers on both sides, response times, what happens during code freezes, and how changes are rolled back. Add it to the statement of work or the onboarding document.

How do Noibu's agents fit an agency's existing review process?

Noibu's Bug Resolution agent creates a pull request for review, the Performance and ADA agents stage changes for review, and the CRO agent lets reviewers edit changes inline before publishing. Agencies route these through the code review and ticketing they already use.

The short version

On a client's store, the agency approves how AI-built changes are made and released, the merchant approves what changes for shoppers, and checkout changes need both. Write the split into the statement of work, name the approvers, and report what shipped. To see what Noibu's agents would surface on a client store, start with Noibu's free website audit.

Noibu is AI for ecommerce operations, the always-on optimization system that keeps online stores running at their best. A team of supervised AI agents finds what's broken, slow, or leaking revenue, does the work, and hands results back for human approval. Trusted by retailers like Free People, Guess, Liquid I.V., and Ariat.

Back to all blogs
‍
‍
See what Noibu’s AI agents would change on your store first.
Free website audit
Share

Keep your store running at its best, with AI agents that never clock out.