AI assistants that answer questions and take action in your systems.
Help your customers and team find information, schedule appointments and submit requests through WhatsApp, web, apps or voice. We configure each assistant with your sources, permissions and authorized actions, and define when it should hand the conversation over to a person.
Resolving a conversation takes context, information and access to your systems.
- 01
Questions in context
People ask the same question in different ways, through text or voice. The assistant needs to interpret the request within the service's defined scope.
- 02
Actions that require integration
Scheduling an appointment, checking a balance or submitting a request requires connections to business systems, authentication and permissions.
- 03
Information and access permissions
Policies, manuals and contracts may be stored in different sources. Responses need to use information authorized for each user.
Your sources and systems, connected to the conversation.
We configure the assistant around the questions and tasks you want it to handle. We connect authorized sources, define actions and permissions, and adapt the conversation to each channel. Human handoff is part of the workflow when the case calls for it.
- 1
Interprets the request
Processes questions through text or voice and uses conversation context to identify the request within the defined use case.
- 2
Consults authorized information
Retrieves information from manuals, policies, catalogs, contracts or tickets according to the user's permissions, with source references available for review.
- 3
Performs defined actions
Looks up information in your systems, schedules appointments or submits requests through the agreed integrations, with authentication and permissions for each action.
- 4
Works across your channels
Shares the assistant's logic across WhatsApp, web, apps, email and voice, with the integrations and conversation adjustments each channel requires.
- 5
Hands off with context
Transfers the conversation to a person according to the agreed criteria and shares the available context. Records questions and unresolved cases so the team can evaluate and improve the service.
Integration, evaluation and ongoing operational support.
- Integration and permissions
- We connect sources and systems with authentication, role-based permissions and actions limited to the use case.
- Quality and traceability
- We evaluate responses, actions and handoffs using representative conversations. We log the sources consulted and actions performed to review performance.
- Scope and channels
- We start with one use case and one channel. Each phase defines deliverables, integrations and criteria for expanding the scope.
- Operations and support
- A dedicated team and 24/7 coverage for critical incidents, with agreed response times.
- Continuity and handover
- Exportable scripts, indexes, configurations and logs, with documentation and handover terms defined in the contract.
Amarello's capabilities and experience.
service through WhatsApp, web, apps and voice, according to the project's scope and integrations
information and source references based on the user's permissions
Amarello implementations across industries since 2008
Experience building conversational assistants
We have built conversational AI assistants and supported their design, integration and operation. The work includes designing conversations, connecting systems, configuring human handoff and reviewing service performance.
Companies Amarello has worked with
Banking · Insurance · Financial services · Retail
We move phase by phase, with verified results.
We start with one use case and one channel. We agree on the questions the assistant should handle, the actions it can perform and when it should hand off to a person. We evaluate response quality and completed tasks before expanding the scope. The timeline is agreed based on the scope, sources and required integrations.
- 1
Use case and scope
- What we do
- We define the use case, channel, sources, permitted actions and human handoff criteria. We agree on how to evaluate responses, tasks and service times.
- Your team's role
- Involve process owners and provide information sources, policies and access needed for the assessment.
- Criteria to proceed
- Scope, permissions, actions and acceptance criteria agreed.
- 2
Testing with real conversations
- What we do
- We test with representative questions and voice notes in a test environment. We evaluate responses, source retrieval, actions and handoffs against the agreed criteria.
- Your team's role
- Provide reference conversations and involve people who can define what makes a response, action or handoff appropriate.
- Criteria to proceed
- Results evaluated, limitations identified and required adjustments defined before the pilot.
- 3
Pilot on one channel
- What we do
- We test the assistant with users on the selected channel. We review conversations, integrations and human handoff.
- Your team's role
- Involve participating users, a channel owner and the team that will receive transferred conversations.
- Criteria to proceed
- Responses, actions and handoffs validated on the channel against the agreed criteria.
- 4
Production and operations
- What we do
- We activate support and performance reviews, assign content responsibilities and document operations. We add use cases and channels based on results and the agreed scope.
- Your team's role
- Assign an assistant owner and people responsible for keeping its sources and policies up to date.
- Criteria to proceed
- Operations and support active, content owners assigned and documentation delivered.
Architecture, integrations, security and deployment.
Channels
WhatsApp Business API, web and apps through a widget or API, email, and voice through telephony or voice notes. The assistant shares common logic, while the required integrations and interaction adjustments are defined for each channel.
Models
Language models, information retrieval with references to your sources, and speech recognition. During the technical assessment, we specify which models process the information. Using your data to train models requires an explicit agreement.
Integration
Integration with CRM, core systems, scheduling, payments and corporate directories, according to the use case. The tools available to the assistant are limited in scope, and authentication and permissions are validated for each action.
Security
Authentication before performing actions, role-based permissions, and conversation and action logs. During the technical assessment, we review security and data location requirements to identify compatible deployment options.
Quality
Evaluation with representative conversations, rules that limit responses and actions, monitoring for changes in performance, and human review of transferred cases. Consulted sources are logged for review; how references are presented to the user depends on the channel's format.
Deployment
Google Cloud Platform (GCP) is our primary runtime environment. Availability on Amazon Web Services (AWS) or private infrastructure is confirmed during the technical assessment, based on the environments validated for the solution.
Data export and handover
The contract defines exports of scripts, indexes, configurations and logs, the documentation to be delivered and the terms for handover to the client's team.
Questions before getting started.
How is this different from a rules-based chatbot?
A rules-based chatbot follows predefined flows, which can be useful for narrowly scoped interactions. An AI assistant can interpret natural-language requests, retrieve information and use tools connected to your systems. It also needs rules, permissions and evaluation criteria to operate within the agreed scope.
Is it for my customers or for my team?
Both. For each use case, we define the questions, sources, permissions and actions involved. The configuration fits the process you want to support, whether that is customer service or internal information access and task management.
Which channels can it use? Does it support voice?
It can be deployed on WhatsApp, web, apps, email and voice. Voice includes telephony and voice notes. Each project defines the channels, integrations and how responses are presented. Shared assistant logic does not remove the need to configure each channel.
What happens if it cannot find enough information to answer?
It is configured to use authorized sources and, when the available information is insufficient, state that limitation or hand off to a person according to the agreed criteria. We evaluate this behavior with representative conversations and review performance in operation.
What actions can it perform in our systems?
It can look up information, schedule appointments or submit requests through the project's integrations. The scope defines the available actions and the authentication and permissions required. Results are tested before those actions are enabled in production.
How do you evaluate responses and improve the assistant?
We agree on criteria for responses, actions and handoffs and evaluate them using representative conversations. We review cases that need improvement and update the relevant sources, instructions or integrations. This process does not mean the model is automatically trained on every conversation.
How is our information protected, and how is our data used?
We define authentication, access permissions and activity logging. During the technical assessment, we specify which models process the information and where they run, based on validated deployment options. Using your data to train models requires an explicit agreement.
How do you support continuity without dependence on Amarello?
The contract defines exports of scripts, indexes, configurations and logs, documentation, rights to components and handover to your team. We provide support to help your team make changes to the solution.
Related solutions
- Customer serviceAI agents for customer service, sales and collections across your channels, connected to your business information.Explore the solution
- AI document processingClassify documents, extract information and validate data against your business rules. Your team reviews exceptions and makes the final decision.Explore the solution
Let's define the first use case for your assistant.
Together, we review the questions, sources and actions you want to enable. We start with one channel and clear criteria for evaluating responses, tasks and handoffs before expanding the scope.