Trained retrieval index
Your docs and last-90-day ticket transcripts chunked, embedded and versioned. Rebuildable with one command.
Accepting September cohort · 2 slots
Your support inbox,quiet in ten days.
A production AI support agent for bootstrapped B2B SaaS — trained on your docs and the past ninety days of tickets, live inside your app in 10 business days for a fixed $1,997. If it doesn't deflect 30% of your tickets in the first thirty days, you get all of it back and keep the code.
No pitch deck. You leave the call knowing yes or no.
Drops into the helpdesk you already run
IntercomCrispHelpScoutZendeskPlainFrontFreshdeskChatwootTawkDiscourse
01Datasheet
The offer, stated as a spec.
No feature marketing. Six numbers describe exactly what you're buying.
Delivery time
10 days
Business days, kickoff to production. Not "sprints".
Fixed price
$1,997
50% deposit, 50% on Day 9. No hourly billing.
Refund threshold
30%
Ticket deflection inside 30 days, or your money back.
Training corpus
Your docs + 90 days of real tickets
Not a generic model with your logo on it. It learns the words your customers actually use, and the answers your team actually gives.
Ownership
Yours, in full
Source, prompts, evals, deploy scripts — handed off Day 10.
Post-launch
30 days hypercare
Included. Retainer after that is optional, month-to-month.
02Ships on Day 10
What lands in your repo.
A concrete list of deliverables — not a roadmap, not "AI capabilities". Working things you own.
Reply engine
Prompt chain with confidence gating. Answers cite the exact doc or ticket. Below threshold it escalates instead of guessing.
Escalation with context
Handoff into the chat tool you already use, carrying the full conversation and the agent's reasoning trace.
Guardrails file
Blocklist for refunds, pricing, promises and anything else you flag. Plain YAML you can edit. Ships with sane defaults.
Weekly report
Deflection rate, top questions, gaps in your docs, confidence distribution. In your inbox Monday 08:00, your timezone.
Handoff pack
Repository, deploy scripts, prompt library, eval suite, and a 20-minute Loom walking through every file in it.
03Ten-day timeline
Every day accounted for.
Roughly ninety minutes of your time across the whole build. No standups, no weekly syncs, no status theatre.
Day 0
You30-min discovery call. Scope and fit.
UsWritten kickoff. Access checklist.
Day 1
YouShare docs, ticket CSV, staging access.
UsBuild environment, ingest, first crawl.
Day 2–3
YouNothing.
UsKnowledge base built. Chunking and embeddings tuned to your corpus.
Day 4–5
YouAnswer three clarifying questions.
UsReply logic, escalation rules and integrations wired.
Day 6
You30-min staging demo. Poke it, break it.
UsLive walkthrough. Every issue you find gets logged.
Day 7
YouNothing.
UsFix everything from Day 6. Ship to pre-prod.
Day 8–9
YouSign go-live. 50% balance due.
UsProduction deploy. We watch the first 200 conversations.
Day 10
YouGet your dashboard login.
UsHandoff pack, 20-min Loom, hypercare begins.
Day 11–40
You15 min a week reading the report.
UsHypercare. Prompt tuning. Regression fixes.
Day 41
YouRead the baseline verification.
UsMeasure against the Day-0 baseline. Under 30%? Refund initiated.
04Live demo
The kind of answer that actually deflects.
Not "here's how AI works". A real customer question, answered the way your best support person would answer it.
Deflection anatomy
Cites the doc. Admits doubt. Escalates cleanly.
Every reply is grounded in a real source, respects a confidence threshold, and hands off with full context when it isn't sure. No hallucinated pricing. No invented features. No "I'm just an AI".
- Answers cite specific docs, so users trust them
- A confidence gate keeps hallucination near zero
- Escalation carries the whole transcript to a human
- Guardrails block the topics you flag — refunds, promises
Customer
How do I upgrade my plan? I'm on Starter and I need the Team seat, but the invoice keeps failing.
Agent
94% confident
Resolved in thread−1 human ticket
05Guarantee
We eat the risk of "AI didn't work".
Fixed price plus a performance floor. Miss the floor and you're made whole — no negotiation.
If your support ticket volume doesn't drop by 30% within 30 days of go-live, you get a full refund of the $1,997 — and you keep the chatbot code and dashboard access.
BaselineYour ticket volume in the 30 days before launch.
MeasurementYour analytics — Intercom, Crisp, HelpScout — not ours.
TriggerYou initiate on Day 45. No approval needed from us.
PayoutBack to the original payment method within 5 business days.
ExclusionA major product change mid-run resets the clock. Fair both ways.
In writingThe clause is in the SOW you sign, not just on this page.
06The alternatives
What you'd otherwise do — and what it costs.
We're not the cheapest and not the biggest. We're the one that ships in ten days, priced flat, with a refund clause behind it.
| Criterion | Intercom Fin | AI agency | DIY weekend | Ten-Day Agent |
|---|---|---|---|---|
| Price | $99+/seat/mo, forever | $15k–$50k | ≈$4k in engineer time | $1,997 one-time |
| Timeline | Instant, generic | 2–6 months | 2–6 weeks, slipping | 10 business days |
| Trained on your data | Docs only | Yes, custom | You decide | Docs + 90d of tickets |
| Production-ready | Yes, generic | Eventually | Rarely first try | Day 9 |
| Performance guarantee | None | Rare | N/A | 30% or refund |
| Code ownership | Vendor lock-in | Sometimes | Yours | Yours, full source |
| Ongoing cost | $99+/seat/mo | Negotiated retainer | Your maintenance | Optional $497/mo |
↔ swipe to compare
07Real questions
What buyers actually ask on the discovery call.
If your objection isn't here, bring it to the call. We answer straight, or we say no.
Isn't $1,997 a lot for a chatbot?
Compared to what? Intercom Fin runs roughly $1,200/year per seat — and still doesn't know your product, your docs or your escalation rules. Our one-time fee is about twenty months of Fin, and you end up owning an agent trained on your specifics with a refund clause behind it. If Fin's generic answers are good enough for you, use Fin. Genuinely.
How does the 30% guarantee actually work?
Baseline is your ticket volume in the 30 days before we go live. We measure through your existing analytics — Intercom, Crisp, Zendesk, whatever you already run. Not our numbers. On Day 45, if deflection is under 30%, you initiate the refund and $1,997 is back with you inside five business days. You keep the code. No approval required from us.
In practice, comparable deployments land at 40–60%. If we miss 30%, we got something wrong.
We already tried a chatbot and it didn't work.
Very common, and usually three causes: it was trained on marketing pages instead of product docs; it had no confidence gate, so it guessed; and it never saw real ticket data, so it didn't speak your customers' language. We fix all three. If we run discovery and conclude your case genuinely can't be improved, we say so and turn the project down.
Do I need to change my chat tool?
No. We integrate with what you already run — Intercom, Crisp, HelpScout, Zendesk, Plain, Front, Freshdesk, Tawk, Chatwoot. If it has an API or an embeddable widget, we wire it. Send us your docs URL and we'll confirm within two hours.
What happens after the 30 days of hypercare?
Two paths. A: you take the code, we hand off, you maintain it — that's the default and it costs nothing. B: retainer at $497/month for ten hours of tuning, monitoring and one new integration slot per quarter. Month to month, cancel any time. Roughly 40% of buyers take the retainer.
Who is this not for?
Enterprise with procurement, legal review or SOC2/HIPAA requirements — we're not set up for it and we'll refer you on. Pre-launch products, or anything under 100 tickets a month — not worth your money yet, come back later. Consumer apps with millions of low-value tickets — different economics, different service.
Can I talk to an existing customer first?
Straight answer: this productized offer is new, so there's no reference customer to call yet. The live demo above runs the same stack, and the Client 01 case study is in progress. If you'd like to be Client 01, it's $500 off in exchange for an on-record case study when we're done.
How do I actually pay?
Stripe for USD, Razorpay for INR, Wise for anything else. 50% deposit when you sign the SOW ($997), 50% on production deploy (Day 9). Invoiced by RavenEdge Pvt Ltd, with GST for Indian clients.
2 slots left · September cohort
Book a 30-min call.No pitch.
Describe your support setup and your top three ticket types. I'll tell you whether we can hit 30% on your specific stack — or point you somewhere better. Either way you get an hour of your week back.
Prefer async? tushar@ravenedge.in — replies within a day.