40 Questions Retail and E-commerce Should Ask Before Trusting AI
Your buyers built their instinct over years. Your brand was made distinct by curated choices. AI tools like Salesforce Einstein and Dynamic Yield can optimise conversion, but they optimise what they can measure. Ask these 40 questions before you let algorithms replace the judgement that built your business.
These are suggestions. Use the ones that fit your situation.
1When Salesforce Einstein recommends which products to feature on your homepage, what customer behaviour is it measuring? Is it only looking at past clicks and purchases, or is it also accounting for brand equity and margin?
2Your buyers have refused to stock certain trends because they clash with brand identity. Does your demand forecasting model know why those decisions matter, or does it only see lost sales?
3If Google Cloud AI predicts high demand for a product category, have you checked whether it is predicting demand that your competitors are already meeting, or demand that only you could create?
4When you use AI to decide which SKUs to discontinue, are you also losing the institutional knowledge about why those products existed in the first place?
5Does your merchandising team review AI recommendations for seasonal products three to six months in advance, or do you only see them when the algorithm thinks demand is already rising?
6If your demand forecasting tool recommends stocking 40 percent more of a trending item, can you trace back through the logic to see whether it is responding to a real trend or to your competitors' inventory?
7When AI personalisation shows different product assortments to different customer segments, how do you check whether you are curating experiences or just showing people what they clicked on last month?
8Are your buyers still making the strategic calls about which new designers, brands, or categories to bet on? Or has AI selection of existing inventory absorbed that decision-making?
9Your merchandising decisions shaped how customers understood your brand five years ago. What will shape it five years from now if algorithms choose your assortment?
10When you compare AI-selected product layouts to buyer-curated layouts in the same period, which one drove higher customer lifetime value, not just higher conversion rate?
Demand Forecasting and Inventory Risk
11What happens to your safety stock levels if you trust AI demand forecasts but the algorithm has never lived through the kind of disruption you are about to face?
12Does your demand forecasting model account for the fact that you have control over what you buy and display, which shapes demand itself?
13When Google Cloud AI forecasts demand for next quarter, does it know the difference between inventory you chose to reduce and inventory that customers stopped wanting?
14If your forecasting tool predicts a 25 percent drop in a category, can you challenge that prediction using buyer experience, or do you have to accept it as fact?
15Are you comparing AI forecast accuracy against a baseline of what your buyers would have predicted with the same data? You need to know if the algorithm is actually better.
16Your demand forecast directly affects what you buy. If the forecast is wrong by 20 percent, what is the cost in dead stock versus stockouts, and which direction is the error more likely to go?
17Does your AI model separate demand signals for bestsellers from signals for distinctive products that only you stock? They need different forecasting logic.
18When you see a demand spike in the model, do you pause to ask whether customers suddenly want it or whether your inventory was low so conversions fell?
19If your forecasting accuracy has been 85 percent historically, and the AI model claims 90 percent accuracy, what metric changed? Is it actually more accurate or just better at predicting what you already know?
20Are you stress-testing your demand model against the last two years of unexpected events? If not, you are trusting a model trained on stability.
AI-Mediated Customer Relationships
21When Dynamic Yield personalises the customer experience, is it optimising for the purchase you will make this week or for the relationship that will keep you buying for five years?
22Your email marketing tool with Klaviyo AI suggests sending different messages to different segments. How do you check whether those differences deepen connection or just increase open rates?
23If ChatGPT powers your customer service responses, how do you audit whether it is solving the customer's problem or just closing the conversation quickly?
24When your AI personalisation engine recommends products based on browsing history, could it be trapping customers in feedback loops where they only see what they have already shown interest in?
25Does your customer service AI flag when a customer is asking the same question repeatedly, suggesting the first answer did not actually solve their problem?
26If you are using AI to decide which customers get offered discounts and which do not, are you inadvertently training price-sensitive customers to never buy at full price?
27When Salesforce Einstein recommends the next product to show a customer, is it choosing based on what that individual customer values or on what the aggregate model knows sells well?
28Your brand built loyalty through distinctive service and human attention. If all customer interactions are now AI-mediated, what personal relationship remains?
29Are you measuring customer satisfaction separately from conversion rate? AI optimisation often improves one while degrading the other.
30When a customer reaches out with a problem, does your AI system route them to the right person, or does it attempt to resolve it algorithmically because that is cheaper?
Brand Differentiation and Competitive Risk
31Your competitor is also using Salesforce Einstein for personalisation. If you both use the same tool with similar data, why would your customers see different experiences?
32When you optimise product pages using AI, are you converging toward the same layout, imagery, and copy that your competitors are also optimising for?
33If you use the same demand forecasting model as five other retailers in your category, and you all stock based on the same forecasts, what stops your assortment from looking identical?
34Dynamic Yield and similar tools make personalisation easier. But if personalisation becomes invisible and uniform across the industry, where is the competitive advantage?
35Are you tracking what makes your brand distinctive three years ago and checking whether AI decisions are eroding those qualities?
36When you use ChatGPT to write customer communications, how different is your voice from every other retailer using the same tool?
37If your buyers stop making instinctive bets on emerging designers or niche categories because AI says they are too risky, could a competitor build a brand around exactly that?
38Are you using AI to defend what you already do well, or are you using it as an excuse to stop doing the things that cost money but built customer loyalty?
39Your brand story is built on choices that made sense at the time even if data did not support them. Which of those choices would your current AI system have rejected?
40When all retailers use AI to personalise at scale, the only differentiation left is in what you choose not to optimise. Are you protecting any areas from algorithmic pressure?
How to use these questions
Before you trust a demand forecast, run it past a buyer who has worked in your category for five years. If they disagree with the model, investigate why before you order based on the algorithm alone.
Measure your customer experience in two ways: conversion rate and repeat purchase rate. If one is rising and the other is falling, your AI personalisation may be optimising the wrong thing.
Set aside one merchandising section or product category and keep it free from AI recommendation tools for the next quarter. Stock it using buyer instinct only. Then compare how it performs.
When you implement a new AI tool, define in advance what kind of error you can tolerate. Is being 15 percent too optimistic about demand worse than being 15 percent too pessimistic? Different answers change which AI outputs you should trust.
At least quarterly, ask your team: what would we do differently if this AI tool did not exist? If the answer is nothing, the tool is just confirming decisions you would have made anyway.