AI Support Chatbot for a Software Vendor: Cost (2026)

The verdict in three sentences

For a software vendor receiving 4,500 tickets a month, a RAG-based AI chatbot connected to the documentation costs €15,000 to €40,000 excl. VAT and resolves 35 to 55% of requests without human intervention. Inference cost is marginal, €0.01 to €0.05 per conversation, and the average cost per ticket drops from about €6 to €3.50. The real issue is not the model but helpdesk integration (Zendesk or Freshdesk) and documentation quality, which drive the resolution rate.

How a RAG support chatbot works

RAG (retrieval-augmented generation) means indexing your sources and having the model answer only from the retrieved passages. For a software vendor, useful sources are the product documentation, release notes, knowledge base articles and, above all, resolved tickets from the last 12 months, anonymised. The bot cites its sources, refuses to answer out of scope and hands over to tier 1 with a conversation summary when it is not sure. Connected to your product API, it can also check a licence status, the installed version or the last failed import, which makes the difference on the most frequent tickets.

Cost per ticket before and after

2026 assumptions: tier 1 human cost of €6 per ticket (team of 7 agents in Lyon, fully loaded cost of €3,850 a month), inference at €0.03 per conversation across all 4,500 conversations, hosting and maintenance at €800 a month.

A 35% rate is achievable with decent documentation, 55% requires documentation kept up to date and actions wired to the product API.

Native helpdesk AI, SaaS platform or custom RAG

Zendesk and Freshdesk offer their own AI agents, billed per resolution. They are quick to switch on, but the bill grows with volume. Custom development costs more upfront and much less in use.

The custom RAG stays connected to Zendesk or Freshdesk: it creates the ticket, fills in the fields, adds the summary and routes it to the right group when a human needs to take over.

Safeguards that protect the customer relationship

A support bot that invents a procedure costs more than one extra ticket. The specification requires answers limited to indexed sources with a link to the article, a confidence threshold below which the bot offers a human, a handover button visible at all times, EU hosting with zero retention at the model provider and a weekly dashboard of unanswered questions. That last point pays off the most: it tells the product team which articles to write first.

Mini case study

Julien, head of support at a Lyon-based inventory management software vendor, serves 3,000 customers with 7 agents and 4,500 tickets a month. Planned growth calls for two new hires, i.e. €7,700 a month in fully loaded payroll. He launches a custom RAG billed €28,000 excl. VAT, connected to Zendesk and the licensing API. By month three, the bot resolves 45% of tickets and the two positions are no longer needed. Net saving: 7,700 − 935 = €6,765 a month. Payback: 28,000 ÷ 6,765, just over 4 months. First response time falls from 4 hours to a few seconds for almost one ticket in two.

FAQ

What resolution rate should we target in year one?

Expect 30 to 35% at launch, then 45 to 55% after three to six months of documentation improvements. Beyond 60%, further gains require deep actions inside the product.

How do we stop the bot from making up answers?

It answers only from retrieved passages, cites the source article and hands over below a confidence threshold set during acceptance testing. A test set of 200 real questions measures accuracy before go-live.

Does the bot work with Freshdesk as well as Zendesk?

Yes, both expose complete APIs to create and enrich tickets. Integration accounts for €3,000 to €6,000 of the budget depending on custom fields and routing rules.

Does our customer data stay in Europe?

Yes if the contract says so: hosting in France or Germany, the model called with zero retention, tickets anonymised before indexing. These commitments are written into the DPA.

Why an 8-week project?

Two weeks to audit the documentation and build the test set, four weeks of development and helpdesk integration, two weeks of acceptance testing with a group of pilot customers.

Let's scope your project. Send us an anonymised export of 500 tickets and your documentation URL: we will estimate your achievable resolution rate, an indicative budget between €15,000 and €40,000 excl. VAT and an 8-week plan. Detailed quote within 48 h. WhatsApp +221 77 596 93 33.