What is Agent Assist?
Agent Assist is AI support for customer service employees during their work. An employee asks a question or is handling a customer interaction. Agent Assist searches the company’s own knowledge for the relevant answer, work instruction or next step and displays the source alongside it. The employee assesses what is appropriate and remains responsible for the customer interaction.
- Your own knowledge as the foundation
- Answers with sources
- The employee remains in control
5 min read · Explanation & practice
From a question to a useful answer.
“How do I help a customer with a return?”
The employee asks the question and adds context where necessary.
Start with the situation. The employee remains the person helping the customer.
The right policy. The right work instruction.
Agent Assist searches for relevant information about conditions, exceptions and next steps.
The source provides guidance. The employee can read the underlying information.
An answer they can work with.
The employee assesses what fits the customer’s question and follows the appropriate work instruction.
You make the decision. Feedback helps further improve answers and knowledge.
Illustration with fictional content. Not a live product view.
How does Agent Assist work?
Agent Assist helps the employee; a chatbot or AI agent talks to the customer itself. Answers come from your own, managed knowledge: articles, work instructions and policies. They do not come from the general internet.
The employee assesses the answer and remains in control. This only works well when the knowledge is up to date and unambiguous, and someone maintains it.
A question in everyday language
The employee asks a question, for example about a return, an exception or an internal policy. They can add context with a follow-up question.
Search managed knowledge
The AI searches by meaning, not just keywords, and uses the passages relevant to the question. This approach is called retrieval-augmented generation (RAG): the answer is based on retrieved company information.
An answer with sources
The employee can review the underlying information, follow a work instruction step by step and help the customer further.
Feedback that improves knowledge
Employees indicate whether an answer is correct. Questions without an answer show where knowledge is missing. Knowledge managers fill those gaps.
How does it differ from other solutions?
Agent Assist and an AI agent are not mutually exclusive. Many organisations use the same knowledge for both: the AI agent answers simple customer questions and Agent Assist helps employees with the rest.
Agent Assist
Supports employees during their work with answers, work instructions and next steps. It is based on the organisation’s own, managed knowledge, with source references.
Chatbot or AI agent
Conducts the conversation independently and hands it over to an employee when necessary. Answers come from the organisation’s own knowledge and connected systems.
Search function in a knowledge base
Displays articles for a keyword. The employee searches those articles for the answer.
General AI assistant
Writes and summarises based on broad, general knowledge. It is based on general training data, unless the assistant is connected to the organisation’s own sources.
When does Agent Assist deliver the most value?
- Many changing procedures
- Actions, conditions and processes change frequently. Employees lose time searching for the right version.
- New employees
- Training is the beginning. During real conversations, you need guidance without constantly having to interrupt an experienced colleague.
- Exceptions
- Everyone knows the basics, but the difficult case is somewhere in a document.
- Multiple teams or channels
- Customers should receive the same answer whether they call, chat or email.
- Reliance on experts
- The same questions keep ending up with the same experienced colleagues.
What do you need to get started?
Useful knowledge is the starting point. Think of existing articles, documents and work instructions. The knowledge does not have to be perfect, but contradictory and outdated information must be removed. Also read how to prepare your knowledge base for AI.
Appoint an owner who takes care of feedback and unanswered questions. Start with one team and a recognisable set of customer questions. Record who is allowed to see which information: internal and customer-facing knowledge should remain separate. Also determine in advance what you want to measure.
Is this Agent Assist solution right for your team?
Assess the solution you are considering or already using. Answer nine questions and discover what you still need to verify in a demo or pilot.
Your selection at a glance.
No component has been fully proven yet. Use the test actions below to verify claims using your own knowledge and way of working.
Some components have already been proven. Use the test actions below to discuss and test the open points before making a choice.
According to your answers, all applicable components have been proven. Document the findings and test in a focused pilot whether the solution also meets requirements in daily work.
Components you marked as not applicable are not included in this overview.
Start with these test actions.
First address what has not yet been demonstrated, followed by what partially meets the requirements. If the answers are equal, we follow the order of the selection check.
Answers with sources
Request a demo using a customer question and source article of your own. Together, check whether the answer can be traced back to the correct passage.
Walk through work instructions
Have a team member run through one work instruction containing an exception.
Integration with the workplace
Test the workflow in your day-to-day environment and make the required integrations and configuration concrete.
Access to information
Have the supplier demonstrate, using different user roles, which information is and is not available.
Feedback and knowledge gaps
Follow one feedback report and one unanswered question from initial identification through to updated knowledge.
Review before publication
Have the complete process from draft to review, scheduled publication and available answer demonstrated.
Multilingual support
Test a fixed set of customer questions in every required language and verify the content with someone who speaks that language.
Data storage and data use
Request written clarity on data storage and model training. Have this assessed against your privacy and security requirements.
Reuse knowledge
Discuss how the same knowledge will be made available to a chatbot or voicebot, including configuration and access per target group.
Take the step toward a focused pilot.
Start with one team and a defined set of customer questions. Decide in advance what you will measure, such as search time, answer quality and questions directed to experienced colleagues. Discuss the results with the employees who use the solution.
View and change your nine answers
- 01 · Answers with sourcesDemonstratedPartially demonstratedStill to be tested
- 02 · Running through work instructionsDemonstratedPartially demonstratedStill to be tested
- 03 · Integration with the workplaceDemonstratedPartially demonstratedStill to be tested
- 04 · Access to informationDemonstratedPartially demonstratedStill to be tested
- 05 · Feedback and knowledge gapsDemonstratedPartially demonstratedStill to be tested
- 06 · Review before publicationDemonstratedPartially demonstratedStill to be tested
- 07 · Multilingual supportDemonstratedPartially demonstratedStill to be testedNot applicable
- 08 · Data storage and data useDemonstratedPartially demonstratedStill to be tested
- 09 · Reusing knowledgeDemonstratedPartially demonstratedStill to be testedNot applicable
This selection check is based on your answers. It is not an independent product assessment and does not replace a technical evaluation or pilot.
Agent Assist in Archie
Agent Assist is part of Archie Helps, within Archie, the AI platform for customer service from VANAD Digital, part of VANAD Group.
Answers with sources
Answers come from your own knowledge, with relevant sources to read through.
Following work instructions
Work Instruction Copilot suggests suitable work instructions and guides employees step by step. They can continue later where they left off.
In Microsoft Teams
Employees can also ask questions in a 1-to-1 chat in Microsoft Teams.
The same knowledge, multiple languages
Questions and answers work in multiple languages from the same knowledge base.
Learning from missing answers
Feedback and the No Answer log show what is unclear or missing.
Supporting a next step
Through the Action Center, connected systems can support a next step.
Your questions.
Clearly answered.
More about how it works, the knowledge base and the role of the employee.
