Skip to main content

AI agents for customer service that resolve, not just reply

A chatbot answers questions, while an AI agent can look up an order, start a return or update a customer record. We design and integrate service agents that act in your systems within strict limits and escalate to your team when judgement is needed.

ISO 9001 & ISO/IEC 27001Certified quality and information securitySince 2014, made in GermanyDiese Seite auf Deutsch

From answering to resolving requests

Much of the volume in support queues consists of repetitive requests that need a system lookup and a small action: where is my parcel, change my delivery address, resend an invoice, cancel a booking within the free period. An AI agent connected to your shop, ERP or ticket system through defined tools can complete these end to end and document what it did in the ticket.

We build such agents on kalAIDoCreate, which provides governance, logging and custom APIs for enterprises and the public sector. OnlineAppCreator handles the integration work with your backend, from order management to CRM, as part of its custom development and company building practice. Engineers design every tool the agent may call, including what it is explicitly not allowed to do.

What to watch out for

  • Verify the customer's identity before the agent reveals order details or changes account data. An email address typed into a chat is not enough on its own.
  • Decisions with significant effects on customers, such as rejecting a warranty claim, should not be fully automated. Data protection law restricts solely automated decisions of this kind, so keep a human in the loop.
  • Measure resolution rate and customer satisfaction rather than deflection. An agent that simply keeps people from reaching a human looks efficient but damages loyalty.
  • Agents that read customer messages can be targeted by prompt injection. Keep tool permissions narrow and validate every action on the server side, independent of the model's output.
  • Plan for peak periods like sales events or service outages. Rate limits of model providers and your own APIs need headroom, and the fallback must route to people gracefully.

What service teams gain

Routine requests closed

Status checks, address changes and document requests are resolved directly in your systems. Your team gets fewer tickets that only require copy and paste.

Safe actions by design

Each tool has limits such as maximum refund amounts, allowed order states or required customer verification. Anything outside those limits goes to a person.

Clean handover

When the agent escalates, the ticket contains a summary, the data it checked and the reason for escalation. Agents in your team do not have to start from zero.

Consistent across channels

The same agent logic serves web chat, email and in-app support. Policy changes are made once and apply everywhere.

How we introduce a service agent

  1. Ticket analysis

    We analyse a sample of past tickets to find frequent, rule-based request types worth automating first.

  2. Tool design

    For each use case we define the API calls, permissions, limits and confirmation steps the agent needs.

  3. Shadow mode

    The agent drafts resolutions that staff approve, so you can measure accuracy before it acts alone.

  4. Gradual autonomy

    Use cases with proven accuracy move to autonomous handling, while monitoring and sampling continue.

Matching products

kalAIDoCreate

Governed AI platform for enterprises and the public sector. An all-in-one AI platform for enterprises and public institutions with compliance features, governance and custom APIs.

Explore kalAIDoCreate

OnlineAppCreator Custom Development

Idea to operating product, built by AI Code Engineers. We take ideas to operating digital products: discovery, AI-native engineering by our AI Code Engineers, launch, SLAs and scale. B2B and public sector, from an ISO 9001 and ISO 27001 certified team in Germany.

See how we build products

Frequently asked questions

Which systems can the agent connect to?

Typically shop systems, order management, ERP, CRM and ticket tools that offer an API. If a system has no suitable API, we can build a small integration service in between.

How is this different from our existing chatbot?

A classic chatbot answers from content. An agent also calls tools to read and change data, which requires stricter design, testing and monitoring.

Will customers know they are talking to AI?

Yes. The EU AI Act requires that people are informed when they interact with an AI system, and we build this into every channel along with an easy way to reach a person.

How long until the agent handles requests on its own?

That depends on ticket volume and integration complexity. The shadow phase gives you data on accuracy, and each use case moves to autonomy only when your team is satisfied.

Let's talk about your project.

15 minutes with an AI Code Engineer, free and without obligation. Or call our team around the clock.