One customer brain that answers, acts, and learns — across the whole relationship. Here's what it does, in detail.
3 seats, 500 credits/month, no credit card.
Inbound lands from every channel. The AI retrieves the customer's history, their ARR, your roadmap and last quarter's commitments, drafts a cited reply with a self-rated confidence score, then sends or queues against per-channel floors. Risky topics always go to a human.
KB, customer history, ARR, renewal date, roadmap and the last 20 messages of the thread — assembled into one prompt before the AI writes a word.
Per-channel confidence floors (App Store + Instagram require 0.95, email 0.85). Public channels need 100+ human-approved drafts before auto-send unlocks.
Anyone writes natural-language rules the AI follows on every draft — scoped to a channel, lens, customer tag, or workspace-wide. Conditional on thread history.
Auto-detected and tone-matched, native to the model — not a bolt-on translation layer.
The AI cites your team's actual answers, not a generic FAQ — and gets sharper with every draft you approve.
Refund, cancel, password, lock/unlock, money, percentages, contracts, and “talk to a human” are blocked from auto-send regardless of confidence.
Most AI support tools stop at a suggested reply. PilotPM's brain can take the action the reply promises — and then prove it worked. Every action is human-approved, metered, and gated: the AI proposes, you confirm.
Approve and the brain issues the refund in Stripe, then posts the confirmation back into the thread — no swivel-chair to the billing dashboard.
Cancellation intent is scored on open threads. The rep sends a retention offer without leaving the conversation — and PilotPM tracks whether the customer actually stayed.
When fifty tickets are the same issue, resolve the cluster together with one approved, consistent reply — instead of fifty copy-pastes.
Unlock the account, reset the state, apply the change — the action the customer needs, behind the same approval gate as everything else.
It ties the complaint to the Sentry error and the lines that broke it. The engineering handoff arrives pre-triaged — clustered, impact-ranked, root cause noted.
Technical questions get answered from how your product actually behaves — with the source citation attached, not a guess from a generic doc.
Resolved tickets become drafted knowledge-base articles, so the next customer self-serves the answer instead of opening another thread.
Human-approved, metered against your credit pool, and logged in the audit trail. The AI proposes; you confirm. Nothing irreversible happens on its own.
Human-approved Metered Audit-logged
/help/yourcompany — the same AI on every page.Import your docs, point your DNS, and customers self-serve before they ever email. AI-only chats are free; only escalations count against your credits.
Your logo, your colors, your DNS. The help site reads from the same KB the inbox AI cites, so answers never diverge.
Scope what each audience sees — consumer vs enterprise, free vs paid — so the right docs and the right tone reach the right reader.
The AI answers in the help widget first. Only when it escalates does a conversation land in your team's inbox.
Every customer mention — Slack, HubSpot, App Store, support tickets, Snowflake usage — flows into one signal stream. The AI clusters them into themes, themes pin to initiatives, initiatives carry real ARR. The roadmap stops lying.
The intake agent reads everywhere your customers talk and turns it into one signal stream — nothing imported by hand.
“Locked out of app” said three different ways collapses into one theme, carrying the customers who raised it and their combined ARR.
Promote a theme to an initiative and it inherits the customer list and the ARR — so “rebuild onboarding” sits next to its real $230K, not a guess from the room.
A per-category automation funnel shows what share of each topic the AI can already handle end-to-end, what's queued for a human, and where your knowledge base has a gap to close. Honest numbers, not a vanity dashboard.
Drafted → approved → auto-sent, broken out by topic. You see which categories are safe to turn loose and which need more calibration.
Each category carries a readiness score driven by real resolution data, so you flip on auto-send where the evidence supports it — category by category.
Where the AI keeps escalating, the scorecard flags the missing knowledge — turning “the AI can't answer this” into a concrete article to write.
Routing, SLA, scorecards and coaching — built into the same platform. The work the CS Director used to do manually now happens automatically.
Round-robin with VIP-first priority, language-matched routing to native-speaker agents, and per-agent queue caps so nobody drowns.
Per-inbox SLA targets with VIP-tiered escalation; overdue and due-soon flags surface in the inbox — and you can prove it in the QBR.
SLA compliance, CSAT percent-positive, reopen rate — per-agent and per-team rollups, exportable for HR review cycles.
Every Monday, team leads get an AI coaching summary per agent: strengths, gaps, and a one-pager to take into the 1:1.
Same data, different view — switch when the work changes. The substrate compounds: every approved reply, promoted theme and corrected draft trains your tenant's AI.
Inbox · AI drafts · Auto-pilot
Call prep · Renewals · QBR
Buying signals · Pipeline intel
Themes · Roadmap from real signal
Audit trail · Controls · DPA terms
We show the dashboard on your actual tools, and what we'd have resolved last week. No slides.