A chatbot follows a script: buttons, keywords, pre-written answers. When a customer phrases something the script did not expect, it gets stuck.
An AI agent understands a free-form message, finds the answer in your knowledge base and takes an allowed action on its own: looks the customer up in the CRM, logs the outcome, creates a task for a colleague. Anything outside what it is allowed to do goes to a person.
An example (an illustration, not a case study). A customer writes on WhatsApp: “My order still hasn’t arrived. What now?” The agent finds the order in the CRM by phone number, reads the delivery status and replies. If the parcel is later than your threshold, it opens a task for a team member with a summary of the chat.
| Scripted chatbot | Automation | AI agent | |
|---|---|---|---|
| Understands | buttons and keywords | structured data | free-form text |
| Does | shows a canned answer | runs a rule | answers and takes allowed actions |
| Breaks when | the customer goes off-script | data arrives in a new format | the question is outside the knowledge base or ambiguous |
| Fits | menus, simple FAQs | moving data between systems | requests that need to be understood |
Support is one of the more mature uses for agents. In McKinsey’s 2025 survey, respondents most often report agent use in IT and knowledge management, including service-desk work.
The best-known study here looked at an assistant helping human support agents, not an autonomous one. Across 5,172 agents, it raised issues resolved per hour by 15% on average, and by about 30% for less-experienced staff (Brynjolfsson, Li, Raymond, Quarterly Journal of Economics, 2025).
Agents make mistakes. A language model can state a wrong answer with confidence. So we do not build an “agent that never errs”; we build a system where errors are visible and their cost is capped.
Forecasts disagree. Gartner (March 2025) predicts agentic AI will autonomously resolve 80% of common customer service issues by 2029. In January 2026 it predicted that GenAI cost per resolution will exceed $3 by 2030, making full automation too expensive for most organisations. Our takeaway: a person stays in the loop; the agent takes routine requests.
Gartner forecasts (June 2025) that over 40% of agentic AI projects will be cancelled by the end of 2027 due to escalating costs, unclear business value or inadequate risk controls. That is why we estimate the payback first, and say so plainly when an agent will not pay off.
Kick-off to pilot takes roughly 3–6 weeks, depending mostly on system access and the knowledge base.
One integrated agent is our Pro engagement, from €1,250, with training, documentation and a month of post-launch support included. Several teams, custom integrations or a local model: Business, from €3,000. Support after month one is optional, from €170 a month.
On top, you pay the model provider for usage and, if needed, an automation platform: by our estimate, a few dollars to a few hundred per 1,000–10,000 conversations. The cost guide has the breakdown and EU market ranges.
With a cloud model, each request is sent to the provider (OpenAI, Anthropic or Google) for processing; some offer regional processing at a higher price. If data must not leave your company, we deploy the model on your own server or private cloud, with hardware quoted separately.
The agent gets only the CRM and tool access it needs for its actions. Your company grants that access, and we hand it back when the project ends.
We do not declare your agent GDPR- or EU AI Act-compliant on your behalf. That assessment stays with your DPO or legal counsel. We sign a data processing agreement (GDPR Art. 28) and give them the data flows, providers and storage locations they need.
CRM: HubSpot, Salesforce, Pipedrive, KeyCRM, SalesDrive, Bitrix24. Messaging and social: WhatsApp, Telegram, Viber, Instagram, Facebook. E-commerce and payments: Shopify, WooCommerce, Stripe. Workspace: Google Workspace, Notion, Slack, Airtable. Platforms: n8n, Make, Zapier. Models: OpenAI, Claude, Gemini.
Don’t see your tool? If it has an API, we can connect it.
We do not publish case studies with numbers yet. A client result will appear here only with the client’s written permission, a baseline and a measurement period. How we measure is on our methodology page.
A chatbot follows a script and gets stuck when a customer goes off-script. An agent understands free-form text, answers from your knowledge base, takes allowed actions in the CRM and hands a person whatever it cannot resolve.
From €1,250 one-off for one integrated agent, a month of support included. Monthly: model usage, a platform if needed, and optional support from €170.
Roughly 3–6 weeks from kick-off. It depends mostly on system access and the state of your knowledge base.
It can. That is why it answers only from the knowledge base and hands over anything that is not there. In the pilot a person reviews its answers, and each error is fixed in the knowledge base or rules.
With a cloud model, requests go to the model provider; with a local model, they stay on your infrastructure. The compliance assessment belongs to your DPO or legal team, and we give them a clear description of the data flows.
No. It takes routine requests; complex and sensitive cases stay with people and arrive with a summary.
Start with a short brief — we reply with an estimate.