How to Create an AI Secretary: A Real Guide to Automation That Works

When you wake up each Monday morning and check your inbox for 127 unread emails, it can be overwhelming. Half of the emails are unsolicited offers, one-fourth don't need to be sent to you, and the last 30 emails actually require your immediate attention; meanwhile, you have 15 unread texts from Telegram, 3 missed calls to your phone, and a meeting starting in 20 minutes that you totally forgot all about.

The typical way to handle this type of situation is to hire a secretary; however, hiring a secretary works great until your secretary is on vacation, is sick or has been overwhelmed by the entire array of emails. Finding a great secretary who understands the crypto market is a challenge, again!

From my experience working with Web3 companies, I have seen companies with monthly revenue between 10–50 million rubles be overwhelmed with taking care of daily business functions and administrative operations to the point where the founding team spends at least four to five hours a day just organizing meetings, answering repetitive questions, and ensuring that administrative processes are running smoothly. That's four to five hours that should have gone into developing the product.

The real solution is to implement an AI secretary into the organization. An AI secretary is an online virtual assistant that works for you 24/7, never grows tired, never forgets anything, and processes requests faster than any human being ever could. This may sound like science fiction, but the reality of this technology is that it's here now; you can have it up and running without having to hire an entire IT team.

In Summary: What is an AI Secretary?

An AI secretary is an agent developed from a machine learning system that can handle all administrative duties. The calendar management, message response, meeting reminders, and processing of incoming request functions implement existing business systems. The distinguishing characteristic of an AI secretary is that it can create context during communication. For example, if a client sends a request saying, "Please change the meeting date to tomorrow", the AI secretary would find the appropriate meeting, check to see if it's available and present multiple choices to the client all on its own.

A Virtual Assistant can be categorized into three different groups in terms of autonomy:

• Reactive Bots: The most basic level of Virtual Assistants, these bots use a set of predetermined commands and responses to respond to user queries. Reactive bots only perform predefined tasks, such as "Show me my calendar" or "Send me a report."
• Contextual Assistants: These Virtual Assistants take into account the conversation history, allow for an understanding of contextual conversations and provide support for the user based on their needs. A user could request to know "What's my next meeting" and the Contextual Assistant would reply with a list of meetings coming up that are enabled by the integration of GPT and NLP with the user's calendar and email systems.
• Autonomous Agents: These Virtual Assistants are capable of becoming a planner for a particular action. For example, if a user requested that "I would like to set up three client meetings for this week", the Autonomous Agent would automatically check each person's availability and send an invitation to the respective participants.

The following is a list of tasks performed by Virtual Assistants:

• Scheduling: Arranging meetings, keeping calendars in sync, and sending reminders.
• Communication: Generating autoreplies from email and messenger apps and categorizing incoming messages.
• Document Handling: Completing various types of forms, searching for data, and generating summaries of meetings.
• System Integration: Adding new leads to a CRM, updating task assignments, and conducting searches.
• Analytics: Generating reports on number of meetings held, average response times, and identifying areas that may have created delays.

How Does It Work?

The backend technology behind Virtual Assistants is the GPT (Generative Pre-trained Transformer), which is a language model that has been trained on billions of documents. GPT is able to understand natural language queries almost as well as a human being, as well as take into account context, and provide coherent responses to user queries.

ASCN.AI utilizes OpenAI's GPT-4, as well as specialized training models that are focused on Web3 and Business Processes.

According to McKinsey, 60% of the employee's hours are spent performing repeated daily activities. The AI Secretary automates these tasks.

Some of the primary functions of an AI Secretary are:

• Email Management: The AI Secretary sorts through all incoming email messages and classifies them according to their category. The AI Secretary sends a pre-written response if a client requests pricing, provides the price list to the client, and informs the manager to follow up with the client.
• Reminders: The AI Secretary keeps track of the employee's date and time of events in the calendar, keeps track of contract and payment due dates using automated notifications sent via the employee's preferred messaging app (e.g., Telegram, Slack), and reminds employees of upcoming tasks using automated messages.
• Booking: The AI Secretary retrieves ticket prices and hotel availability using API connections from online travel services and sends the employee the option to book an event to their preferred method of communication using automated messages.
• Order Processing: The AI Secretary sorts and organizes requests into categories, takes care of the administrative tasks associated with fulfilling an order request by reassigning the request to the relevant department or manager, and logs the request with a unique tracking number.

One project implemented an AI Secretary and has been able to decrease the order fulfillment time from 4 hours to 12 minutes and increase the order conversion rate by 18% — many clients simply do not have the time to look for competitors during the order fulfillment process.

Automating Document Management

Document Management is a difficult task for many businesses. The majority of businesses have difficulty managing the volume of documents they receive on a daily basis and spend a great deal of time searching for, opening, closing, and storing documents. An AI Secretary removes these tasks from an employee's daily responsibility and allows them to focus on other activities.

Some examples of functionality that can be automated with Document Management:

• Automatically send any correspondence (emails, snail mail, faxes) to employees to a defined email address upon receipt.
• Automatically generate and print monthly invoices for an employee's account upon successful completion of a transaction.
• Automatically generate and print monthly financial statements upon receipt of all required information (address, account number).

Approval of contracts can sometimes take days due to inefficiencies, even though the actual processing time is typically only one hour in length.

AI has capabilities that allow it to do the following:

• Extract data from documents (dates, sums, etc.) into accounting systems, as well as notifying accounting departments.
• Generate documents: contracts or commercial proposals can be created using templates and inserting data from a customer's record within CRM. The AI will create a PDF document and send it for signature.
• Summarise meetings: audio files will be transcribed into written format. Key decisions will be highlighted and separated, as well as the creation of tasks to assign responsibilities.

Research published in the Harvard Business Review states that by automating the document management function, organisations will reduce operational costs between 30–40% and accelerate the approval process between 5 to 7 times.

All-in-One Ecosystem

The AI can serve as a link between all dislocated work platforms. For example, this might include Google Calendar, Microsoft Outlook, Google email, Telegram, CRM (e.g. AmoCRM, Bitrix24 or HubSpot), task managers (e.g. Trello, Asana or Notion) and payment systems.

As an example: a customer requests a consultation within Telegram, and the AI's capabilities are activated by reviewing the manager's Google calendar and proposing 3 time slots that the manager is available. After the client has selected a time, the AI creates an 'event' on the manager's calendar, creates the client's contact record in the CRM and tags it as 'cold lead.' The AI is also able to assign a task to the manager to follow up with the client and remind both parties one hour before the scheduled meeting.

Main types of API integrations:

• Google Calendar API (for calendar integration)
• Microsoft Graph API (for email with Microsoft Outlook)
• AmoCRM, Bitrix24, HubSpot (for customer records and CRM management)
• Telegram Bot API, WhatsApp Business API (for messaging)
• Stripe, Robokassa and cryptocurrency exchange APIs (for payment processing)

All data flows through a 'chain'. A manager closes a transaction → then, the AI generates and issues invoices for payment. Once the customer makes payment, a task is created and the logistics department is notified. No manual steps are required in this process.

How to Build an AI Secretary: A Complete Stepwise Plan

Selection of a Platform

The initial query that should be considered is whether to create a platform from the bottom up or to use a pre-built platform.

Building from scratch would require ML Engineers, Backend Developers, DevOps — cost of $3–$5 million and time of 6 months to 1 year. While these two aspects make this way too expensive and time consuming for the majority of companies, it makes much more sense to select a NoCode/LowCode platform with preselected options:

• OpenAI GPT + custom middleware. Use the OpenAI API and develop the logic yourself. Flexible but requires programming knowledge.
• Google Dialogflow. A visual builder interface, Google Calendar integration. Good for easy cases but limited customisation.
• Rasa (Open Source). Complete control over NLP. Requires programming skills.
• ASCN.AI NoCode. Over 100 integration choices, a visual builder interface, built-in GPT. Will launch in 2–3 days with no programming.

My recommendation: for Startups/Small Business — ASCN.AI NoCode; for large companies — OpenAI + custom middleware.

Defining Scenarios

Do not try to create all automated systems at once; otherwise, the system will get overloaded and will be ineffective.

Start with the three most important tasks you want automated:

• 1st Level Technical Support. Most Frequently Asked Questions (FAQ) account for about 70% of your requests. Technical Support will be directed to an operator for assistance.
• Calendar Management. Reduces time spent to organise meetings through automated coordination.
• Response Processing. For every request received: check availability, determine estimated completion period, create lead (request), and notify manager.

Write out for every scenario:

• What Triggers it — (e.g., a message or event)
• What Input Data is Required — (e.g., contact, request number, email, etc.)
• How Decision Logic is Applied — (e.g., if an item is in stock, send a price list)
• How Actions are Performed — (e.g., sending of messages or database entries)
• What Output Data is Produced — (e.g., confirmation, request number, etc.)

Configuring and Training the Model

Language Models that utilise the ASCN.AI platform have already been trained. The Agent has specific guidelines to follow based on how we want our business to operate. For example:

"You are an AI Secretary helping to manage the scheduling of manager meetings for Company X. You will facilitate your clients by responding to their requests and reminders professionally, while also ensuring that all of the information needed to schedule a meeting is clear and accurate. If there are any questions about the client's request, you will request clarification from the client. Lastly, if you are not able to complete a task on your own that the client has requested, you will direct them to a Human (a member of the Company)."

The AI Secretary uses an API to communicate with calendar services (Google Calendar API or Microsoft Graph API), CRM systems, and payment platforms. All tokens and keys needed to access these services are securely stored in the "Secrets" section of the system.

Testing and Launch

1. Unit Testing — Testing the functionality of each Block of Code in isolation.
2. Integration Tests — Ensuring that the tests conducted in Unit Tests perform as expected when all Blocks of Code are combined together.
3. Pilot Testing — Testing the overall system with a limited number of users.
4. Iterative Launch — Incorporating additional Features into the system as needed.
5. Monitoring — Collecting and Reviewing Logs to assist in identifying the Reasons for Requests.

Real-World Examples

Client #1 — Crypto Project Support

Situation: Our Crypto Exchange receives 500+ Requests Daily, 60% of which are frequently asked questions. Time to respond to Requests is 4–6 Hours.

Solution: The AI Secretary from ASCN.AI was integrated with Telegram and a Knowledge Base to provide a faster response time.

• Time needed for an initial response decreased from four hours to thirty seconds.
• Decrease of 65% in operator load due to automation.
• Conversion rates from registration to purchase increased by 22%.

Client #2 — Marketing Agency Calendar Management

The marketing agency utilized AI to significantly decrease the number of hours it took to organize meetings among 15 employees. Previously, they would spend two to three hours each day coordinating before each meeting, but after implementing the AI secretary, it took only 15 minutes to organize. On top of that, they reduced the number of times they needed to reschedule meetings by 80%, thus saving them over 10 hours per week.

Client #3 — Crypto Education Platform Lead Generation

An educational platform focused on cryptocurrency wanted to create leads without having to have their own call center. The educational platform implemented an ASCN.AI Telegram bot to have conversations, qualify leads and sell courses without needing a human present.

• Conversion rate increased from 7% to 18% on dialogues to payments.
• Lead cost decreased from 1200₽ down to 340₽.
• The bot is capable of conducting up to 300 dialogues per day without human involvement.

Popular Market Solutions

There are numerous market solutions available; however, only a few provide the degree of automation that ASCN.AI does.

• Microsoft Copilot is integrated into Office 365 and is costly.
• Notion AI has limited integrations and serves only as an assistant to the document creation process.
• Clara serves as an email-based meeting scheduler and has highly limited functionality.
• ASCN.AI NoCode allows the user to create customized AI agents that can perform any task and connects to numerous other external systems.

Best Practices for Implementation

Customization to Meet Business Needs

• Adapt to the company's industry and specific location, utilizing their terminology throughout the knowledge base.
• Configure your AI to use a tone of voice that is consistent with the company's culture and internal processes.
• Put employees' personal preferences and schedules into consideration when implementing AI.

Data Security

• Employ encryption keys using the Secrets Section of ASCN.AI to store all API keys securely.
• Only provide the minimum amount of access to databases and systems as possible.
• Only grant access to people who need the ability to use the AI Secretary.
• Log all actions for later audit and review.
• Restrict access using Parameterized Queries.
• Adhere to legal requirements by obtaining Consent and also enable Data Deletion.

Team Training

• Train the team to ensure proper usage and to learn how to create queries and the proper phrases used.
• Provide advanced training on Variables and how to set up workflows.
• Provide expert training to assign responsibility for monitoring and retraining.
• Offer a reporting channel to provide feedback and report errors.

FAQ

What do you need to create an AI secretary?
You will need a Platform (either ASCN.AI or Dialogflow), Service API Keys, a description of the Scenario and a Knowledge Base. If you are using ASCN.AI, you can have your AI Secretary up and running in 2–4 hours and the monthly cost will start at 3000 rubles.

Which technology is the best to use?
The best combination will be to use OpenAI GPT-4 and ASCN.AI NoCode.

What Common Mistakes Should I Avoid Creating An AI Secretary?
You should never try to automate everything in one go; always do tests; train your teams; monitor your Security and your overall effectiveness; and monitor your effectiveness regularly.

How can an AI Secretary Help Small Businesses?
By using an AI Secretary, small business owners can reduce operating costs by a minimum of ten to fifteen times per business day. The AI Secretary is available 24/7 and enables the small business owner to grow and increase the number of clients without having to hire additional staff.

Additional Capabilities

Multi-Agent Systems

You can assign different types of agents to one Head Secretary. For example, the Sales Agent will manage Leads; the Support Agent can solve Problems for Clients; and the Analytics Agent will produce Reports. The Head Secretary can delegate responsibilities to the various agents.

Automation in Other Areas

• Marketing: the generation of Content, the generation of Leads through Automation.
• Finance: the ability to manage Budgets, process Payments, and perform Analyses.
• Finding Clients: parsing databases of potential Clients, automated Email campaigns, and scoring Leads.

Emerging Trends in Development

• Multimodality: combining Voice, Image, and Video Processing.
• Proactivity: Artificial Intelligence is able to "predict" a client's needs and make suggestions without waiting for a response.
• Emotion Recognition and Style: using Emotional Intelligence is able to identify and then Adapt to the Moods of the Clients.
• Autonomous Decision-Making: Artificial Intelligence will be able to make decisions without Human Interaction.
• Web3 Integration: automated Trading in Cryptocurrencies.

Summary of Recommendations for Launching ASCN.AI

• Start with the most critical Tasks and track the effectiveness of the AI Secretary's efforts.
• Actively engage and train your Team to use the AI Secretary effectively.
• Continue to improve and use Data Based Feedback to improve the AI Secretary.
• Ensure your Security from Day One.
• Regularly review your AI Secretary's processes and adjust Scenario Scripts as needed.

How ASCN.AI Can Help You Make Money

While Automation with the AI Secretary does reduce your Operating Costs, it can also create additional Sources of Revenue.

• The AI Secretary's ability to respond to Clients at any time, day or night, has shown to increase Conversions by 18 to 25%.
• By using an AI Secretary, a Business' Managers will be able to free up hours of their time for Active Selling.
• As a Business grows, the Number of Leads grows and will create Additional Sales Revenue without the need to equally increase Operating Expenses.
• With a high mark-up for assembling AI Secretaries to meet Client needs, the Sales Growth from this source is significant.
• The AI Secretary is also capable of monitoring the Market overnight and will be able to provide updates for Trading.

Example: A Crypto Education Company went from a Revenue of 875,000 rubles per month to 2,250,000 rubles per month due to the AI Secretary implementation. The total cost was 150,000 rubles and it paid for itself in three days.

Using an AI Secretary is a long-term investment with a potential Return on Investment for the first year of between 500 and 2000 percent.