The conversion data coming out of THG Ingenuity's AI Shopping Assistant deployment is not incremental. It is a structural break from everything the industry assumed about conversion rate optimization. THG Ingenuity, in collaboration with Google Cloud, launched an AI Shopping Assistant built on Gemini Enterprise Agent Platform that produced 8x higher conversion rates versus site average, a 5.5x increase in first-time buyer conversion, and a 20.8% average order value uplift. For returning buyers, CVR increased 4.61% and AOV increased 9.6%. These numbers came from Myprotein, one of the world's largest sports nutrition brands. They are not projections or pilot data. They are live results on a high-traffic, high-competition ecommerce store.
If your store does $1M–$20M in annual revenue and you're running conversion rate optimization the old way: A/B testing button colors, tweaking landing page copy, adjusting checkout flow, you are competing with a rowboat against a submarine. The game has changed. Here is how to run the same playbook at your scale.
What Actually Happened at Myprotein
Before we get to implementation, understand what the AI Shopping Assistant actually did. This is not a chatbot that answers FAQ questions. It is a conversational AI that sits inside the product discovery and purchase journey and guides customers through decisions that most stores leave entirely up to the shopper.
A first-time buyer lands on a sports nutrition site with 400 SKUs, 12 protein categories, and 30 flavor options per product. The traditional experience: scroll, click, read product descriptions, get confused, leave. Conversion rate on first visit: 1–2% on a good day.
The AI Shopping Assistant changes the interaction. The buyer answers 3–4 questions: What are your fitness goals? Do you have dietary restrictions? What's your budget per month? The AI then narrows 400 SKUs to 3–5 specific recommendations with explanations tied directly to the buyer's stated goals. The customer isn't browsing anymore. They're being served. That's the mechanism behind 5.5x first-time buyer conversion.
The AOV uplift comes from a different mechanic. The AI understands purchase context. If someone buys a whey protein, the assistant recommends a creatine that's compatible with their goals, a shaker bottle that matches their order size, and a subscription frequency that makes financial sense for their usage. These are not generic upsell pop-ups. They are contextually relevant additions that a knowledgeable sales associate would suggest. Customers accept them at dramatically higher rates.
Why the Old CRO Playbook Runs Out of Runway
I watched a similar dynamic play out in capital raising. At Angel Investors Network, early on, we tried to optimize pitch deck formats, investor targeting lists, and follow-up email sequences. The marginal improvements were real but small. The structural improvement came when we built a systematic qualification process that ensured every investor-founder meeting was high-context: the investor already understood the deal, the founder already understood the investor's thesis. Conversion on meetings doubled. Not because of better presentation. Because the match quality improved before the meeting started.
AI Shopping Assistants do the same thing. They improve match quality before the purchase decision. The customer who reaches the cart has already been qualified: their needs match the product, the price point fits their stated budget, and the use case has been confirmed. You're not trying to convert a cold browser. You're confirming a warm decision.
Traditional CRO never touched this layer. It assumed the product-customer match happened organically through browsing. It optimized the checkout, not the discovery. AI Shopping Assistants optimize discovery, which is where 85% of the conversion opportunity lives.
The Mid-Market Implementation Playbook
You are not Myprotein. You don't have THG Ingenuity's engineering team or Google Cloud's enterprise contract. Here is what you can actually deploy in the $1M–$20M revenue range.
Phase 1: Fast Simon for Shopify stores (Days 1–30). If your store runs on Shopify, Fast Simon's AI Personalization platform launched built specifically for brand merchandisers with documented double-digit conversion increases. It addresses the discovery problem directly: AI-powered search, personalized collection pages, and merchandising rules that adapt to individual shopper behavior. The implementation timeline for a Shopify store is 2–4 weeks. You don't need a developer. The platform integrates through Shopify's app ecosystem.
The specific Fast Simon features to prioritize: AI Search (replaces standard Shopify search with intent-understanding search that surfaces the right product even when the query is vague or uses informal language), AI Collections (dynamically re-ranks product listings based on individual shopper signals), and Smart Upsell (contextual recommendations at product page and cart). Start with AI Search. It has the fastest visible impact on conversion because it reduces the zero-results and wrong-results problem that causes 30–40% of search-initiated sessions to end without a purchase.
Phase 2: Conversational guidance layer (Days 31–60). This is the mechanic that drove Myprotein's 5.5x first-time buyer conversion. You need a product discovery quiz or guided selling tool. For stores under $5M revenue, Octane AI's quiz builder or a custom Typeform-to-Klaviyo sequence can approximate this. For stores in the $5M–$20M range, platforms like Rebuy or Nosto offer more sophisticated recommendation engines that sit natively inside the product discovery flow.
The quiz design is critical. Keep it under 5 questions. Each question should directly inform a filtering decision that narrows your product catalog. Bad question: "What's your fitness level?" Good question: "How many days per week do you currently train?" The second question produces a specific, actionable answer that maps to product recommendations. Design the quiz to the product decision tree, not to customer psychology.
Phase 3: Return buyer optimization (Days 61–90). The Myprotein data shows 4.61% CVR uplift and 9.6% AOV uplift for returning buyers. This is a different problem than first-time conversion. Returning buyers already know your catalog. Their friction is decision fatigue and subscription timing. The AI opportunity here is purchase prediction: identifying when a returning buyer is likely to need to reorder based on their previous purchase history and typical usage rates.
Klaviyo's AI predictive sending, Recharge's AI subscription timing tools, and Yotpo's loyalty-plus-recommendations stack all address this. The specific implementation: set up a predictive reorder sequence that sends personalized restock reminders 3–5 days before the customer is likely to run out of their most recent purchase. Include a one-click reorder button. Returning buyer sessions that start from a predictive reorder email convert at 3–5x the rate of organic returning buyer sessions.
The Specific Numbers to Measure
Running an AI Shopping Assistant without defined measurement is like standing a watch with no gauges. Here are the metrics and the baselines.
First-time buyer CVR. Benchmark your current first-time buyer CVR before deployment. Most stores run 1.5–2.5% on first-visit buyers. A well-deployed discovery quiz should move this to 4–6% within 60 days. If it doesn't move, your quiz design is wrong: revise the questions or the recommendation logic.
AOV by session type. Segment AOV for quiz-assisted sessions versus unassisted sessions. Expect 15–25% AOV uplift in quiz-assisted sessions from day 1. If you see less than 10%, your upsell recommendations aren't contextually relevant. Audit the recommendation logic against actual purchase patterns.
Zero-result search rate. Before AI Search: most stores see 12–20% of searches return zero results or irrelevant results. After AI Search: this should drop below 3%. Track it weekly for the first 60 days. Zero-result searches are lost revenue on an explicit demand signal.
Email-to-purchase conversion for predictive reorder sequences. Baseline for standard promotional emails: 1.5–3% click-to-purchase. Baseline for predictive reorder emails sent at the right timing: 8–15%. If your predictive reorder emails aren't outperforming promotional emails by at least 3x, the timing model needs recalibration.
The Catalog Preparation Work Nobody Talks About
The AI Shopping Assistant results at Myprotein didn't happen because they connected a tool. They happened because Myprotein's product catalog was prepared for AI interpretation. This is the work that most mid-market operators skip, then wonder why their AI recommendations are poor.
Every product needs: a structured title that includes the product type, key differentiator, and relevant attribute (not "Vanilla Flavor" but "Impact Whey Protein: Vanilla, 5.5g BCAAs per serving, Grass-Fed"). Every product needs: a description that answers the three questions a knowledgeable customer would ask: What is it? Who is it for specifically? When do you use it? Every product needs: accurate tag and attribute data that the AI can use for filtering: dietary properties, use case, compatible products, price tier.
This catalog work takes 2–4 hours per 100 SKUs for a team member who knows the products. For a 500-SKU catalog, that's a 10–20 hour project. Do it before you launch any AI feature. The AI is only as good as the structured data it reads.
Q: Our store is on WooCommerce, not Shopify. Can we still run this playbook?
Yes. Fast Simon supports WooCommerce as well as Shopify. Rebuy and Nosto also have WooCommerce integrations. The implementation timeline is typically 2 weeks longer on WooCommerce due to less standardized theme architecture. Budget for a developer day to handle the integration and QA. The ROI math is the same regardless of platform.
Q: What's the minimum catalog size where AI personalization meaningfully helps?
The discovery problem that AI Shopping Assistants solve becomes significant at around 50+ SKUs. Below 50 products, customers can browse the entire catalog in 3–5 minutes without getting overwhelmed, and the AI recommendation layer doesn't materially shorten the decision path. Above 200 SKUs, AI-assisted discovery is effectively mandatory for competitive first-time buyer conversion. The Myprotein use case, with hundreds of SKUs, is the clearest example of why.
Q: How do you justify the platform costs at lower revenue levels?
Run the math on your current first-time buyer CVR. If you're at 2% and the AI tool moves you to 4%, you've doubled first-time buyer revenue from the same traffic. Fast Simon's pricing starts at $99/month for basic plans and scales with order volume. At $1M annual revenue, if 30% of revenue comes from first-time buyers ($300K), and you improve first-time buyer CVR by 50%, you're generating $150K in incremental revenue for $1,200 in annual platform cost. The math works at $500K annual revenue.
Q: What's the biggest implementation mistake mid-market stores make?
Deploying the AI tool on an unprepared catalog, then attributing poor results to the tool. The AI recommendation quality is directly proportional to product data quality. A quiz that recommends the wrong products because SKU attributes are missing or inaccurate will produce worse results than no quiz at all. Catalog preparation is not optional. It is the foundation.
Q: How long before we see statistically significant conversion data?
For stores with 10,000+ monthly sessions, you'll have statistically significant data within 30 days. For stores with 2,000–5,000 monthly sessions, plan for 60–90 days before drawing conclusions. Split the traffic: send 50% through the AI-assisted discovery flow and 50% through your existing flow for a clean comparison. Don't turn off the old flow during testing. You need the control group.
Doctrine Connection
The Myprotein results are not magic. They are the outcome of building a system that closes the gap between what a customer needs and what they purchase. In military terms, this is fire control: the process of ensuring your asset reaches the right target with the right timing. Traditional ecommerce is unguided. You launch products at a broad audience and hope for a hit rate above 2%. AI Shopping Assistants are guided fire control. Every customer interaction is aimed. The hit rate goes up by a factor of 5–8 because the guidance system is working. The operators who deploy this system in 2026 will have conversion rates that their 2024-era competitors cannot replicate with budget or headcount alone.
Research from Salesforce's State of Commerce report consistently shows that personalization is the single highest-ROI investment in ecommerce, with AI-driven personalization delivering 4–10x the return of rule-based personalization. The Myprotein numbers confirm the thesis at production scale. The tools to replicate this exist for mid-market stores today. The catalog work takes a week. The platform integration takes two to four weeks. The measurement setup takes two days. By day 90, you will have a documented conversion lift number. That number changes every board conversation, every investor conversation, and every unit economics model you build going forward. Deploy it.