Product Updates8 min readby José

Build in Public, Week 2: The Demo, 9 Viral Loops, and Taming Slack

Week 2 of building Sherlock Calls: the interactive demo shipped, 9 viral growth loops designed, Slack responses made production-grade with full thread history, and 300+ SEO combination pages generated.

TL;DR — The short answer

  • 1

    Week 2: the interactive demo shipped, 9 viral growth loops designed and implemented, and Slack responses made production-grade with full thread history and correct mrkdwn formatting.

  • 2

    Multi-turn investigations now maintain full context across Slack messages — the product works as a real conversation, not a one-shot query.

  • 3

    300+ combination pages auto-generated for every provider pair — an SEO asset that compounds with zero ongoing maintenance.

The interactive demo: showing the investigation, not describing it

The demo is the hardest thing to get right for a Slack-native product. You cannot write a paragraph that conveys what it feels like to watch Sherlock pull Twilio error codes, correlate them with ElevenLabs latency data, and deliver a sourced case file in a Slack thread. You have to show it.
The demo at /demo is a Slack-like interface with real AI-generated responses. The same Anthropic tool-use pipeline that runs in production generates the responses — not canned screenshots. Six pre-loaded investigation topics cover the most common voice AI failure modes: dropped calls, TTS latency spikes, billing anomalies, webhook failures, cross-provider correlation. Tool calls fire, findings accumulate, the case file appears in real time.
No login required. A skeptical voice AI engineer should be able to evaluate the product in 90 seconds. We added a CTA popup after 3 questions: 'Add to Slack / Try again / Go to homepage'. The demo is a conversion device — not a product preview.

9 viral growth loops: turning investigations into installs

A Slack-native B2B tool has a natural viral surface: every response is visible to the team. Every investigation ends in a thread that teammates can read. We designed 9 explicit growth loops to turn that visibility into installs.
Loop A: Case Cards — After any successful investigation, Sherlock generates a public shareable case card at /cases/[id]. Sharing your first case earns +5 credits.
Loop B: Viral Slack footer — Every response ends with 'Case closed · Invite a teammate → · Add a provider →'. Every investigation is a team-expansion event.
Loops C+D: Digest and provider-role invites — Weekly digests include 'Forward to #support / #revops' buttons. When Sherlock identifies a provider root cause, it suggests inviting the teammate who owns that provider.
Loop G: Vendor Ticket Export — Export any case file as a pre-filled support ticket for Twilio, ElevenLabs, Vapi, or Genesys. Saves 20–30 minutes of copy-paste per escalation.
Credit burn email sequence — 3 emails triggered at 50%, 10%, and 0% credit balance. Prevents lockout. Drives upgrades at the moment the team has proven the tool works.
Referral program (/partners) — 20% lifetime recurring commission on referred workspaces.

Slack: from working to production-grade

The Slack integration in v0.1 worked. The Slack integration in v0.6 was production-grade. The gap between those two statements is larger than it sounds.
Full thread history. In v0.1, Sherlock only saw the current message. In v0.6, Sherlock loads the complete conversation thread before responding. Multi-turn investigations now maintain context — say 'dig deeper on the ElevenLabs latency issue you found' without re-explaining which issue, which call batch, which time window. The investigation becomes a conversation.
Slack mrkdwn formatting. Slack uses its own markdown dialect — bold not bold, backtick code blocks, > for quotes. Sending standard Markdown to Slack renders raw characters: visible asterisks, no formatting applied. Sherlock's responses now render correctly in Slack.
Serverless keep-alive. Vercel serverless functions have a maximum execution duration. Long investigations were timing out before the response finished. The Slack event handler now uses waitUntil to stay alive during long investigations.
Credit exhaustion mid-stream. Running out of credits mid-investigation now stops tool calls immediately and delivers a partial answer with an explanation — not a silent termination.

300+ combination pages and provider marketing pages

Sherlock can investigate any combination of connected providers simultaneously. We auto-generated /for/combining/[slug] pages for every provider pair — over 300 combinations at current provider count. Each page is provider-aware: 'Sherlock for Twilio + ElevenLabs' explains exactly what cross-provider investigations look like for that pair, what failure modes it can identify, and what a real investigation output looks like.
These pages capture long-tail search intent no single provider page can capture: engineers searching 'Twilio ElevenLabs debugging tool' or 'how to correlate Retell calls with HubSpot'. 300+ pages, zero ongoing maintenance — the combination logic runs at build time from the provider data object.
We also launched dedicated /for/[provider] pages for all 8 active integrations. Provider-aware, engineer-readable content — not generic SaaS copy.
Week 3 covers the biggest technical changes yet: AES-256-GCM encryption, ClickHouse and PostgreSQL integrations, 4 free voice AI tools, transparent token billing, and the Admin Activity Monitor.

Explore Sherlock for your voice stack

TwilioElevenLabsVapiRetell AIBland AIGenesys

Frequently asked questions

How does the Sherlock Calls interactive demo work?

The demo at usesherlock.ai/demo is a Slack-like interface showing a live Sherlock investigation with real AI-generated responses. No login required. Ask about 6 pre-loaded investigation scenarios and see tool calls, findings, and the sourced case file in real time. A CTA appears after 3 questions: add Sherlock to Slack, try again, or visit the homepage.

Does Sherlock maintain conversation history in Slack threads?

Yes. As of v0.6, Sherlock loads the complete conversation thread before every response. Multi-turn investigations maintain full context — follow up on a previous finding without re-stating context. Thread history is loaded via the Slack API and passed to Anthropic as part of the investigation.

Share

Ready to investigate your own calls?

Connect Sherlock to your voice providers in under 2 minutes. Free to start — 100 credits, no credit card.

Start for free
← Back to the blog