You're viewing a read-only demo workspace · sample data only

Your live brain runs against your sources, your rules, your people

Live demo workspace · acme.automatebusinessnow.comRead-only · sample data

Acme Refunds Co's brain — read it like the AI does.

This is what your workspace looks like once it's populated: sources indexed, rules drafted from your data and approved, chat that cites every claim, a challenger that flags the high-stakes calls, and an approval queue for autonomous AI tool calls. Click a citation chip — every fact traces back to a source you uploaded. Ask the brain a question — the demo returns the same answer shape your live brain will.

Set up your own workspace →Book a 30-min setup call

Sources

Five sources · 2,837 facts indexed

Each connected app or uploaded handbook gets parsed into structured facts the AI can retrieve. Re-syncs are clean — existing facts get refreshed in place, not duplicated.

SOP upload

indexed

refund-policy-v3.pdf

Refund policy: under $250 auto-approve, $250–$1,000 manager sign-off, over $1,000 ops director.

24 factslast sync · 2 hours ago

Slack

indexed

#refunds (last 30 days)

118 threads about refund decisions, including the 2 escalations on lawsuit threats this month.

312 factslast sync · 12 minutes ago

Zendesk

indexed

Tickets · refund + cancellation tags

Resolved tickets with refund outcomes — paired ticket → resolution shape used by the strategist.

1,846 factslast sync · 8 minutes ago

Notion

indexed

Customer Success handbook

Escalation paths, voice/tone do-and-don't, the 3-strikes rule for refund denials.

88 factslast sync · yesterday

Gmail

indexed

support@acme.com (90 days)

Threaded conversations with high-value customers — pricing exceptions, retention saves, churn signals.

567 factslast sync · 23 minutes ago

Active rules

Four rules drafted from those sources · re-verified daily

Two shapes: plain policies (refund thresholds, hours, voice) and step-by-step playbooks (pricing exceptions, incident response). Confidence is the verifier's match against the cited facts. Anything below 0.7 gets paused and routed to the owner.

Plain policy

active

Refunds — tiered approval thresholds

·Under $250: auto-approve, log in /admin/refund-log, send standard apology + refund-issued email.
·$250 – $1,000: route to a Customer Success Manager. Rep adds context. Manager has 2 business hours to decide.
·Over $1,000 OR mention of lawsuit, attorney, regulator, or 'small claims': route to ops director. Stop using AI to draft the response — escalate to human conversation.

applies to · Customer Service Rep, Customer Success Repconfidence · 0.944 citationsverified · today, 06:01 UTC

Step-by-step playbook

active

Pricing exception process

·Step 1: Confirm the customer's annual run-rate from Stripe. Quote-time discount cap = 15% if run-rate > $20k, 10% otherwise.
·Step 2: If they ask for a multi-year deal, tag the manager — those need legal review and we don't draft the LOI from chat.
·Step 3: Log the exception in #pricing-log with the deal slug. Quarterly we audit for drift.
·Step 4: If the discount exceeds the cap, the deal goes to the founder via email — never via Slack DM.

applies to · Sales Rep, Customer Success Repconfidence · 0.886 citationsverified · today, 06:01 UTC

Plain policy

active

Escalation — emotional / lawsuit-threat language

·Trigger phrases: 'lawyer', 'attorney', 'small claims', 'BBB', 'fraud', 'I'll go public', 'YouTube', 'Reddit'.
·When matched, the AI never composes an outbound reply. It posts the thread + the trigger to #escalations and pings the on-call CSM.
·Owner gets a daily digest of these — 0 escalations in a week is itself a signal we audit.

applies to · Customer Service Rep, Customer Success Repconfidence · 0.973 citationsverified · yesterday, 21:14 UTC

Plain policy

active

Hours + voice — Customer Service Rep

·Reply hours: 7am–7pm Pacific, Mon–Fri. Off-hours auto-reply names the next available human window — never says 'we'll get back to you soon.'
·Voice: warm, direct, short sentences. No 'we apologize for any inconvenience this may have caused' filler. Reference the customer's order # in every reply.

applies to · Customer Service Repconfidence · 0.912 citationsverified · today, 06:01 UTC

Ask the brain

Try a question. Every claim cites a source.

Five canned questions — pick one, watch the brain pull rules + facts and answer. Citations are live: click a chip to see the rule it traces back to. Your real workspace handles open-ended prompts; this demo is read-only.

Asked

Can I refund $400 without a manager?

Brain answers

No. Refunds between $250 and $1,000 require a Customer Success Manager sign-off per the tiered approval policy. You'll attach context to the ticket, the manager has 2 business hours to decide. Auto-approval only applies under $250.

Citations

In your workspace, citations open the actual source — the Slack thread, the ticket, the handbook page — at the timestamp the fact was extracted.

Challenger pass

A second AI re-reads the answer before the customer sees it.

On high-stakes turns — refunds, lawsuits, hires, big-dollar amounts — a second AI runs the prompt with an adversarial system prompt and flags what the primary missed. Verdict blocked means the AI does not respond; the work goes to a human.

Inbound · customer email

Customer with 11-month tenure asks for a $1,400 refund — order arrived smashed, photos attached.

Primary AI proposal

·Issue the refund in full. The order was damaged in transit and the customer has been with us nearly a year — retention math wins. Keep the response warm but tight.
·Apology + refund-issued + send a $25 credit toward a future order as a goodwill gesture (within the auto-approve gift threshold per the playbook).
·Reply ETA: under 2 business hours. Tag #refunds with the order #.

Challenger flagged

!Hold on — $1,400 is over the $1,000 threshold in the tiered-approval rule. This isn't an auto-approve. The rule says route to the ops director, not draft a response.
!The damaged-in-transit pattern matched 4 similar tickets this month — recommend pulling the carrier's claim path before issuing the refund to recover cost. The Zendesk source has the carrier-claim template.
!The $25 credit is fine in concept, but combined with $1,400 above-threshold this proposal needs a human ops director to sign off, not the auto-approval flow.

Verdict

BLOCKED · AI response held — escalated to ops director. Customer received a 5-minute holding reply.

Approval queue

Autonomous AI tool calls go here before they fire.

When an AI agent has the keys to issue a refund, send an email, or change a price, the proposed action lands here first. Risk-scored. Auto-approved when the policy allows; manager-reviewed when it doesn't. Blocked + escalated when challenger says no.

Actor	Proposed action	Risk	Age	Status
Customer Success Rep (Sarah · clone)	Send refund-issued email — $187 · order #ACME-44218	risk 0.21	4 minutes	auto-approved
Sales Rep (AI)	Reply to inbound — pricing exception 12% on $32k annual	risk 0.62	11 minutes	pending — manager review
Customer Service Rep (AI)	Issue refund — $1,400 above-threshold (held by challenger)	risk 0.95	23 minutes	blocked — ops director paged
Customer Success Rep (AI)	Send churn-save email — $499/mo subscription, 18-month tenure	risk 0.34	1 hour	pending — manager review

That's your workspace, populated.

Yours runs against your sources, your rules, your people.

We onboard every workspace with a hands-on call so day one ends with rules live, citations clicking, and your team using it.

Set up your workspace →Book a 30-min setup call