SoftBank Resolves 70% of Website Inquiries

Summary

SoftBank put an AI agent at the first point of contact on its corporate website.
Continuous improvement raised the self-resolution rate from about 50% to 70%.
When routine inquiries are automated, sales staff can focus on complex consultations and proposals.

A division of labor that began at a website flooded with first questions

When a business customer visits the website of a telecom or IT company, what they want is clear. They want to quickly check the pricing, service specifications, or contract terms they need right now. But when a company is large and offers many services, customers easily get lost in a complex web page. In the end, they look for the fastest way to reach a contact desk and leave a question, and the burden falls entirely on the sales staff.

The Customer Growth Division, which handles marketing and sales inside the corporate business unit of SoftBank Corp., set out to solve exactly this problem. The division does business with about 3.4 million small and medium-sized business (SMB) customers across Japan. The needs of SMB customers vary widely by industry and size. Sales staff have to understand each customer’s business environment deeply and carry out tailored proposals and complex consultations.

But on the front line, a large volume of routine inquiries poured in, asking about a product’s basic specs, list prices, or simple contract procedures. The first response matters a great deal to the customer, so demands grew, from inside and outside, to give answers that are more consistent, of higher quality, and above all faster. The division had to fix an inefficiency: time that should go to the deep proposals that really matter was tied up in basic guidance.

This post looks at how SoftBank redesigned its first point of contact with customers, based on a customer story that Sierra, an AI agent platform company, published on July 14, 2026.

Three reasons a more refined system was needed

SoftBank had not left its website without any automation. Earlier, it had put an AI chatbot and an AI search feature on the corporate customer website to help customers find information on their own. But the existing tools alone were not enough to satisfy what the front line needed.

The Customer Growth Division had three clear goals in deciding to switch to a conversational AI agent. The first was to build an environment that gives an answer right away, at the very moment a customer needs the information. The second was to set up a system that accurately measures the self-resolution rate, which shows whether a customer solved their problem all the way through without help from a support representative or a salesperson. The third was to use this to raise the overall level of customer experience one step further.

The division needed a tool that understands context more precisely and wraps up a problem through natural conversation.

For its first target, the Customer Growth Division chose the first response to SMB customers visiting the corporate customer website (https://www.softbank.jp/business/). It picked this spot first among many business areas for two reasons. It was the channel where customer contact happens most often, and it was the path on which customers and the organization could feel the effect fastest once the customer experience improved.

Starting from website knowledge, and sharpening intelligence in the field

The practical worry when it decided to bring in the new AI agent was the time it takes to learn. SoftBank’s corporate business covers a wide range of services, so there is a huge amount of product knowledge to learn. It takes a new employee a considerable amount of time to take it all in and handle customers well.

Abe, of Section 2 of the Marketing Strategy Department in the Marketing Division of the Customer Growth Division, shortened this preparation period by using the Sierra platform. Abe described the process like this.

“Sierra feels like a highly polished AI agent platform that is both easy to implement and easy to improve over time. In our case, by leveraging information from our website during the initial setup, we were able to train the AI agent — in a short period of time — the product knowledge and FAQs that would take a human considerable time to learn.”

In practice, the team likely built the first knowledge base by connecting, as knowledge sources, the product brochures, price tables, and frequently asked questions (FAQ) pages that were already public and verified on the softbank.jp/business domain. They did not rewrite new manuals from scratch. Because they used website resources that were already approved, a quick rollout was possible.

The way it was run after the agent went live was also systematic. Rather than expecting the system to work perfectly the moment it was released, the team kept adding to its answers while watching the question data that real customers asked. Abe added the following about this process.

“SoftBank’s corporate business covers a wide range of services, and the self-resolution rate — which started at around 50% at launch — has now risen to 70%, thanks to iterative improvements to help the AI agent better understand customer intent and the relevant service.”

The specific launch date was not disclosed, but the initial self-resolution rate was around 50%. That means about half of the inquiries were still not resolved. When a business customer simply asked, “How much does it cost?”, if it was not clear which service they meant, a wrong answer would likely have gone out.

Using Sierra, the team picked out the questions that were not being answered well and quickly worked out what information was missing. Then, as they refined how the agent asks about and listens for the customer’s real intent and the service in question, the self-resolution rate climbed to 70%.

The boundary where people and machines hand work over

An AI agent cannot handle every question, and it should not try to force every conversation to a close. SoftBank drew a clear boundary so that the agent would not hold on to questions it cannot resolve.

When a question comes up that the agent cannot resolve, the agent naturally guides the customer to an inquiry form. The line was probably cases that public information alone cannot solve, such as a custom quote or individual contract terms. When the customer submits the details through the inquiry form, a salesperson then contacts them directly and takes over the consultation.

Thanks to this division of labor, the inquiries that now reach sales staff have been narrowed to complex, tailored consultations and matters that need individual handling. In practice, a summary of the conversation with the agent and the specific service information were probably passed along together on the inquiry form, reducing the hassle of a salesperson asking the customer the same questions again from the start.

The agent handles about 100 inquiries a day. Of these, 70% end with the customer finding the information they need in the conversation with the agent and finishing the matter on their own. Only the complex inquiries, or those needing individual handling, go through the inquiry form to a human salesperson.

The key metric here, the 70% self-resolution rate, is not a score that people gave by reading the transcripts one by one. It is a ratio that Sierra’s monitoring feature calculated by analyzing the whole context of each conversation and having AI judge automatically whether the customer’s problem was resolved within it. To keep this automatic judgment reliable, the operations team probably pulled samples regularly and had people review them, adjusting the criteria. Because such an objective measurement system stood behind it, the organization could trust the results and keep improving.

People who regained the essence of sales, and the expansion ahead

As repetitive basic inquiries were filtered out at the front, the way the sales organization works changed too. With 70 routine inquiries out of about 100 a day resolved automatically, salespeople no longer had to break up their day answering simple confirmation questions.

Tajima, the Director of the Marketing Strategy Department in the Customer Growth Division, assessed the value of this change like this.

“We expect that as the self-resolution rate of the AI agent continues to improve, customers will be able to access the information they need even more quickly. Sales representatives will also be able to dedicate more time to addressing individual customer challenges and advancing proposals. We see this as an initiative that achieves both improved customer experience and greater sales productivity, and we intend to continue developing it.”

Customers check the prices and specifications they want on the website right away, without waiting for a salesperson’s reply. Customers who need deeper consulting, on the other hand, get a much fuller and more focused proposal from a specialist salesperson who is not worn out by basic questions. It shows that the purpose of bringing in the AI agent was to raise the quality of the customer experience and to help people focus on the in-depth proposal work where they create the most value.

Building on this success, SoftBank plans to widen the deployment of AI agents beyond the corporate website to a variety of customer touchpoints. Its next goal is to complete a consistent support system in which customers can reach accurate information at the moment they need it, whichever route they come by.

Three principles other companies should take from this case

The SoftBank case shows that you can create real improvement in work without changing a huge call center infrastructure all at once. Business leaders thinking about the digital transformation of customer response or sales organizations can find three practical principles in it.

First, start at the narrow entrance where customer contact is most frequent and the effect of improvement can be felt right away. Many organizations try to automate the whole vast range of internal work and never get started. SoftBank chose a clear target: the website inquiry desk, where customers arrive first, and within it the routine questions about products and prices. Because the target was narrow and sharp, it could build the system in a short time on knowledge already organized on the website.

Second, what matters more than a perfect launch is the cycle of improvement after launch. SoftBank also did not record a 70% resolution rate from the start. It started from a half success of about 50%, but by looking continuously at failed conversations and refining the way of asking about customer intent, it raised the metric to 70%. What decides success is having an operating system in which you can easily see which questions the agent is weak on and add knowledge.

Third, clear measurement metrics and a reallocation of sales time. SoftBank made the self-resolution rate, in which AI judges whether the problem was solved in the conversation, its key metric. And its aim was to turn the time that sales staff freed from routine inquiries toward high-value work, namely complex tailored proposals. Bringing in an AI agent is not about bringing in technology. It is about redesigning how the organization spends its time.

Sources

Sierra — How SoftBank’s Customer Growth Division delivers fast, high-quality customer service with AI
https://sierra.ai/customers/softbank
Published: 2026-07-14, accessed 2026-09-24
Sierra — Japanese edition of the same customer story
https://sierra.ai/jp/customers/softbank
Published: 2026-07-14, accessed 2026-09-24