The Future of Digital Workers: Next-Generation Sales Strategies Pioneered by AI Agents

In recent years, innovative technologies in the AI field, such as Large Language Models (LLMs) and agent frameworks, have emerged one after another, leading to significant advancements in areas like business automation and sales support. Among these, the concept known as "Digital Workers" is garnering attention. This refers to a system that incorporates AI that works in an "agentic" manner, executing, learning, and improving tasks just like a real human to contribute to specific business outcomes.

In this article, based on the theme "The Future of Digital Workers," we will explain the importance of agent technology, concrete use cases, and key points for implementation and design in an easy-to-understand manner. In particular, while covering cutting-edge topics such as voice agents that support phone calls and multi-product strategies, we will introduce content that tends to be technical by replacing it with familiar examples as much as possible.

1. What is Agentic?

1-1. Definition of Agents and Agentic Experience

The term "Agentic" has been heard frequently in the recent AI field. Broadly speaking, an "agent" refers to a system that not only answers inquiries but also has the ability to "plan, reason, reflect, and learn continuously." A key feature is that they "think" and improve, much like a human does when tackling a new problem.

For example, one speaker stated the following:

For complex tasks that could not be handled by conventional "rule-based" or "just a combination of pipelines" automation, agentic systems try to find answers while exploring. That is where the great potential lies.

1-2. Agents as Digital Workers

In business areas such as sales activities, "Digital Workers" utilizing agent technology are receiving particular attention. Digital workers are expected to aim for results similar to humans and to automatically and flexibly handle diverse tasks, from sending emails to handling phone calls and gathering information via web scraping. However, agent technology is not a panacea, and it is necessary to wisely determine "what to execute with agents and what to execute with existing mechanisms."

In this way, the idea of clearly separating agent-based and non-agent-based tasks is being adopted in the field.

2. B2B Sales Transformed by Digital Workers

2-1. AI Agents Supporting Sales Teams: "Alice" and "Mike"

In B2B sales, a wide range of tasks occur, from acquiring customer lead information to approaching via email or phone, arranging business meetings, and following up. Introducing AI agents here significantly changes workflows, for example, as follows:

1. Discovery of Potential Customers (Leads)

   • Automatically identify website visitors and match them with CRM (Customer Relationship Management) data to identify high-priority customers.

2. Sending Personalized Emails

   • Agents crawl company sites and social media to generate highly personalized emails based on the other party's needs and background.

3. Follow-up via Phone Calls

   • Voice agents make calls to handle questions and even coordinate business meetings.

   • Multilingual support is possible even if the customer uses a language other than English.

As a concrete example, "Alice" is an agent that automates sales activities via email and text. On the other hand, "Mike" is a voice agent specialized in phone calls, handling inbound and outbound calls 24/7 without breaks.

By utilizing these digital workers, sales representatives can not only reduce a significant amount of manual work but also perform follow-ups without missing the right timing.

2-2. The Difficulty of Utilizing Agents in B2B

On the other hand, utilizing agents for B2B is not so simple. For example, the "criteria for identifying high-quality leads" differ from company to company, and "what constitutes high-quality content" is highly subjective. Furthermore, voice agents face the following challenges:

• Recognition accuracy of voice models: Even just having different microphone environments, such as headsets or AirPods, leads to differences in the quality of voice input.

• Natural interaction: It is highly difficult to reproduce human-like turn-taking (the timing of speaker changes) and the use of pauses.

• Multilingual support and latency reduction: It is necessary to support various languages and accents while achieving response speeds that approach real-time conversation.

As such, to fully entrust B2B sales tasks to AI, it is essential to have a design that deeply understands the entire business workflow, rather than just relying on the performance of a language model.

3. Technical Infrastructure and Multi-Model Strategy

3-1. Rapidly Changing AI Models

Large language models like GPT and Claude are currently in a period of rapid transformation, with new versions being released on a monthly basis. Furthermore, new models and services are appearing one after another in the fields of speech recognition and speech synthesis. Companies developing digital workers must optimally combine these models on a task-by-task basis and repeatedly conduct evaluation and replacement tests.

Additionally, architectures that use different models depending on sub-tasks within the same agent, such as for "email text generation" or "tool calling," are becoming common.

3-2. Utilization of External Services and Platform Design

Behind the scenes of digital workers, integrations with numerous APIs and third-party services are at work. This is because it is necessary to incorporate various functions, including voice calls, web scraping, security, and data warehouses.

However, this increases the number of dependent services, creating a risk that if one fails, it could trigger a chain reaction of problems. Therefore, a platform design that establishes redundancy and monitoring systems, allowing for failure switching at a level invisible to the customer, is required.

4. The Challenge of Multi-Product Deployment

4-1. The Aim of Launching 'Alice' and 'Mike' Early

Normally, it is a significant resource burden for a startup to run multiple products simultaneously. However, since sales operations involve diverse communication methods such as email, text, phone, and social media, the channels that companies prioritize vary. Therefore, the strategy of providing both the email-centric 'Alice' and the voice-centric 'Mike' early on is aimed at achieving 'antifragility'—the ability to adapt regardless of which becomes mainstream.

4-2. The Dilemma of Rebuilding and Upgrading

Because the evolution of the AI field is so rapid, there are many cases where one is forced to make the decision to "rebuild from scratch" a system that was already somewhat complete. The more users an old system has, the more technical and support costs arise during the transition period.

To address this dilemma, many companies get by with a method where "one team maintains the old product while another team works on a crash project to build the new architecture." In many cases, it is necessary to eventually "rip off the bandage" by prioritizing the new product, and the timing and speed of decision-making determine competitiveness.

5. The Importance of Design and User Experience

5-1. Visualization and Feedback Loops

One of the difficulties with agents is that their actions tend to become a "black box." If users cannot see why a specific sentence was generated or why a call was made at that particular time, it can create distrust.

For example, it is important to devise ways to clearly display information obtained by the agent crawling the web or data referenced to generate email text.

5-2. Designing Decision-Making Flows

Furthermore, the design of the "decision-making flow" beyond the user interface (UI) level is key. If you fully anticipate how an agent will combine multiple tasks and how it will prioritize them during the design phase, you can avoid unnecessary confusion in later operations.

AI agents and digital workers have the potential to spread not only to sales but also to various operations such as customer support and human resources management. However, it is not a case of "automating everything is good"; it is becoming more important to judge which areas and tasks should be turned into agents and where one should stick to conventional systems or human-centered approaches.

Especially in B2B sales, understanding the "complex workflows of each client company" and how to produce "truly effective outcomes" there is a major point that determines the success or failure of introducing digital workers.
In addition, new technologies such as voice agents are predicted to continue to evolve, and significant reductions in latency and natural dialogue control will likely be remarkably improved. On the other hand, the need for new security measures and privacy compliance will also increase.

Ultimately, what is important is an attitude that prioritizes "customer outcomes" above all else. By flexibly verifying every time a new model or service appears and boldly rebuilding if necessary, the true "digital worker" will likely become widespread once team structures that support both "speed and quality" and designs that users can trust are in place.

Related Articles