I start my day by asking ChatGPT questions about an upcoming trip I am
planning, including where I am heading, with whom I am traveling, and the
expected weather. ChatGPT provides an itinerary of where to visit, places to
eat, and what to bring on my trip.
Five minutes later, I visit several ecommerce websites to find and buy the
recommended items. The search box barely fits two keywords, and after
several tries, I give up and visit a second site and then a third.
I run out of time and VPN into my client’s network to begin my work. Today,
I am reviewing how the customer support teams respond to product information
inquiries.
It turns out that only 30 percent of the questions can be answered by reps
when they search the CRM. When they don’t find the answer, they scavenger
hunt through Sharepoint sites, Slack feeds, and multiple content management
systems to find answers.
ChatGPT is a game-changer for search
Compare ChatGPT’s natural language capabilities and the verbose responses
against the difficulties customers and employees face when querying for
information. Even before ChatGPT skyrocketed to fame, I advised product,
technology, and innovation leaders on how
AI search is a digital transformation force multiplier
and why
CIOs should invest in AI search.
But ChatGPT raises the stakes with its capabilities and media attention, and
I believe
ChatGPT will energize the consumerization of natural language search.
We may not be able to offer ChatGPT-like experiences to customers and
employees today, but you should review search platforms built with large
language models and
relevance generative answering capabilities
to stay ahead of people’s expectations.
Here’s how I unpack people’s ChatGPT-driven expectations and the five AI
search capabilities customers and employees will expect in search
experiences.
1. Natural language search
Customer and employee journeys start with how they browse results using
faceted search
capabilities and what they enter into keyword search boxes. I first
implemented faceted search back in 1997 while CTO for a SaaS company serving
the newspaper industry. While this capability has advanced significantly
since then, most websites only support basic keyword searching
capabilities.
I expect customers will expect
natural language search
capabilities with ecommerce customers asking questions like, “What are the
hot t-shirts teenagers are buying for the summer,” or banking customers
entering, “What low-risk small-cap funds should I consider if I plan to
retire in ten years?”
Natural language search can also improve customer support and employee
experiences. SaaS companies should target more customer self-service
capabilities through natural language search to reduce customer service
costs and improve customer experience. Enterprises can
reduce tribal knowledge
and improve productivity by enabling employees to ask questions and find the
information they need to do their jobs faster.
2. Case classification
Let’s consider the use case where a field or remote worker needs technical
help and opens a support ticket to the IT service management desk. Service
requests are a broad use case, and enterprises often have HR, legal,
facilities, and other support services. Additionally, online forms are the
primary way customers ask questions or escalate issues to B2C and B2B
businesses.
Such forms often require comprehensive information to help prioritize,
route, and service tickets efficiently. But having too many fields on the
form is an end-user nightmare, and many will abandon their quest for help
when the process appears too daunting. This is a worst-case scenario for
service desks because they lose the opportunity to address the customer’s or
employee’s needs.
Case classification
uses machine learning and natural language processing to bridge the gap.
Forms use short open text fields instead of dozens of dropdowns, and the AI
displays matching categories and terms for the end-user to validate.
Case classification and natural language search are two ways to reduce
engagement friction and increase the likelihood of people asking questions.
3. Respond to search with smart snippets
After people ask questions, what do they want? Answers!
Customers and employees will expect generative AI capabilities and long-form
answers in the near future. Today, companies can provide
smart snippets
that match a resulting document’s content with the end-user’s question.
When an employee asks, “What mental health options does my insurance cover,”
they prefer seeing a snippet with the answers and avoiding reading through a
long document. Snippets can be incredibly useful for answering policy
questions or finding the right people to contact when a field or remote
worker has a technical question.
4. Click-worthy recommendations
After a question is answered, most end-users want suggestions on where to go
next to learn more, transact, contact an expert, or take another step in
their journey.
Commodity search indexes address these needs by ranking search results based
on the categories and rules defined by subject matter experts and
implemented by software developers. It’s a brute force and inefficient
development methodology that’s expensive to maintain and often provides
underwhelming results. According to one study,
85 percent report manual tuning is required to improve search result relevancy.
Most users don’t want a stream of results that can be overwhelming and
time-consuming to review. They seek
click-worthy recommendations
using machine learning to predict likely courses of action. These
recommendation engines leverage
search analytics
and context on who the user is, when they are searching, and what they
queried to provide their recommendations.
Returning to my example, an employee asking about mental health coverage
wants their question answered through a snippet. But they should also be
shown links on where to get help and whom to call if they face an emergency.
5. Simple to implement integrations
The first step for enterprise leaders in product, technology, innovation,
and data is to recognize that developing natural language search will
require centralizing access to content and data resources. Today, the CRMs,
CMSs, and other SaaS tools search their siloed data repositories — but
natural language search will need training and connectivity to these primary
data sources. Furthermore, the native keyword search boxes built into these
platforms will need to be replaced with natural language search
interfaces.
Top search platforms
make this easy to implement with out-of-the-box data integrations to common
platforms so that search snippets and recommendations tap into all of the
enterprise’s knowledge. These platforms also include APIs, dev tools, and
low-code dev options for easily embedding natural language search interfaces
across SaaS and applications.
And enterprises need simplicity, especially when customer and employee
expectations increase rapidly, and there’s an opportunity to leapfrog the
competition.
This post is brought to you by Coveo.
The views and opinions expressed herein are those of the author and do
not necessarily represent the views and opinions of Coveo.
