The verdict in three sentences
An AI assistant wired to your documents (RAG) resolves 40 to 60% of level-1 requests without human intervention, 24/7. The build is scoped between 15,000 and 50,000 EUR over 6 to 12 weeks, with a monthly inference cost of 200 to 2,000 EUR depending on volume. The real question isn't "AI or not" but custom build vs SaaS, and where your data lives.
Custom build or SaaS (Intercom Fin, Zendesk AI)?
SaaS solutions are fast to activate but charge per resolution and keep your data in their ecosystem. Custom offers control and a mastered marginal cost.
| Criterion | SaaS (Intercom Fin, Zendesk AI) | Custom build (RAG) |
|---|---|---|
| Setup cost | 3,000 - 10,000 EUR | 15,000 - 50,000 EUR |
| Timeline | 2 - 5 weeks | 6 - 12 weeks |
| Cost per resolution | 0.90 - 1.50 EUR/resolution | ~0.02 - 0.15 EUR (tokens) |
| Data control | limited (vendor cloud) | total (your hosting) |
| Tone/logic customisation | medium | total |
| Internal IS integration | standard | custom |
| Vendor lock-in | high | none |
Simple rule: under 1,500 conversations/month, a SaaS may be enough. Beyond that, custom's cost per resolution becomes unbeatable.
Monthly inference cost (tokens)
The recurring cost depends on conversation volume, answer length and chosen model. 2026 orders of magnitude:
| Monthly volume | Economy model | Premium model | Equivalent SaaS |
|---|---|---|---|
| 500 conversations | ~40 EUR | ~120 EUR | ~600 EUR |
| 2,000 conversations | ~150 EUR | ~450 EUR | ~2,400 EUR |
| 5,000 conversations | ~350 EUR | ~1,100 EUR | ~6,000 EUR |
| 10,000 conversations | ~700 EUR | ~2,000 EUR | ~12,000 EUR |
At 5,000 conversations/month, custom costs roughly 6 to 17 times less marginally than per-resolution SaaS.
Performance and GDPR
A well-designed assistant doesn't try to handle everything: it resolves the repetitive and escalates the rest cleanly.
| Criterion | Realistic 2026 target |
|---|---|
| Automatic resolution rate | 40 - 60% of L1 requests |
| Escalation rate to human | 40 - 60% (assumed) |
| CSAT maintained | >= human agent level |
| Data hosting | EU cloud, GDPR-compliant |
| Log retention | configurable, anonymisation |
| Knowledge base | your docs, never shared |
Mini case study
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
Sofia, customer support lead at a 45-employee SaaS SME in Bordeaux, handles 3,800 tickets/month with 4 agents. Kolonell delivers a RAG assistant wired to the help base and CRM for 34,000 EUR in 10 weeks, inference cost ~280 EUR/month.
Result: 48% of tickets resolved automatically, i.e. ~1,800 tickets/month off agents' plates. That frees the equivalent of 0.8 FTE (~32,000 EUR/yr saved) and cuts first-response time from 4 h to under a minute. ROI reached in about 13 months, customer satisfaction stable then rising.
FAQ
Will the AI make things up?
A RAG assistant answers only from your documents, with citations and guardrails. We measure the hallucination rate at acceptance and escalate any uncovered case to a human.
SaaS or custom: the switchover threshold?
Below ~1,500 conversations/month, a SaaS like Intercom Fin is often more cost-effective. Beyond that, custom's cost per resolution (0.02-0.15 EUR) crushes the SaaS model (0.90-1.50 EUR).
Where does our data live?
On a GDPR-compliant European cloud, under your control. No customer data trains a third-party model; logs are anonymisable and retention configurable.
What does monthly operation really cost?
Between 200 and 2,000 EUR/month of inference depending on volume, plus hosting. It's the item to model at scoping to avoid surprises.
What resolution rate should I target?
A realistic 2026 target is 40 to 60% of level-1 requests. Aiming for 100% is counterproductive: better to escalate complex cases cleanly.
Let's scope your project. Give us your monthly ticket volume, your channels (email, chat, WhatsApp) and your knowledge base: indicative budget 15,000-50,000 EUR plus inference. 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.
