A skincare brand in Rotterdam was paying €4,200/month for two part-time support reps and still missing weekend WhatsApp messages. Their actual ticket volume was 2,800 conversations a month across website chat, WhatsApp, and Instagram DMs. The breakdown of those conversations was almost comically predictable: 41% "where is my order," 18% "do you ship to [country]," 14% return or refund requests, 11% product questions answered word-for-word on the product page, and the remaining 16% were genuine edge cases that needed a human.
Roughly 84% of their inbound was answerable from data they already owned. They were paying senior salaries to copy-paste tracking links.
They tried Tidio Lyro first and got to about 52% deflection on website chat alone. WhatsApp was a separate workflow. Then they looked at Zendesk AI Agents — quoted around $115/month per agent for Suite Professional plus AI Agents add-on metering, and their support lead estimated 4-6 weeks to migrate macros and triggers from their existing setup. Intercom Fin priced out at $0.99 per resolution which, at their volume, would have been roughly $2,300/month before seat fees.
They switched to SimplyBoost Growth at $89/month. The agent indexed their Shopify store, their Zendesk help center articles, and their shipping policy PDF in under 20 minutes. Within three weeks they hit 71% deflection across all three channels. One support rep handles the escalations now. They redeployed the other to retention email — a role that actually moves revenue.
The pattern I keep seeing: most "support" volume isn't really support. It's order status lookups, policy questions, and product clarifications that already exist in your CMS. An AI customer support agent doesn't replace the hard work of customer service. It eats the boring 70% so your humans can do the 30% that matters.
What FAQ deflection actually looks like in practice
Industry benchmarks put modern AI customer support agents in the 60-78% deflection range when they are configured properly. Tidio's published Lyro data hovers around 70%. Intercom's Fin claims similar numbers, charging $0.99 per resolution on top of seat fees. Zendesk's own AI Agents documentation describes roughly 38% automated resolution in their broader customer base — and that gap is the most important number on this page.
The 38% figure is not Zendesk being bad at AI. It is what happens when you bolt an AI layer onto a ticketing system designed for humans to triage everything. Tickets still get created. Macros still fire. Agents still see the queue. The AI handles the easy questions, but the workflow assumes humans are the default.
A standalone AI customer support agent inverts that. The agent answers first. Tickets only exist when escalation is required. That is the difference between deflecting 38% of inbound volume and deflecting 65-75%.
For a store doing 3,000 inquiries a month at $7.50 average cost-per-ticket, the math works out like this. Zendesk AI Agents at 38% deflection: 1,140 tickets deflected, $8,550 saved monthly, minus Zendesk Suite Professional at $115/agent and AI Agents add-on at $50 per 100 automated resolutions — net roughly $5,000-$6,000 in savings depending on seat count. SimplyBoost at 70% deflection: 2,100 tickets deflected, $15,750 saved monthly, minus $89/month flat for the Growth tier. The flat-price model is why this works at small scale.
How the agent decides what to answer and what to escalate
The deflection number is only useful if the agent does not bluff. We built a hard ground-truth rule: the agent answers from your indexed knowledge base, your product catalog, and your synced order data — nothing else. If it cannot cite a source, it does not answer. It escalates.
Three things drive accurate deflection:
Source-bound retrieval. The agent reads your help center, product pages, return policy, shipping FAQ, and any PDFs or Notion docs you connect. Chatbase does this well at the document-QA level but stops short of operational data. SimplyBoost goes further — Shopify and WooCommerce order lookups, subscription status checks, tracking link retrieval. Roughly 35-40% of "support" tickets are actually order status questions. If the agent cannot pull live order data, your deflection ceiling is around 45%.
Confidence-based escalation. Every response is scored. Below a configurable threshold (default 0.72), the agent hands off instead of guessing. Intercom Fin uses a similar mechanism. The difference is what happens at handoff — Fin opens a conversation in Inbox at $0.99 per resolved AI conversation plus seat licensing. SimplyBoost routes to your existing tool: email to support@yourdomain, a Slack channel, a webhook to Zendesk or Freshdesk, or WhatsApp to an on-call number. We do not charge per resolution.
Topic-level escalation rules. Refunds over $100, account deletion requests, anything involving the word "lawyer" or "chargeback" — these go straight to a human regardless of confidence score. You define these in the dashboard. ManyChat and Tidio handle this with keyword triggers; we use intent classification, which is more robust to phrasing.
Where competitors win, honestly
Zendesk AI Agents is the right choice if you already run a 20+ seat support org on Zendesk Suite and your CSAT measurement, SLA reporting, and workforce management are tied into Talk, Guide, and Explore. Ripping that out to save $89/month is not the trade.
Intercom Fin is the right choice if your business model is high-touch B2B SaaS where every conversation should feel like a sales motion and the $0.99/resolution price scales sub-linearly with your ACV. Fin's reasoning quality on ambiguous questions is genuinely strong.
Chatbase is the right choice if all you need is a website widget answering FAQ off a document corpus and you do not need WhatsApp, Instagram, Messenger, order lookups, or human handoff routing. At $40/month for the Hobby tier it is hard to beat for that narrow scope.
SimplyBoost is the right choice when you are doing 500-15,000 monthly conversations across multiple channels, want flat pricing instead of per-resolution metering, need order data integration, and want the same agent to handle sales-qualifying questions (sizing, fit, availability) as well as post-purchase support. That is the gap we built into.
The escalation policy template we ship by default
When we onboard a new account, the default escalation policy looks like this. You edit it from the dashboard:
1. Order-related questions: answer if order is in customer's name and status is retrievable. Escalate if multiple orders match or status is "exception."
2. Returns: answer with policy and initiate label if value under threshold (default $75). Escalate above threshold.
3. Product questions: answer from catalog and FAQ. If product not found, offer alternatives and escalate.
4. Billing: answer subscription status, next charge date, cancellation steps. Escalate any dispute or refund request.
5. Complaints with negative sentiment score above 0.6: acknowledge, log the conversation, escalate immediately to human via your chosen channel.
6. Off-topic: politely decline and offer relevant topics. Do not improvise.
This is what gets the deflection rate above 65%. Without rules 5 and 6, agents either over-escalate (deflection drops to 40%) or hallucinate (CSAT drops).
What it costs and what to do next
Three tiers, all flat-priced, no per-resolution fees:
- Starter $39/month — up to 1,500 conversations, website widget, 1 knowledge source, email handoff. Right for sub-100-ticket-per-week businesses.
- Growth $89/month — up to 5,000 conversations, all channels (website + WhatsApp + Instagram + Messenger), Shopify/WooCommerce order lookup, Slack/webhook handoff. The tier 80% of customers land on.
- Scale $169/month — up to 15,000 conversations, custom integrations, A/B testing on responses, priority support, GDPR DPA on request.
Data is hosted in Frankfurt, EU. We're registered in the Netherlands (KVK 87456346). Full DPA available for GDPR-bound buyers.
The practical recommendation: if you're paying more than $1,500/month for support tooling and deflecting under 50% of inbound, run a 14-day trial against your real ticket corpus. Index your help center, connect your order system, send 200-500 real historical questions through the agent, and measure deflection plus accuracy. That test takes a Tuesday. The math after that is straightforward.
See full pricing and start a trial at simplyboost.io/pricing.