Try Oscar Chat — free popups & forms, AI chat from €60/mo
1. Pricing Model Honesty
What good looks like: You can predict your monthly bill on the worst month of the year. Flat per-seat pricing, or usage pricing with a hard cap you control. The pricing page shows what happens at 2×, 5×, and 10× your current volume — in numbers, not adjectives.
Red flags: Per-message or per-conversation credits with no cap. “AI resolutions” billed differently from “AI conversations.” Overages described as “flexible.” If the vendor can’t tell you what your December bill looks like during a traffic spike, you’re not buying software — you’re signing a blank check. Credit-based billing also punishes success: the better your bot works, the more popular it gets, the more you pay.
2. Training Quality
What good looks like: Two training paths that work together — structured Q&A pairs for your most-asked questions, and a knowledge base where you can paste long documents (policies, product manuals, FAQs). Just as important: a defined fallback. When the bot doesn’t know an answer, it should say so, offer a human handoff or a contact form — never improvise. Bonus points for a testing mode that generates likely customer questions and shows you weak answers before you publish.
Red flags: “Just drop in your website URL and the AI figures it out.” Scraped-site bots hallucinate prices, invent policies, and answer confidently about products you discontinued in 2023. If there’s no way to review and correct answers before going live, you’ll be apologizing to customers within a week.
3. Languages
What good looks like: Real multilingual depth — the bot detects the visitor’s language and answers natively, not through a visibly machine-translated layer. It matters most where the language gap is the whole problem: a clinic with international patients, a store selling across borders. Ask the vendor: how many languages, and is the quality tested by native speakers or just “powered by a translation API”? For a deeper look, see Multilingual AI Chat: How to Sell in 50+ Languages Without Hiring Translators.
Red flags: Checkbox i18n — a settings dropdown listing 100 languages the bot can technically output, none of them well. Test it yourself: paste a real customer question in your second-biggest market’s language and read the answer. If it reads like a 2015 Google Translate screenshot, your international visitors will notice before you do.
4. Conversion Tooling Beyond Chat
What good looks like: Chat is one touchpoint. The toolkit should also catch visitors who never open the chat: exit-intent popups for abandoning shoppers, smart forms that ask the right question at the right step, lead capture that feeds your CRM. When chat, popups, and forms share one brain, a visitor who dismissed a popup doesn’t get asked the same question again by the bot. See our Popup Display Rules Tutorial for how to show the right form at the right moment.
Red flags: A chatbot that is only a chatbot. It will answer “where is my order?” beautifully while your cart abandonment rate doesn’t move a millimeter. If the vendor sells popups as a separate product with separate billing, you’re assembling a stack — and stacks leak leads between the tools.
5. Integrations That Matter
What good looks like: Native connections to the systems you already run — your store platform (Shopify), your email tool (Klaviyo), your CRM (Salesforce, Zoho) — so a lead captured at 2 a.m. is in the right pipeline by morning. Webhooks for everything else, so your developer isn’t reverse-engineering an API that was an afterthought.
Red flags: “Integrations” that turn out to be a Zapier-only bridge with a markup on every task. Or a logo wall where half the logos are “coming soon.” Pick your three most critical tools and verify each integration exists today, in the product, not on the roadmap slide.
6. Team and Agency Fit
What good looks like: Workspaces. Multiple websites, multiple clients, multiple brands — each with its own widgets, training data, and team permissions, under one login. If you run three stores or manage clients, you should not be logging in and out of three accounts like it’s 2012.
Red flags: One account = one website, and “agency plans” that mean “buy five accounts and we’ll invoice you once.” Also watch for per-site pricing that quietly triples when you launch your second store. Multi-site is where flat pricing models earn their keep — and where per-message billing goes to die.
7. Support and Exit
What good looks like: You can export your training data — Q&A pairs, knowledge base, conversation history — in a usable format. Your data is yours. Support answers with humans who can look at your actual bot, and there’s a real answer to “what happens to my setup if I cancel?”
Red flags: No export. Your carefully trained bot is hostage, and the vendor knows it. Support that’s itself a chatbot with no escalation path (yes, chatbot companies do this, and yes, it’s exactly as frustrating as it sounds). If leaving means rebuilding your entire knowledge base from scratch, factor that lock-in cost into the “cheap” plan.
The No-BS Scoring Rubric
Score any vendor 0–2 on each criterion: 0 = red flag present, 1 = acceptable, 2 = genuinely good. A serious tool scores 11+ out of 14. Anything scoring 0 on pricing honesty or training quality should be disqualified outright, no matter how pretty the demo is — those are the two failures you can’t patch later. We applied this rubric in our Oscar Chat vs Chatbase comparison and in our Oscar Chat vs Tawk.to comparison. Once live, measure what matters with our guide to 12 Chatbot Metrics That Actually Matter.
Oscar Chat’s honest self-assessment: Strong on the criteria buyers get burned on — flat €60/month per-seat pricing with no per-message credits, conversion tooling built in (free popup builder and free smart form builder alongside the AI chat), real 50+ language support, and workspaces for multi-site teams and agencies. Weaker spots, stated plainly: we’re a newer brand, and our integration catalog is smaller than the incumbents’ — it covers Shopify, Klaviyo, Salesforce, and Zoho natively, but if you need a niche tool connected on day one, check with us first. We’d rather lose a sale on honesty than win one you’ll regret in December.
Frequently Asked Questions
How much should an AI chatbot cost for a small business?
For most small businesses, €50–150/month per seat is the sane range — if it’s flat. The danger isn’t the headline price, it’s the model: per-message or credit billing can double or triple your bill in a good sales month. Always model your busiest month, not your average one, before signing.
What’s the difference between Q&A training and a knowledge base?
Q&A pairs are precise: you write the exact question and the exact answer, so critical information (prices, policies, hours) is always correct. A knowledge base is breadth: you paste long documents and the bot draws answers from them. Good bots use both — Q&A for what must never be wrong, knowledge base for everything else.
Do I really need multilingual support if most customers speak English?
Check your analytics for the visitors who don’t. If 15% of your traffic is international and your bot only answers in English, you’re not saving money — you’re quietly abandoning that segment. This was the founding insight behind Oscar Chat: clinics and stores serving people who don’t speak the site’s language lose them at the chat window first.
Can a chatbot actually increase sales, or is it just support deflection?
Support-only bots deflect tickets; conversion-equipped bots make money. The difference is tooling: exit-intent offers, smart lead forms, and chat that can recommend and capture — not just answer. If a vendor’s case studies only mention “tickets resolved,” it’s a support tool wearing a sales costume.
What should I check before migrating from another chatbot?
Three things: can you export your training data from the old tool (if not, budget time to rebuild it), does the new tool import or at least structure that data sensibly, and what happens to your historical conversations. Migration pain is almost always data pain.
Is a free popup or form builder actually free?
With Oscar Chat, yes — the popup builder and form builder are free to use; the €60/month per seat is for the AI chat. With most vendors, “free” means a capped trial: limited popups, limited submissions, their branding on your site. Ask what’s capped before you build your lead capture on it.
The cheapest chatbot is the one you don’t have to replace in six months. Run the checklist, score honestly, and pick the tool that’s still the right answer on your busiest day of the year.
