When athletes realize they’re underperforming, they have roughly two choices: to try harder or to try something different. When I started snowboarding, my jumps were often tail-heavy. I used to go flying from the ramp and land on my back, sometimes on my neck, again and again, until my body was so sore it couldn’t take anymore. Finally, I realized that my center of gravity was too far back when taking off, and when I moved it a bit towards the front, things started working. The number of jumps didn’t bring the improvement, changing the technique did.
When e-commerce is performing on average, it is underperforming, as only 2% of visitors conclude their purchase. That means that 98% of visitors don’t make a purchase (for comparison: in many brick-and-mortar stores the numbers are the other way around). To improve its sales, an e-store must either increase the customer flow or develop the technique with which they meet the client.
Let’s say, that Jane sets up an e-store and aims at M€1 in sales revenue. Executed smartly, a store of that size could only just make a living for its owner. Jane sells interior decoration items, with the average price on the plus side of €40, and on average a visitor buys one product per visit. This means, that Jane must get 2000 buying customers each month to reach her goal of a million. To get those 2000 buyers, the store needs all in all 100 000 visitors a month (if the conversion rate is the above mentioned 2 %).
Google and Facebook happily offer Jane their assistance, because digital marketing is what makes digital sales, right? If Jane were to decide to buy all her traffic, and 98% of the visitors leave without making a purchase, every month she would be paying for 98 000 clicks that don’t convert to sales. Annually that would mean 1,18 million contacts, that don’t increase the cash flow. In real life, Jane would have to buy perhaps 70% of her traffic with advertising and the rest would find their way organically (but even that would have to be supported with SEO, which isn’t free either). Jane will quickly learn, that the cost of the increased traffic is easily 20% of the new revenue, and the whole exercise probably a waste of money.
The smaller the store, the easier the merchant believes that they need to increase the number of visitors to increase the number of purchases. The larger the store, the more intensively the merchant monitors the conversion rate, meaning the 2% in this case. Consults specialized in conversion optimization believe of course, that this rate can be improved by making the purchase decision journey easier and more appealing. This is naturally worth doing, but even marginal improvement might take a lot of work, and also, in this case, the limit of cost-effectiveness is easily overstepped.
The rate can be boosted with a chat or a chatbot. In theory, they should fill the void of human interaction, which is usually missing from e-stores. This goal has a built-in conflict. If there is a human behind the chat, the cost of human labor that was supposed to decrease will return. If we continue with our example of Jane and her goal revenue of a million euros, I would say that, given the option, possibly 10% of store visitors would be interested in chatting. When the labor is done with human effort, 100 000 monthly visitors would convert to 10 000 chat discussions. You would need to hire 10 people to take care of that workload. Your salary costs just increased by 40 000 euros a month, almost half a million euros a year. To outsource the chat, it would cost €4 per discussion, which isn’t cheap either. Therefore, sales should grow from a million to a whole other scale for it to make sense to hire a chat crew in the first place.
For that reason, Jane is offered an AI chatbot next. The problem is that an AI chatbot doesn’t give out an intelligent impression before it’s been taught. The bot will handle simple customer service encounters and contact requests pretty smoothly but if there are thousands of items in the inventory, it will take an immense amount of time to teach the qualities and possible customer questions to it. And even then, the bot will get overwhelmed with several questions, such as “If I smash this to the wall, will it break?”
The problem lies in the fact that an average bot doesn’t retrieve its information from the existing database but creates its own. The situation is similar to one where a sales rep should provide customer service without ever having a look into the store’s shelves or reading a single product description: “Thanks for stopping by at our home appliance store. I have no idea what we are selling, but based on your questions, I will try to learn a little more about the industry.”
A groundbreaking solution is to combine AI with the product database and make it the info desk of the store, not just a sales rep. Onboarding is quick because the machine doesn’t need to learn everything as it can retrieve whatever information is needed from the database. In addition to finding the required information – “the clay flowerpot breaks easily, the plastic one doesn’t” – it can offer alternative and additional products, personalized discounts, and other carrots. On top of the product information, an AI-driven shopping assistant utilizes recommendation algorithms, in other words, it produces personalized product recommendations based on, for example, what others have looked and purchased.
The difference between a ”traditional” customer service bot and a shopping assistant that is linked to the product database is approximately the same as between a summer intern and an experienced sales rep. If the conversion rate of Jane’s customers will increase from two to three percent by utilizing a shopping assistant and the average purchase grows from forty to sixty euros with the help of additional sales, she will now only need 47 000 visitors a month, of which 1400 will complete their purchase. Instead of profitability getting out of Jane’s reach, it has now settled in.