In my previous article — Become Amazon in your space with AI/ML: What B2B must learn from B2C — I discussed how, despite the huge promise of B2B commerce, this space faces significant challenges for both pre-sales and post-sales due to limitations of existing technologies like “search.”
In this article, let me start by first giving you a preview of the advancements in AI/ML technologies and then discuss how product discovery on your site can be easy and effortless for your customers with next-generation self-service and a powerful agent-assistant tool for your support staff. The result will be increased sales and strong, loyal relationship with customers.
Advancements in AI/ML – Deep Learning, Natural Language Processing (NLP) and Computer Vision (CV)
Deep learning is part of a broader family of machine learning techniques that uses large neural networks trained with enormous datasets for a given space, to understand patterns in data. Once trained, a network can interpret, with high accuracy, complex sentences written in natural language processing (NLP) or identify objects and patterns in an image via computer vision (CV). These technologies can be applied to bring automation to a wide range of business use cases, resulting in increased sales, reduced cost, and a drastic increase in productivity.
Deep learning techniques have become increasingly more sophisticated in the last three years due to advancements in hardware, such as graphical processing units (GPUs) and tensor processing units (TPUs), that allow you to quickly perform the very large matrix operations at the heart of neural network computations. For example, the GPT3 model from OpenAI is trained on about 45 Terabytes of textual data and has 175 billion parameters to perform natural language tasks rivaling human performance – e.g., it can write articles, perform a sophisticated automated conversation, and even detect the similarity between two sections of text.
Previous deep learning NLP techniques were based on RNN (recurring neural networks), a technique that had the problem of forgetting the past context – e.g., when you summarize an article you need to know the previous context. RNN would forget the previous context. This problem has now been solved by a new category of neural networks called transformers, which has led to a significant boost in the accuracy of text interpretation-based tasks such as text summarization, performing conversation, Q&A, and more. Examples of transformers includes BERT and GPT, which also require less training than RNN-based implementations. Transformers are also used for computer vision (CV), which is a field of artificial intelligence that enables computers to derive meaningful information from digital images.
Solving Product Discovery when a customer is looking to buy
Use Case 1: Customer is looking to buy a product with a complex spec
Consider a B2B site with over a million products, with each product having a wide range of features or attributes. The reference data probably includes a structured database with the products and their features, with each product having its own data sheets, instruction sheets, user guides and more. In addition, there is pricing, inventory, and order status that come from applications from where data needs to be accessed in real time.
Consider a customer using a self-service chatbot looking for a product with a request like this:
“I need a Hermetic thermostat with operating temperature between 100 and 300 C, pin count of 4 and that is EU RoHS compliant.”
Note: here, Hermetic refers to the manufacturer type.
To provide the answer, you need a powerful chatbot (or even voicebot) that can have a broad range of conversations with the customer, figure out in real time if there is any information missing from the query, and even have a sub-dialog if needed. Once it has the complete query, it should be able to get the answers directly from your reference data.
To provide answers from your reference data, you need an AI/ML Answer Engine on the back end that can take a complex query, interpret it, and then get answers (not documents) directly from the reference documentation — structured, un-structured, and real-time.
The chatbot should even have a memory of previous conversations to implicitly complete missing information in a query. This way, customers can easily help themselves to discover the right products and get all the necessary information they need to make a decision on the purchase — right then.
This level of sophistication is hard to provide with “search.” As a result, today you probably have live chat, voice- or email-based support — where the support engineer spends a lot of time using search and other applications to find the answers. Sometimes this could take 15 to 20 minutes. Sometimes they need to seek the help of an expert.
Your customers are busy professionals; if they don’t get the right information easily, they will go somewhere else.
Note: for self-service, you can use chat or voice or upgrade your existing search with an “AI-based computational search.”
Use Case 2: Customer is looking for product features, configuration, installation and more
As I mentioned in the last section, let’s say you have a million products on your site. For each product, there are datasheets, instruction sheets, user guides, product guides, and marketing material as PDF or HTML documents. Some of these documents were prepared by your internal teams, while others may be from partners, manufacturers, or distributors. And they get updated regularly.
If the customer has questions on features, configuration, or installation, they should be able to use a self-service chatbot or a computational search interface and ask any question to get the right answer from your reference documents. They should also have a link to the exact section or page in the document (datasheet, user guide, instruction sheet, etc.) for further reference.
If they need help with troubleshooting a device, they should be able to take a picture of, say, the LED indicators on the device and input that image in the chatbot. Give them the answers, not documents to read.
Use Case 3: Customer sends email with a question; reply automatically
In B2B commerce sites, a lot of times your customers and distributors may prefer to send you an email with their questions — not only on products specs and pricing, but also on the status of an order, order replacement, or to find the location of a distributor in their area. With AI, it is now possible to interpret the email (subject and description), understand what the customer is looking for, and then to reply with an answer automatically.
Use the latest AI/ML technologies to provide customers with exactly the right products based on their specific requirements, the right answers to installation and configuration, or even help them troubleshoot issues themselves with high accuracy, using self-service, or with the help of a live agent who is equipped with AI-based productivity tools. The result will be long lasting customer loyalty and high employee retention.
In the next article I will be writing about post-sales support — how you can provide a highly sophisticated service to customers post purchase.
About the Author:
Prosenjit Sen is a serial entrepreneur and currently the CEO of Quark.ai, an “autonomous support” platform that uses deep learning, natural language processing (NLP), and computer vision (CV) to bring automation to sales support and field support. He was previously employee No. 5 as part of the founding team of Informatica, a pioneer in online data integration technology. Prosenjit is a mentor for the Alchemist Accelerator and the Bay Area IIT Startups Accelerator. And he is the co-author of the book “RFID for Energy & Utility Industries.”
Sign up for a complimentary subscription to Digital Commerce 360 B2B News, published 4x/week. It covers technology and business trends in the growing B2B ecommerce industry. Contact editor Paul Demery at [email protected] and follow him on Twitter @pdemery.