Why Reading eCommerce Product Reviews Still Requires Brands’ Human Intelligence?
Tons of product reviews are written and shared by consumers in online stores every single hour and day. Discovering what they say and how they feel about your products in a reasonably short-enough time seems hard to grasp. The good news is that a series of AI-based tools are claiming that this is finally possible:
You can now hand over this job to smart algorithms while engaging yourself into work where your time, mind and spirit are put at better and more enjoyable use other than reading product reviews, one at a time.
But can you fully rely on how these machine learning powered algorithms assist you and what they deliver to you on your quest towards:
- building new generations of better products
- identifying exactly what creates troubles for consumers
- having more satisfied customers and
- a vibe in the online stores that determines more and more visitors to loading up their shopping carts with your products instead of competitors’?
Nowadays, reading aggregated product reviews from online stores, discovering most talked about relevant topics and classifying them, measuring consumer sentiment around the most sought-after product features and even replying to customer reviews are more and more desired by brands across various consumer goods sectors and expected to be produced automatically, without even a mouse click.
Leaving the voice of customers entirely in the hands of machine learning is regarded as the salvation for many digital marketers, eCommerce professionals and product managers dealing with increased amounts of sources where consumers speak out loud about their products. Everyone hopes to find the holy grail of consumer feedback analysis delivered effortlessly.
However, even if machine learning (ML) has been evolving, not all ML-powered results are the straightforward answers to brands’ needs when interacting with customers’ voice in online stores.
There are 2 main situations when dealing with eCommerce product reviews that should not be replaced by machines:
#1 – When replying to customer reviews in online stores
Nobody on the receiving side appreciates automated replies, further than a receipt confirmation.
There is a high degree of irritation on our side as consumers when it comes to the now famous automated replies that some brands send out to customers and further take them through elaborated and what it seems endless self-service loops towards a solution to their complaints. In some cases, that full automation is part of a well-thought strategy, as this Harvard Business Review article explains, meant to determine consumers to give up on the way towards solving their claims. This approach might have been proving its viability, as the article points out, in less competitive industries like airlines, internet, cable, and telco, meaning that alienating unhappy customers has not made yet a visible impact on the loss of their market share and profitability.
At it looks like, this approach has worked well so far for a narrowed group of market-dominant service providers, and on particular customer support channels – usually hidden from the public eye or not consulted at the moment of purchase.
However, the situation is different when complaints take the form of product reviews and are displayed publicly in renowned global and local online stores.
Online product reviews call for bringing in radical transparency and authenticity to customer support due to their unique characteristics:
- they are public, visible to everyone searching the store
- they are always actual, flowing in on an on-going basis
- they are mainly from existing customers
- they have the highest influence degree on potential new customers compared to any other online source of feedback
- they have amazing power over any other previous research consumers might have undertaken because they “talk” directly to them, one step before the checkout
- it’s impossible not to notice their volume, histogram, and ratings
- they attract other store visitors’ comments and questions
- when they remain unanswered by brands, that shows a certain lack of care, unwanted at the selling point
While at FeedCheck.co we often meet marketers who look for automated review reply capabilities, we also often hear them wanting to diversify the way they respond to customers online, correct misunderstandings they sometimes have and present solutions to help them as well as future consumers.
Adopting the common approaches used by large market-dominant brands, as highlighted by the HBR article, and thus multiplying them and make them the customer support norm almost everywhere, is not that a good strategy, especially for the eCommerce space.
#2 – When trying to understand the exact context of complaints and make good sense of it
Which words consumers use to describe their experiences with products and brands can be identified and measured by machine learning but putting them into the actual contexts is far from reach without human intelligence assistance.
This is often a surprise for many people who have never gone deep into this matter to see the entire process behind. And that would be just fine if these tools would work as people imagine. Only that they’re not there yet, while some people’s assumption is.
Review text analysis with machine learning gives a brand a series of statistics over so many words consumers use in their feedback. The intelligence component of these analytics does not understand the text per se like we do, as many people wrongly assume, but only identifies the noons, verbs, attributes and adjectives through what is called part-of-speech tagging process. The results are only estimates based on the data previously used in the training process of the machines.
Based on that, and for refined results, human intelligence comes into play and defines what’s needed and what’s not for a market niche’s specific product review analysis.
While the most popular industry where ML has been adopted is the hospitality industry, the consumer goods sectors carry on a significant variety factor that requires a niche specific setup. Taking into account that product reviews are still in their infancy time, expecting high performance from a ML solution in the product reviews space is only a nice dream at this moment. Unless you want to be delivered a set of charts and reports that you can carefully review, take note of their existence and then start questioning every single indicator trying to find meaning for it and how it will help you further.
After getting a decent understanding of the results, you may still remain in a foggy area where you can’t really tell much about the actual context in which your products had been reviewed.
In order to achieve that you need to actually read your reviews and get in touch with your consumers’ real-life stories from using your products.
Another notable aspect is the volume of your product reviews. ML typically works on large data sets. The lower the volume of reviews, the higher the isolated appearance of various words and key phrases that correlated with the volume of reviews make the qualitative analysis little relevant for further actions.
In the meantime, until further progress of the contextual text analysis, FeedCheck offers a set of capabilities that help brands navigate through and explore their product reviews for various business needs, from measuring how products’ value propositions have been received and understood by consumers to evaluating consumer sentiment for various product features over time.
If you want to know how we could help your product marketing, sales, merchandising and other brand functions to make the best use of their product reviews across your eCommerce space, get in touch with us.
Image by https://pixabay.com/users/geralt-9301/