Automatic replies to customer requests
Requests from email and WhatsApp get a reply prepared by the AI agent from company systems. A person approves, the reply goes out.
AI automation for companies.
AI agents that answer customers, read documents, prepare quotes and work inside the company ERP. With human checks where it matters and data kept in your own account.
The processes we automate most often
Automatic data extraction from invoices and documents
OCR and AI read invoices, IDs and forms and send the data into the system. Nobody types; a person checks only uncertain fields.
Answers from the ERP, without anyone opening the ERP
Sales ask on WhatsApp or by voice “how much is left in Brașov?” and the agent answers with the stock, price or invoice from the system.
Automatic generation of quotes and contracts
The agent fills company templates with the customer’s data and prepares the document. A person reviews and sends.
Follow-up and reports without human effort
The agent tracks what is still unanswered, sends reminders and delivers the daily or weekly report on its own.
Instant answers from company documents
Employees ask in natural language and get the answer from procedures, contracts and history, with the source cited.
How we build
- Pick one process and define a measurable result
- Connect the agent to company data: email, documents, ERP, CRM
- Rules in code: what it does alone, what needs approval, what it never does
- Pilot on real cases, measured before and after
- Data stays in the company account; AI processing in an EU region
FROM THE PORTFOLIO
Frequently asked questions
Does my company data end up at OpenAI or Google?
Not in the way you fear. Your database, emails and documents stay in your systems; nothing is copied elsewhere. When the agent has to answer a question, it sends the AI model only the information needed for that answer, encrypted and on servers in the European Union. For sensitive data, such as ID documents, we can use models that run on your own servers, with no external service at all.
Can I trust what the AI does? What if it makes a mistake?
The agent does not make important decisions on its own. Anything with consequences, an email sent to a customer, an order, a payment, first reaches a person who approves it. These rules are built into the application; they do not depend on the model “behaving”. We also start with a pilot of a few weeks on real cases, so you see the results before relying on it.
How much does it cost and how long does it take?
It depends on the process, but we do not start with a big project. We pick one process together, the one that costs you the most time, and run a pilot of a few weeks at a fixed cost agreed in advance. If the results are good, we extend step by step. You get the plan with stages and cost before we start.
Do I have to change the software we use?
No. The agent connects to what you already have: ERP, CRM, email, WhatsApp, company files. Your people keep working where they work; part of the work just gets done on its own.
Which process would you like to test?
Tell us what you want to build, what is holding you back and which systems you use. That is where we start.
FAQ
Questions about working with Graalsoft
Where is Graalsoft based and what AI services does it provide?
Graalsoft is a software development and AI automation company based in Bucharest, Romania. We build AI agents, business process automation, document processing and integrations with ERP, CRM and operational systems.
Which business workflows can Graalsoft automate with AI?
Examples include OCR-to-ERP invoice processing, normalising quotation requests from email, WhatsApp and call-centre records, support and warranty workflows, and operational follow-up. Integrations depend on the available APIs, access permissions and agreed process.
How are AI agents different from a standalone chatbot?
A chatbot mainly provides a conversational interface. An AI agent can also carry out permitted steps in a business workflow, such as preparing an ERP record, routing a request or drafting a reply. Human approvals, validation and activity logs define what the agent is allowed to do.
Does AI automation require replacing our existing ERP or CRM?
Not necessarily. A project can connect existing systems through APIs and agreed integration methods. The analysis determines what can be integrated, which permissions are needed and where a person must approve an action.
How should an AI automation project begin?
Start with a clearly described workflow, representative sample documents and the expected outcome. Agree the pilot scope and measurable criteria, such as handling time, error rate and the number of cases that need human review.
Can AI agents include human approval and operational controls?
The solution can be designed with defined roles, approval steps and activity records. These requirements should be agreed during analysis, especially for actions that affect business records or customers.
IMPLEMENTATION
From a workflow to a measurable pilot
Define the input, permitted actions, access rights and human approvals. Test representative cases against the current baseline before expanding.
Documents → ERP
Invoices, purchase orders and goods-receipt documents → field extraction → business-rule checks → operator review of exceptions → ERP import.
Pilot measurement: Extraction accuracy, handling time per document and the share requiring human review.
Requests → quotations
Email, WhatsApp and call-centre requests → structured products, quantities and customer details → ERP or catalogue checks → draft quotation → sales-team approval.
Pilot measurement: Time to a draft quotation, missing information and manual data entry.
Support and warranties
Requests and warranty documents → completeness checks → classification → assignment → draft replies and status updates. Authorised staff retain commercial decisions.
Pilot measurement: First-response time, backlog, missing documents and escalations.
Connected operations
Email, CRM, ERP and service-desk events → tasks, reminders and follow-up → exception queues and reporting, with access roles and activity records.
Pilot measurement: Overdue tasks, hand-offs, reconciliation errors and completion time.