What Should an Ecommerce Brand Get Right Before Scaling With AI?
Get the foundations right before scaling with AI: mobile site speed first, then your existing customers, then product and inventory, then brand story. Sabir Semerkant, Founder of Growth by Sabir, argues on The Ecommerce Toolbox that AI scales whatever is already there, so a slow, leaky store only gets bigger problems faster.
This recap of episode 151 of The Ecommerce Toolbox: AI in Retail answers a question many teams are asking right now: what should an ecommerce brand get right before scaling with AI? The conversation between Noibu President and Co-Founder Kailin Noivo and Sabir Semerkant puts site speed, customers, product and brand ahead of new AI features and bigger ad budgets.
- AI scales what a store already is. A slow site or a one-time-buyer base gets bigger problems, not bigger growth.
- Sabir Semerkant's order: mobile site speed, then existing customers, then product and inventory, then brand story.
- His speed target is a page that becomes interactive in 1.7 seconds, with 2.5 seconds as the forgivable ceiling.
- AI agents do the work under guardrails: success criteria, a performance test before anything goes live, and a second model auditing the first.
Who is Sabir Semerkant?
Sabir Semerkant is the Founder of Growth by Sabir and runs Rapid 2X, a live coaching program for ecommerce and DTC brands. He has worked in ecommerce for 25 years, co-founded VaynerCommerce, and rebuilt brands including Vitamin Shoppe and Ashley Stewart after bankruptcy. On the episode, he says more than 200 brands have gone through Rapid 2X, and that the whole program runs on AI agents.
Why does AI make an ecommerce brand's existing problems bigger?
AI speeds up whatever a business already does, good or bad. Sabir Semerkant's view is that a slow site with vibe-coded new features is a slower site, and a weak customer base with more AI-generated ad creative is more wasted budget. He opens the episode with the line that sums up his approach.
“When you scale garbage, you have a massive garbage dump. That's what you have. Scaling problems give you more problems.” — Sabir Semerkant, Founder, Growth by Sabir
Sabir Semerkant ties this to the bankrupt brands he has rebuilt: businesses that report strong fourth-quarter numbers can still end up in bankruptcy filings in January when the underlying problems were never addressed. What to do: before adding an AI feature or raising spend, list what that change would amplify, and deal with the weak points first.
Why should site speed come before ROAS work?
Brands often arrive asking how to improve ROAS or get more traffic. Sabir Semerkant's first move is site performance, because a page that is slow to become interactive loses conversion and makes every paid click worth less. His target is 1.7 seconds to the first interaction, with 2.5 seconds as the upper limit, measured on mobile rather than on a designer's desktop.
The speed rule also applies to AI-built work. Sabir Semerkant connects AI coding agents to a performance testing tool so that every new theme page or product detail page has to pass a time-to-interact check.
“So, you don't take that theme page or product detail page; you don't take it live. You stage it first in a staging theme on Shopify, for example, and then you production it after it passes those tests.” — Sabir Semerkant, Founder, Growth by Sabir
What to do: set a mobile speed threshold, test every new page or feature against it in staging, and only promote changes that pass. For the revenue side of slow pages, see how much revenue slow page speed costs an online store and which Core Web Vitals actually move conversion.
Is your ad spend buying customers or one-time orders?
After speed, Sabir Semerkant looks at the customers a brand has already acquired, using RFM analysis: recency, frequency and monetary value. He says that on average about 90% of a Shopify store's customers buy only once, which means the ads are buying transactions rather than relationships.
“No, that's not your customer acquisition cost; it's your order cost.” — Sabir Semerkant, Founder, Growth by Sabir
Sabir Semerkant tests segmented reactivation offers to learn which groups of past buyers still care about the brand, and he holds off on loyalty programs until repeat buying is healthier. He also points out that unengaged email profiles cost money in tools that bill per profile. What to do: compare ad spend to orders and repeat orders, not just clicks. Check the site before you blame the ads, and look at why ad platform ROAS differs from store data.
What happens to product and brand once speed and customers are in order?
The third stage is product. Sabir Semerkant applies the 80/20 rule to inventory, arguing that about 20% of products drive 80% of revenue while slow stock ties up cash and warehouse space. He recommends clearing dead inventory, for example through a flash sale, and putting that cash behind proven bestsellers.
The fourth stage is brand story. Sabir Semerkant argues that product-page details such as return policies or fabric content are not a brand, and that without a clear reason to care, blog posts, ads and social content all sit on a weak foundation. What to do: review which products earn their space, then write down why the brand exists before scaling content.
How do you keep AI agents from scaling mistakes?
Sabir Semerkant uses a mix of AI models and puts every task through evals: a clear problem, a definition of success and the tests the output must pass. A second model then audits the first for gaps and edge cases, over three rounds. He also checks numbers in AI analysis, because models can get arithmetic wrong.
“What AI is missing is that context, that intelligence, somebody who can steer it to say that this is what good looks like or what best looks like.” — Sabir Semerkant, Founder, Growth by Sabir
What to do: give every AI task written success criteria, require a second check before anything reaches the store, and verify the numbers. For more on this, read how to verify AI-generated ecommerce insights before you act on them and why AI needs business context to improve an ecommerce site.
Where do Noibu's AI agents fit in a foundations-first approach?
The same test-before-live discipline runs through Noibu's agents. The Performance agent measures Core Web Vitals from real shopper traffic, ranks the worst-performing page groups and stages a change for review. The ROAS agent shows return on ad spend scored on orders, not just clicks, and recommends where to shift spend and targeting toward what converts. Nothing ships without the team's approval, and the agents work in the Noibu console and in AI tools such as Claude through Noibu MCP.
Ecommerce Toolbox listeners can try the agents on their own store with the podcast listener free trial, which runs until two changes are live and measured.
Where can I listen to the full episode?
Watch or listen to Growth by Sabir: Fix the Foundation Before You Scale With AI on the episode page, which includes the full human-reviewed transcript. The episode is also on Apple Podcasts and YouTube.
Frequently asked questions
What order should an ecommerce brand work on its foundations?
Sabir Semerkant works in a fixed order: site performance first, then the customers a brand has already acquired, then product and inventory, then brand story. Each stage takes about six weeks in his Rapid 2X program. The order matters because each later stage depends on the earlier one, so a faster site makes every customer, product and brand improvement count for more.
Does vibe coding new features make an ecommerce site slower?
It can. Sabir warns that building new features or landing pages with AI on a site that already has performance problems simply scales those problems. His safeguard is to give AI coding agents a performance test that every new page must pass in a staging theme before it goes to production.
How can a store tell whether it is acquiring customers or just buying orders?
Run an RFM analysis on existing customers: recency, frequency and monetary value. Sabir says that on average about 90% of a Shopify store's customers buy only once. If most buyers never return, the money spent on ads is buying one-time orders, and the real cost to compare is cost per order, not cost per customer.
Should a store add a loyalty program before it has repeat buyers?
Sabir advises against it. Installing a loyalty program while most customers buy only once makes it look like loyalty does not work, when the real problem sits earlier in the journey. He starts with segmented reactivation campaigns to learn which customer groups still care about the brand, and adds loyalty later.
How do you check AI agent work before it goes live on a store?
Give every task clear success criteria and tests to pass, then have a second model audit the first for gaps and edge cases. Sabir runs three rounds of audit and correction, and he double-checks any numbers in an AI analysis because models can make arithmetic mistakes. Store changes are staged and only go live after they pass.
Where do Noibu's agents fit in a foundations-first approach?
Noibu's Performance agent measures Core Web Vitals from real shopper traffic, ranks the worst-performing page groups and stages changes for review. Noibu's ROAS agent shows return on ad spend scored on orders, not just clicks. Nothing ships without the team's approval, which matches the test-before-live discipline described on the episode.
About Noibu
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.