What Freddy Does
Freddy is the AI layer embedded across Freshworks apps — not a separate product, not a separate license screen. Features surface inside Freshsales (lead scoring, deal insights), Freshdesk (suggested responses, ticket summarization), Freshservice (similar-incident detection, virtual agent), and Freshchat (intent classification, generative answers). Three families exist under the Freddy umbrella: Freddy Self Service (deflection bot), Freddy Copilot (in-product assistant for agents and reps), and Freddy Insights (analytics and forecasting). Knowing which family a feature lives in matters for billing and for tuning, because the underlying models and quotas differ.
Freddy Copilot
Sits in a side panel next to the agent or rep interface. Drafts replies grounded in conversation history and your knowledge base, summarizes long ticket threads in 3-5 bullet points, suggests next-best actions (“offer a refund,” “escalate to engineering,” “schedule a follow-up call”). The summary feature is the most-used; turn it on first because it produces immediate productivity gain without retraining agents. Copilot calls are metered per “AI session” — a session is a continuous interaction window, typically 30 minutes — and the per-tier session quota matters for tenants with high agent counts.
Predictive Scoring
Deal scoring in Freshsales uses a model trained on your portal’s last 12 months of closed-won and closed-lost deals. Lead scoring uses similar inputs at the contact level. Ticket priority prediction in Freshdesk uses ticket text plus historical priority assignments to predict the right priority on creation, reducing manual triage time. Models retrain monthly. They underperform when the historical dataset is small (less than 200 closed deals or 1,000 closed tickets); below that threshold, rule-based scoring is more predictable.
Conversational Bot
Freddy chatbot builder ships two modes: flow-based (you draw branching dialogue trees) and LLM-backed (Freddy Self Service answers from your KB without explicit flows). Production bots usually combine both — flows for structured intents (reset password, check order status) where the path is fixed, LLM for open-ended questions. Configure the bot from Freshchat -> Freddy AI -> Bot Builder. Top-5 supported intents typically deflect 30-50% of inbound conversations once tuned.
// Trigger Freddy Self Service answer via Freshchat conversation API
fetch(`https://api.freshchat.com/v2/conversations/${convId}/messages`, {
method: 'POST',
headers: { Authorization: `Bearer ${jwt}`, 'Content-Type': 'application/json' },
body: JSON.stringify({
message_type: 'normal',
actor_type: 'agent',
actor_id: freddyBotId,
message_parts: [{ text: { content: 'How can I help?' } }]
})
});Pricing
Tiered. Basic Freddy features (similar-ticket detection, simple intents) are bundled in mid plans. Advanced AI — Copilot drafting, generative summaries, AI Agent for autonomous resolution — sits in Enterprise or as add-on Freddy AI Copilot / Freddy AI Agent SKUs. Pricing is per agent per month for Copilot, per resolved session for AI Agent. Price the rollout before enabling broadly because a tenant of 200 agents can rack up a five-figure monthly add-on bill in a quarter.
Common Failure Modes
The top three. First, enabling generative answers without curating the KB — the bot will confidently quote a 2022 article that contradicts your current pricing. Audit and tag KB articles with freddy_eligible and exclude the rest. Second, expecting Freddy Insights forecasts to be accurate in a tenant with inconsistent stage-exit criteria; the forecast learns from your historical noise. Third, treating Copilot output as send-ready — it is draft-quality and needs a human review pass; agents who hit Send without reading will produce escalations.
What to do this week
Turn on ticket summarization for one team and measure average handle time before and after for two weeks. The delta usually justifies the broader rollout, and the data shuts down the inevitable “is AI worth it” debate.
