The verdict in three sentences
An AI qualification agent only makes sense if your sales team already loses measurable time on off-target requests: above 300 inbound leads a month, the math turns favorable. For a B2B software vendor, a serious project costs EUR 10,000 to 22,000 excl. VAT to set up, then EUR 0.05 to 0.20 per conversation in API costs. The real gain does not come from the chatbot itself but from its integration with the CRM and the sales team's calendars, which turns a conversation into a dated, qualified meeting.
What a qualification agent actually does
The typical scenario: a visitor lands on the pricing or demo page, the agent starts the conversation and asks five qualification questions inspired by BANT or MEDDIC (company size, current tool, project timeline, indicative budget, role in the decision). Depending on the answers, it routes to one of three outcomes: direct booking of a slot with a sales rep, a tailored resource plus enrollment in nurturing, or a documented answer for support requests and job applications.
Each exchange is summarized and pushed into the CRM (HubSpot, Pipedrive, Salesforce) with a score, so the rep does not have to requalify by phone. Guardrails matter as much as the model: a knowledge base limited to your approved content, refusal to quote a firm price, and handoff to a human as soon as a question goes off-script.
| Project component | Content | 2026 budget (EUR excl. VAT) | Timeline |
|---|---|---|---|
| Scoping and scenarios | Qualification grid, flows, tone | 1,500 - 3,000 | 1 week |
| Conversational agent | Prompting, knowledge base, guardrails | 3,500 - 7,000 | 2 - 3 weeks |
| CRM integration | Contact creation, score, summary, owner | 2,000 - 5,000 | 1 - 2 weeks |
| Meeting booking | Shared calendar, routing rules by region | 1,000 - 3,000 | 1 week |
| Testing and GDPR compliance | Test sets, notices, retention period | 1,000 - 2,500 | 1 week |
| Dashboard | Conversion rate by source and outcome | 1,000 - 1,500 | 3 days |
| Total | 10,000 - 22,000 | 6 - 9 weeks |
Recurring costs and risks to budget for
Usage cost depends on conversation length and the model chosen. A five-question qualification uses 3,000 to 8,000 tokens on average, i.e. EUR 0.05 to 0.20 per conversation depending on the model. For 600 conversations a month, the API bill stays between EUR 30 and 120, far below the cost of half a day of SDR time.
| Recurring item | 2026 order of magnitude | Comment |
|---|---|---|
| Language model API | EUR 30 - 120 / month | 600 conversations, mid-range model |
| Hosting and monitoring | EUR 50 - 150 / month | Logs, alerts, backup |
| Maintenance and answer tuning | EUR 300 - 800 / month | Monthly review of failed conversations |
| Widget or platform license (if SaaS) | EUR 0 - 400 / month | Zero with custom development |
| Hallucination risk | Low with a closed knowledge base | Regression tests before every update |
| GDPR risk | Medium | Visitor notice, no sensitive data collected |
The three main risks are well known. The agent may promise a feature that does not exist: the fix is a closed knowledge base and a fallback sentence. It may put off a good prospect with too many questions: beyond five, drop-off climbs sharply. Finally, the EU AI Act has required since 2025 that users be clearly told they are talking to a machine.
Mini case study
Claire, marketing director at an HR software vendor in Lyon, receives 600 requests a month. 70% are off-target (students, micro-businesses under 10 employees, support requests). Her two SDRs spend about 12 minutes per request sorting them, i.e. 120 hours a month.
After deploying an agent for EUR 16,000 excl. VAT, the agent automatically filters out or redirects 380 requests. The SDRs now handle only 220 pre-qualified conversations, i.e. 44 hours a month: 76 hours recovered, valued at EUR 45 fully loaded, or EUR 3,420 a month. The qualified meeting rate rises from 9% to 17%, i.e. 102 meetings instead of 54. With EUR 600 in monthly recurring costs, the project pays back in about 6 months on time saved alone, before counting any extra contracts.
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
FAQ
Does an AI agent replace an SDR?
No, it absorbs the sorting of low-value requests, which often make up 60 to 70% of inbound volume. The SDR focuses on the 30% of qualified leads and on outbound prospecting.
Do you need a CRM for the project to work?
Yes, without CRM integration the agent produces isolated conversations that nobody uses. Budget EUR 2,000 to 5,000 excl. VAT for a clean integration with HubSpot, Pipedrive or Salesforce.
How many questions should the agent ask?
Five questions at most, two of them closed-choice. Beyond that, drop-off frequently exceeds 40% on mobile.
How long before results can be measured?
A first assessment is reliable after 8 to 10 weeks and about 1,000 conversations. That is the volume needed to compare the meeting rate before and after.
Is prospect data GDPR-compliant?
It is if you inform the visitor, limit collection to business data and set a retention period, usually 3 years for a prospect. Hosting logs in the European Union simplifies the impact assessment.
Let's scope your project. Send us your monthly lead volume and current CRM: we will price a qualification agent between EUR 10,000 and 22,000 excl. VAT, delivered in 6 to 9 weeks. Detailed quote within 48 h. WhatsApp +221 77 596 93 33.
Mohamed Bah
Fondateur, Kolonell
Passionate about digital and entrepreneurship in Africa, Mohamed has been helping Sénégalese businesses with their digital transformation since 2020. Founder of Kolonell, he believes every SME deserves a professional and accessible online présence.
