# Owlish

> Owlish is a no-code AI agent platform for customer support. Businesses ingest their knowledge (websites, PDFs, FAQs) into Owlish, customise the agent's brand and persona, and deploy across their website, social channels, and team chat apps. Agents ground answers in your content, cite sources when enabled, and hand off cleanly to human operators when needed.

The product is built by Chevvi Pty Ltd (Sydney, Australia). Marketing site at https://owlish.bot, console at https://console.owlish.bot, embeddable widget at https://widget.owlish.bot, OpenAPI specification at https://owlish.bot/openapi.json, and public API reference at https://api.owlish.bot/docs.

## Documentation

- [Welcome to Owlish](https://owlish.bot/docs/): Overview of what Owlish does and where to start.
- [Build your first agent](https://owlish.bot/docs/quick-start/build-your-first-agent): 5-minute walkthrough — paste your website, let Owlish discover your brand, attach sources, chat with the agent.
- [Concepts](https://owlish.bot/docs/quick-start/concepts): The four primitives — agents, knowledge, sessions, channels — and how they relate.

### Agents

- [Create an agent](https://owlish.bot/docs/agents/create): The 5-step wizard (URL + use case → discovery → sources → review → create).
- [Playground](https://owlish.bot/docs/agents/playground): The agent's main page — Content / Brand / Behavior / Model tabs plus a live test chat.
- [Tone and fallbacks](https://owlish.bot/docs/agents/tone-and-fallbacks): Writing instructions that set voice, refusal behaviour, and unknown-question handling.

### Knowledge base

- [Knowledge base overview](https://owlish.bot/docs/knowledge-base/overview): Folders, sources, processed-MB caps per plan, and the citation pipeline.
- [Add a website](https://owlish.bot/docs/knowledge-base/websites): Crawl marketing sites and help centres; allow/exclude patterns; scheduled re-sync.
- [Add files](https://owlish.bot/docs/knowledge-base/files): Upload PDFs, DOCX, CSV, TXT, Markdown. OCR fallback for scanned PDFs.
- [Direct Response](https://owlish.bot/docs/knowledge-base/direct-response): Paste a question + canonical answer for FAQ-style content.
- [Citations & re-training](https://owlish.bot/docs/knowledge-base/citations): How citations are produced, debugging wrong answers, when re-sync runs.

### Deploy

- [Web widget](https://owlish.bot/docs/deploy/widget): Embed snippet, allowed domains, citation toggle. Available on every plan.
- [Customize the widget](https://owlish.bot/docs/deploy/widget-customization): Brand, content, behaviour — all configured on the agent's Playground tabs.
- [Slack](https://owlish.bot/docs/deploy/slack): OAuth install, DMs, channel mentions, threaded replies. Available on Growth and above.
- [Microsoft Teams](https://owlish.bot/docs/deploy/teams): Tenant install, personal chats, channel mentions, and threaded replies. Available on Growth and above.
- [Email](https://owlish.bot/docs/deploy/email): Give an agent its own support address; forward your inbox and it answers customer email automatically. Custom domains and delivery health. Available on Growth and above.
- [Discord](https://owlish.bot/docs/deploy/discord): Coming soon.

### Helpdesk

- [Conversations](https://owlish.bot/docs/helpdesk/conversations): The Inbox (Open / Resolved) and Live conversations views — every session, every channel.
- [Human handoff](https://owlish.bot/docs/helpdesk/human-handoff): Triggers (visitor asks, agent decides, operator pulls), Claim / Barge in / Resolve / Return to AI flow.

### Settings

- [General](https://owlish.bot/docs/settings/general): Workspace name, ID, cap-reached fallback, transfer ownership, delete workspace.
- [Members](https://owlish.bot/docs/settings/members): Invite teammates, Owner / Admin / Operator roles, seat limits.
- [API keys](https://owlish.bot/docs/settings/api-keys): Create, scope (Sources Read/Write/Delete), revoke.
- [Privacy & DSR](https://owlish.bot/docs/settings/privacy): Visitor data subject requests — search, delete, and controlled anonymization workflows (GDPR / CCPA).

### Billing

- [Plans](https://owlish.bot/docs/billing/plans): Free ($0, 0.5 MB KB, 7-day retention), Starter ($49/mo, 10 MB KB, 30-day retention), Growth (20 MB, 90 days), Scale (40 MB, 365 days).
- [Usage & change plan](https://owlish.bot/docs/billing/usage): Current usage meters, change plan, add-ons, annual vs monthly, invoices.

### Skills

- [Skills](https://owlish.bot/docs/skills/overview): Two skills ship today — human handoff and email escalation. Lead capture and web search are on the roadmap.

### Owlish developer resources

- [Owlish developer platform](https://owlish.bot/docs/developers/overview): REST API, scoped authentication, MCP setup, rate limits, and current programmable boundaries.
- [Owlish OpenAPI specification](https://owlish.bot/openapi.json): OpenAPI 3.1 JSON with unique operation IDs, typed inputs and responses, and per-operation required scopes.
- [Owlish API catalog](https://owlish.bot/.well-known/api-catalog): RFC 9727 Linkset connecting the public API endpoint to its OpenAPI description, human documentation, OAuth metadata, and MCP manifest.
- [Owlish interactive API reference](https://api.owlish.bot/docs): Scalar reference for the knowledge-folder and knowledge-source REST API.
- [Owlish OAuth protected-resource metadata](https://api.owlish.bot/.well-known/oauth-protected-resource/mcp): RFC 9728 metadata for the MCP resource with authorization-server discovery and supported least-privilege scopes.
- [Owlish MCP server manifest](https://owlish.bot/mcp/server.json): MCP Registry-compatible server.json for the Streamable HTTP endpoint at https://api.owlish.bot/mcp.

### Account & support

- [Your account](https://owlish.bot/docs/account): Profile, data export, account deletion.
- [Help & support](https://owlish.bot/docs/help): Contact (support@owlish.bot), find an answer, report a doc problem.

## For AI agents

- [Agent Onboarding Skill](https://owlish.bot/agent-onboarding/SKILL.md): Anthropic-format Agent Skill. Fetch this to onboard a user to Owlish — sign-up walkthrough, the 5-step agent wizard, API-key flow, and source-management automation via the REST API or MCP server. Clearly marks which steps require the human and which the agent can drive end-to-end.
- [Agent Onboarding Facts](https://owlish.bot/agent-onboarding.json): JSON summary of signup, Free evaluation, Growth/Scale API access, self-serve scoped keys, MCP transport, and the absence of a public API sandbox.

## Marketing

- [Home](https://owlish.bot/): Product overview, channels, use cases.
- [What it is](https://owlish.bot/what-it-is/): Deeper tour of the product — what ships today, what's grounded vs. generated, who it's for.
- [Pricing](https://owlish.bot/pricing/): Plan comparison, feature matrix, FAQs.
- [Owlish Partners](https://owlish.bot/partners/): Early-access partner program for agencies, consultancies, and software companies selling white-label AI customer support services with Owlish as the platform layer.
- [Roadmap](https://owlish.bot/roadmap/): Public roadmap across phases; what's shipped, in progress, and next.
- [About](https://owlish.bot/about/): Why Owlish exists, principles, founder, and Chevvi the company.
- [Contact](https://owlish.bot/contact/): Direct lines for sales, support, and the founder.

## Blog

- [Owlish blog](https://owlish.bot/blog/): Practical guides, product notes, comparisons, and support-operations playbooks for grounded AI customer support.

- [AI Chatbot for Notion Knowledge Bases: A Safe Public-Docs Setup](https://owlish.bot/blog/ai-chatbot-for-notion/): A practical Notion source-boundary guide for customer-support chatbots. Includes a Public–Private–Proof gate, four-route worksheet for public pages, approved exports, Direct Responses, and handoff, plus a six-prompt citation and refusal test.

- [How to Build an AI Chatbot for Developer Documentation](https://owlish.bot/blog/ai-chatbot-for-developer-documentation/): A technical support guide for public API documentation, SDKs, migrations, and changelogs. Includes a Developer Documentation Answer Contract, an eight-prompt release gate for version conflicts and refusal behaviour, source hierarchy, cited-answer checks, and a clear private-debugging handoff boundary.

- [6 Front Alternatives for AI Customer Support Teams](https://owlish.bot/blog/front-alternatives/): Commercial buyer guide for teams using or evaluating Front. Separates the three real decisions—keep the shared inbox, layer a source-grounded AI answering front line, or migrate to a fuller service suite—and includes an original Front Workflow Retention Map plus a reproducible Keep–Layer–Migrate proof sheet for source evidence, handoff, account-data boundaries, and billing events. Compares Owlish, Help Scout, Intercom Fin, Zendesk, Missive, and Crisp with current official-source pricing notes, clean public pricing-page screenshots, candid no-native-Front-integration positioning for Owlish, and a trademark disclaimer. Checked August 2026.

- [How to Build an AI IT Helpdesk Bot Without Admin Access](https://owlish.bot/blog/ai-it-helpdesk-bot/): Internal-support playbook for IT managers that distinguishes approved knowledge answers from identity-bound status checks, privileged administrative actions, and security incidents. Includes the original IT Request Permission Ladder, eight-prompt pilot, source-scope and escalation workflow, and candid guidance on where Owlish fits as a cited knowledge and handoff layer rather than an ITSM or identity-action system. Checked August 2026.

- [AI Chatbot Prompt Injection: A Support-Team Safety Guide](https://owlish.bot/blog/ai-chatbot-prompt-injection/): A defensive guide for customer-facing AI support agents. Includes a five-lane Support Agent Input & Source Boundary Map, a seven-prompt release gate, source-scoping guidance, citation review, and human handoff rules without claiming a prompt makes an agent immune.

- [AI Chatbot vs Live Chat: A Support Routing Framework](https://owlish.bot/blog/ai-chatbot-vs-live-chat/): A buyer and support-operations guide for deciding which requests deserve a cited AI answer, a live human, or an honest after-hours handoff. Includes the copyable Evidence–Judgment–Access routing card, a 12-request worksheet, and an outside-hours acceptance check.

- [How to Add an AI Chatbot to a HubSpot Website](https://owlish.bot/blog/ai-chatbot-for-hubspot/): A practical HubSpot CMS rollout for a cited public-knowledge support widget, including a native-versus-separate support decision card and a 12-prompt acceptance run.

- [Website Chatbot Placement: A Support-First Guide](https://owlish.bot/blog/website-chatbot-placement/): A page-by-page method for placing a public support widget where its answer source, human route, and mobile clearance all hold up.

- [AI Chatbot Client Onboarding Checklist for Agencies](https://owlish.bot/blog/ai-chatbot-client-onboarding/): A client-safe launch packet for agencies: map approved sources to answer boundaries and human routes, test the published chatbot, and assign post-launch change ownership.

- [Blog](https://owlish.bot/blog/): Notes from the Owlish team — product updates, deflection patterns, operator workflows.
- [Customer Support Knowledge Base Template for AI Chatbots](https://owlish.bot/blog/customer-support-knowledge-base-template/): Template-first knowledge workflow for teams preparing an AI support agent. Includes a copyable AI Support Source Map covering customer intent, answer lane, canonical source, risk, owner, update trigger, citation probe, and handoff condition; a worked policy example; a 12-question published-channel launch test; and practical guidance on choosing website, file, Direct Response, and human-handoff lanes without exposing private-data or exception decisions.
- [RSS feed](https://owlish.bot/rss.xml): Subscribe to Owlish Blog updates.
- [How to Add an AI Chatbot to a Framer Website (2026)](https://owlish.bot/blog/ai-chatbot-for-framer/): Practical Framer chatbot setup guide for founders, agencies, and small support teams. It explains when to use sitewide Custom Code versus a provider-supported inline Embed, then adds the original Framer Support Widget Placement & Source Map (placement, safe question, canonical source, and out-of-scope route) and nine-prompt published-site test for cited answers, reworded questions, policy details, unknowns, stale content, account-data boundaries, human handoff, allowed domains, and phone layout. Includes Framer-specific custom-code mechanics, no-code Owlish positioning, an honest no-authenticated-account-actions boundary, and a launch checklist. Checked against current Framer documentation, August 2026.
- [6 Help Scout Alternatives for AI Customer Support (2026)](https://owlish.bot/blog/help-scout-alternatives/): Commercial buyer guide for teams using or evaluating Help Scout that separates the three real decisions — keep the shared inbox and Docs, layer a standalone AI answering front line, or migrate to a deeper help desk. Includes a practical Keep / Layer / Migrate decision map, a reproducible 20-question switch test covering source evidence, refusal, handoff, account-specific boundaries, and billing events, a cost-shape worksheet, six persona-based alternatives (Owlish, Intercom Fin, Zendesk, Freshdesk, Gorgias, and Tidio Lyro), current official-source pricing notes, real public product-page screenshots, a clear note that Owlish has no documented native Help Scout integration, and a trademark disclaimer. Checked August 2026.
- [6 Crisp Alternatives for AI Customer Support Teams (2026)](https://owlish.bot/blog/crisp-alternatives/): Commercial buyer guide for teams using or evaluating Crisp. Separates the real decision—keep the shared inbox, layer a source-grounded AI front line, or migrate to a deeper service suite—and includes an original Crisp Exit / Keep Worksheet, current official-source pricing notes, a contact-export and conversation-export migration check, six persona-based alternatives (Owlish, Intercom Fin, Zendesk, Tidio Lyro, Chatbase, and SiteGPT), clean public product-page screenshots, and a trademark disclaimer. Checked August 2026.
- [How to Add an AI Chatbot to Your Webflow Site (2026)](https://owlish.bot/blog/ai-chatbot-for-webflow/): Practical Webflow chatbot setup guide for founders, agencies, and support leads — choosing between a sitewide Footer-code widget and an in-page Code Embed, checking the Webflow plan requirement, defining the support boundary before ingestion, preparing one current source for each answer, installing a provider script safely, and running the original Webflow Support Widget Launch Card (scope, citation, contradiction, no-answer, account-data, handoff, and real-page/mobile/domain checks). Includes Webflow-specific custom-code limits and testing gotchas, an FAQ, and where Owlish fits as a no-code website/PDF support agent with citations, domain controls, and human handoff, while honestly excluding live account actions without a verified integration. Checked against current Webflow documentation, July 2026.
- [AI Email Support: Automate Replies Without a Bot Loop](https://owlish.bot/blog/ai-email-support-automation/): Channel-workflow guide for turning an existing support inbox into a controlled AI support lane rather than a blanket autoresponder. Includes an original four-lane Email Reply Ladder (ignore automated mail, send a cited answer, draft for review, or hand off), the three checks an auto-send reply must pass (scope, source, stakes), a no-send list for account-specific and high-stakes messages, a Gmail/Outlook forwarding and thread-continuity test pack, authentication and deliverability checks, a 25-thread review gate, and where Owlish fits (grounded replies with citations, existing-inbox forwarding, threaded answers, delivery-health visibility, and human handoff) and does not (live account data, binding decisions, or a full enterprise ticketing suite). Checked July 2026.
- [Why Your AI Chatbot Gives Outdated Answers (And How to Fix the Sync)](https://owlish.bot/blog/ai-chatbot-outdated-answers/): Knowledge-base workflow guide on why RAG-based AI support agents answer from a stale index instead of live content — the gap between editing a source and re-embedding it, three distinct failure modes (structural staleness where the source changed and the index didn't, orphaned content where a superseded document never gets removed and two chunks disagree, and undated content with no timestamp for time-aware ranking to use), how to catch it (read citations, check last-synced timestamps, spot-check monthly), and the three real sync levers with their honest limits (manual "Sync now" with a 30-day per-source cooldown that only starts after a successful run, scheduled auto-sync gated by plan and limited to Website-crawled folders only — Off on Free/Starter, Monthly on Growth, Weekly or Monthly on Scale, not available for uploaded documents or Direct Response answers — and a logarithmic time-decay ranking factor with an 0.85 similarity floor that quietly deprioritizes aging content as a safety net, not a substitute for updating the source). Includes a practical freshness checklist, an FAQ, and where Owlish fits (citations on by default, manual sync, Growth/Scale scheduled website sync, time-aware ranking, manual content-date correction) and where it doesn't (document-management systems needing webhook-triggered re-ingestion). Checked against Owlish's own ingestion architecture and pricing, July 2026.
- [How to Audit Your AI Support Agent's Conversations: A Practical QA Checklist](https://owlish.bot/blog/ai-support-agent-conversation-audit/): Support-operations playbook for the ongoing habit most teams skip — actually reading a sample of live AI conversations on a schedule, against a rubric, distinct from pre-launch testing or a metrics dashboard. Opens with the perception-reality gap (Laivly's 2026 AI Deployment Index and a Sinch survey, via CX Dive July 2026: two-thirds of CX leaders call their AI project a success even as 53% went over budget and 43% are delayed, and three-quarters of enterprises have rolled back an AI deployment, top reasons being data exposure, hallucination/brand risk, and inability to diagnose what went wrong), why CSAT and dashboards can't substitute for reading transcripts (Intercom's July 2026 research: CSAT captures under 10% of conversations; Verint/Calabrio: manual contact-center QA typically covers 1-3% of interactions), four failure classes only a transcript read catches (confidently-wrong-but-cited, resolved-by-the-numbers-not-for-the-customer, tone/policy drift, mistimed escalation), a stratified weekly sampling method (escalations, flagged/low-confidence conversations, top questions by volume, a random baseline), a five-point rubric (grounding, tone/policy fit, escalation timing, resolution truth, repeat pattern), closing the loop from a bad answer to a fixed one, and a cadence guide sized to 30-60 minutes a week with a named owner. Covers where Owlish fits (Chat Logs per-agent session browser with inline "Used N sources" citations, the analytics dashboard's auto-flagged "Conversations needing review" list, the Helpdesk Inbox for escalations, and Revise Answer for correcting and pinning a fixed reply) and where it doesn't (no CSAT/star-rating capture, no manual flag-for-review workflow, no built-in audit-scoring UI). Checked July 2026.
- [Drift Alternatives (2026): Where to Go Now That Drift Is Being Sunset](https://owlish.bot/blog/drift-alternatives/): Comparison/buyer guide built around Clari + Salesloft's March 5, 2026 confirmation that Drift is being sunset, with 1mind named the exclusive successor — the official Salesloft newsroom announcement, drift.com now redirecting to Salesloft's Chat Agents page, and the August 2025 OAuth breach (via Cloudflare's own incident post) as backdrop. Splits "Drift alternative" into the two jobs Drift actually did (sales/pipeline chat vs. customer support chat) and routes each to a different tool: 1mind (the official successor — enterprise sales-AI, no public pricing, not a support tool), Owlish (grounded cited answers, human handoff, flat session pricing, for the support-chat job), HubSpot Service Hub (CRM-native, Breeze Customer Agent at $0.50/resolution on Pro/Enterprise only), Fin/formerly-Intercom ($0.99/outcome, now under Salesforce), Zendesk (full help desk), and Crisp (flat per-workspace widget with its own "Hugo" AI agent). Includes a persona-based summary table, a how-to-choose section by which job Drift was doing for you, an FAQ, sources, and a trademark disclaimer. Includes real screenshots of each vendor's public site captured July 2026. Checked July 2026.
- [How to Add an AI Chatbot to Your Wix Site (2026)](https://owlish.bot/blog/ai-chatbot-for-wix/): Practical setup guide for adding a grounded AI support agent to a Wix site — what changed when Wix retired AI Site Chat for new sites in March 2026 in favor of Wix Smart Chat, the plan-tier gate that decides whether Custom Code runs on a published site at all, exactly where to paste the embed snippet (Settings → Custom Code → Body – end → All pages, then publish), the alternative HTML iFrame embed element, grounding the agent in Wix's JavaScript-rendered pages plus PDFs, Wix-specific gotchas (Spaces by Wix doesn't run custom code, a December 2025 Wix security change that broke some existing embeds, Stores/Bookings being account-specific systems the bot can't see by default), a launch checklist, an FAQ, and where Owlish fits (free web widget on every plan, domain allowlist from Starter, human handoff and shared inbox from Growth) and where it doesn't (Stores orders, Bookings appointments, or the Spaces mobile app). Checked against Wix's own Help Center and developer docs, July 2026.
- [How to Build a Telegram Support Bot](https://owlish.bot/blog/telegram-support-bot/): Practical channel workflow guide for building a Telegram support bot — Telegram's default bot privacy mode and when to disable it, setting up a bot with @BotFather (no app review required), when Telegram Business fits a solo operator, training on trusted sources instead of raw group history, designing a handoff packet for a channel with no Slack-style threads, measuring resolution instead of reply volume, and where Owlish fits (native Telegram deployment on Growth+, auto-registered webhook, shared knowledge base and citations, human handoff into the same helpdesk inbox as every other channel) and where it doesn't (Bot API features like paid broadcast channels or in-chat payments). Cites Telegram's own Bot API docs and Pavel Durov's March 2025 1-billion-user announcement, checked July 2026.
- [How to Build a Google Chat Support Bot](https://owlish.bot/blog/google-chat-support-bot/): Practical channel workflow guide for building a Google Chat support bot — why Google Chat is almost always an internal channel (Workspace Business/Enterprise accounts only), the real Google Cloud setup (project, Chat API, service account JSON, HTTP endpoint), the private-vs-public Marketplace distinction that lets an internal bot skip Google's review entirely, DMs vs. mention-triggered Spaces, training on IT/HR/ops sources instead of raw Chat history, handoff design, and where Owlish fits (Google Chat channel on Growth+, service-account-JSON connection, shared knowledge base and citations, one helpdesk inbox across channels) and where it doesn't (Workspace admin automation). Checked against Google's own Chat API and Marketplace publishing docs, July 2026.
- [The 8 Best AI Chatbots for Customer Support in 2026 (Compared)](https://owlish.bot/blog/best-ai-chatbots-for-customer-support/): Ranked, persona-based buyer's guide comparing 8 AI customer support chatbots — Owlish, Intercom Fin, Zendesk AI, Chatbase, SiteGPT, Tidio Lyro, Gorgias, and Help Scout — with pricing and features checked against each vendor's own pricing page, docs, or help center in July 2026 rather than marketing copy or third-party summaries. Frames the three real differentiators (flat vs. per-outcome/per-resolution billing; admin-only source tracking vs. a genuine customer-facing citation trail; a dropped ticket vs. a clean full-context handoff), gives each tool a persona-based "best for" (Owlish for grounded cited answers and clean handoff on a flat plan; Intercom Fin for a unified omnichannel inbox at $0.99/outcome; Zendesk AI for teams already on the Zendesk suite, flagging that its per-resolution overage isn't published on its own pricing page; Chatbase for budget-conscious DIY builders watching the credit meters; SiteGPT for solo founders wanting a flat-rate website bot; Tidio Lyro for small ecommerce teams bolting AI onto live chat; Gorgias for Shopify brands needing live order-data lookups; Help Scout for a human-first helpdesk with AI as an opt-in add-on), a 3-step demo-proof test (grounding, handoff, cost), an FAQ, sources, and a trademark disclaimer. Includes real screenshots of each vendor's public site captured July 2026. Checked July 2026.
- [Agentic AI in Customer Service: What It Actually Means (and How to Deploy It Safely)](https://owlish.bot/blog/agentic-ai-customer-service/): News-to-evergreen buyer-education explainer on agentic AI for customer support — what "agentic" actually means (an agent takes an action toward a goal; a chatbot only answers, with IBM/McKinsey/Zendesk definitions), the autonomy ladder mapped onto real support work (Gartner's four levels — Observe, Advise, Act with approval, Act autonomously — plus Salesforce's parallel maturity model, and why most support value sits on the lower rungs while most risk sits on the higher ones), "agent washing" as Gartner's own term for relabeled chatbots (40% of agentic AI projects canceled by 2027) and a four-question buyer test (does it act or only answer, where do answers come from, what happens when it doesn't know, how is autonomy bounded), the June 2026 vendor landscape as orientation not ranking (Zendesk AI agents, Intercom/Fin at $0.99 per outcome, Salesforce Agentforce's self-reported 85%, Ada, Sierra, Decagon), the reality-check tension between the forecasts (Gartner's 80%-by-2029, Salesforce's 79% of leaders) and customers losing patience (CX Dive: 60% repeat themselves only once, 14% fully trust AI on complex requests) plus the data-readiness gap (Microsoft: ~80% can't share data across teams), a five-step safe-rollout playbook (start grounded not autonomous, draw the action boundary explicitly, keep a human on the exceptions, disclose it's AI per EU AI Act Article 50 from 2 August 2026, measure resolution not containment), a FAQ, and where Owlish fits (deliberately Level-2 territory done well — grounded cited answers in a web widget plus Slack/Teams/Discord, clean human handoff from Growth, flat session-based pricing not per-resolution/outcome — Free $0 / Starter $49/mo or $39 annual / Growth $149/mo or $119 annual / Scale $449/mo or $359 annual) and where it doesn't (not a full enterprise contact-center suite; does not run autonomous back-office actions like refunds unattended). Cites IBM, McKinsey, Zendesk, Gartner, Salesforce, CX Dive, Microsoft, and the EU AI Act, checked June 2026.
- [How to Replace Canned Responses With an AI Support Agent (2026)](https://owlish.bot/blog/replace-canned-responses-with-ai/): Support-operations playbook for retiring canned responses (Zendesk/Gorgias macros, Intercom/Help Scout saved replies, Freshdesk canned responses) in favor of a grounded AI agent — the two jobs macros did (speed + approved wording) and why they break at scale (stale templates sent silently, wrong-macro picks past a few dozen, placeholder personalization that generates repeat tickets, and the resolution ceiling where a template can inform but never check an order or issue a refund), the critical split between the two things "replace with AI" means (customer-facing grounded self-service that makes the ticket disappear before it's created vs. an agent-facing copilot that drafts replies inside your helpdesk — Owlish is the first, not the second), the part of canned responses worth keeping (canonical exact-wording answers for refunds/security/legal, served as a grounded Direct Response the agent cites and edits live rather than a template an agent pastes), a three-bucket migration (informational → retire and ground in the upstream source; canonical → curated direct answer served verbatim with citation; action → keep in the human/agent workflow behind a clean handoff), the failure modes (importing macros verbatim teaches old policy at scale, deleting canonical control invites inconsistency and liability, measuring deflection instead of resolution, letting the knowledge base go stale, losing brand voice), how to tell it's working (repeat-contact rate, resolution not deflection, escalation quality, answer-to-source audits), and where Owlish fits (grounded cited answers from websites/files, Direct Responses for canonical wording, web widget + Slack/Teams/Discord, human handoff to a shared inbox, flat session-based pricing — Free $0 / Starter $49/mo or $39 annual / Growth $149/mo or $119 annual / Scale $449/mo or $359 annual) and where it doesn't (agent-assist reply drafting, or a back-office action engine). Cites Gartner, Zendesk's 2026 CX research, and the NBER "Generative AI at Work" study, checked June 2026.
- [How to Add an AI Chatbot to Your Squarespace Site (2026)](https://owlish.bot/blog/ai-chatbot-for-squarespace/): Practical how-to for adding an AI support chatbot to a Squarespace site — the one plan limit that decides whether you can add a chatbot at all (Code Injection requires the Core plan or higher and is not available on Basic; June 2026 plans are Basic, Core, Plus, Advanced), why Squarespace ships AI for building a site but no grounded customer-support agent, the two install patterns (sitewide Code Injection → Footer vs. a single-page Code Block) and why Footer injection before the closing body tag is the right path for a floating widget, deciding the bot's narrow job from real contact-form/inbox/Scheduling questions, reconciling pages so each topic has one canonical dated answer, grounding the agent in your site and PDFs, exactly where to paste the snippet (Settings → Advanced → Code Injection → Footer), restricting to your domain and testing in a private window against Squarespace's CDN cache, handoff rules, Squarespace-specific gotchas (caching, cookie consent, Acuity/Scheduling, Member Areas, mobile layout), what to measure, and where Owlish fits (web widget on every plan including Free, domain allowlist from Starter, human handoff from Growth) with the honest no-Commerce-order/no-Acuity-booking-lookup caveat. Cites Squarespace's code-injection docs and pricing, and CX Dive's 2026 CX trends, checked June 2026.
- [Per-Resolution AI Support Pricing: What Zendesk, Intercom Fin, and Ada Actually Charge](https://owlish.bot/blog/per-resolution-ai-pricing/): Pricing explainer and buyer guide on per-resolution (outcome-based) AI support pricing — what the model means (you pay a fixed amount, roughly $0.75–$2.00, each time the AI closes a conversation on its own), why everything hinges on the vendor's slippery definition of "resolution" (Intercom Fin: "no further help requested after Fin's last answer"; Zendesk: fully resolved with no human; some vendors a silence timer, which Ada calls the "containment trap"), what the recognizable vendors charge as of June 2026 (Fin $0.99/resolution with a 50/mo standalone minimum; Zendesk ~$1.50 committed / $2.00 pay-as-you-go with figures now gated to sales; Ada unpublished and custom; Help Scout $0.75; Gorgias ~$0.90–$1.00), the paradox that your bill rises as your AI improves (Ada's own words: the model can "punish success"), a dollars-and-cents walk-through of how the resolution definition swings a 10,000-chat month between $6,000 and ~$10,000, a five-question contract checklist, when per-resolution is actually the right call (high-volume high-deflection, spiky seasonal traffic), how to predict and cap your bill, the four pricing models compared (per-resolution, per-seat, per-conversation, flat/tiered), and where Owlish fits (flat tiered session pricing — Free $0 / Starter $49/mo or $39 annual / Growth $149/mo or $119 annual / Scale $449/mo or $359 annual, extra sessions $29 per +100 — metered by conversation not resolution, so improving the agent doesn't raise the bill; grounded cited answers and human handoff) and where it doesn't (full enterprise contact center with telephony, CRM routing, and standardized outcome billing). Cites Intercom/Fin, Zendesk, Ada, and Computer Weekly, checked June 2026.
- [How to Design Your AI Support Agent's Voice and Tone](https://owlish.bot/blog/ai-support-agent-voice-and-tone/): Conversation-design how-to for an AI support agent's voice and tone — why you design two separate systems (voice and tone live in the instructions/system prompt; facts live in the grounded knowledge base) and the rule that keeps you safe (instructions describe behavior, the knowledge base holds facts, so anything that changes when a price or policy changes belongs in a source not the persona), how to write the opening message that sets the whole tone (Intercom's A/B test lifting CSAT from 72.8% to 78.4% by warming only the greeting; say who's talking, set a realistic scope, signal the exit), writing the persona as concrete testable behavior rather than backstory (formality and length defaults, what to do with uncertainty, boundaries, brand phrasing, plus pinned exact-wording answers via Direct Response), matching verbosity to the question, designing the fallback where tone meets trust (refuse cleanly, cite what it knows, hand off with full context — with the Air Canada/Moffatt invented-policy case as the cost of guessing), situational tone for frustrated/locked-out/billing/pre-sales moments, a pre-launch voice test script, a lean copy-paste instructions template that contains zero product facts, and where Owlish fits (instructions field is explicitly the system prompt/personality kept separate from ingested sources, first-class greeting and suggested prompts, citations on by default, full-transcript human handoff, no-code, web widget and Slack, flat plans from $39/mo annual / $49/mo monthly) and where it doesn't (full enterprise contact center with telephony and CRM routing). Cites Intercom's June 2026 conversation-design A/B test.
- [AI Customer Support Hallucinations: Real Cases, New Rules, and How to Prevent Them](https://owlish.bot/blog/ai-customer-support-hallucinations/): Trust-and-safety guide on hallucinations in AI customer support — what they are (answers generated from training patterns rather than your sources), why they differ from stale or wrong-but-real retrieval, three real-world patterns (the invented policy behind the Air Canada / Moffatt tribunal ruling that made a fabricated refund policy legally binding in Feb 2024, the confident fabricated discount, and the plausible wrong procedure that quietly inflates ticket volume), what they cost (trust, legal exposure, and the EU AI Act Article 50 disclosure obligations that apply from 2 August 2026), a six-layer prevention framework (ground in your sources, refuse cleanly, show citations, hand off before improvising, pin high-stakes answers via Direct Response, keep sources current), a five-step loop for measuring your own hallucination rate, a seven-question buyer's checklist for vendors, and where Owlish fits (grounded cited answers, clean refusal, context-carrying handoff, pinned answers, flat session-based pricing from $39/mo annual / $49/mo monthly) and where it doesn't (full contact-center suite). Cites the EU AI Act, McCarthy Tétrault and AI Business on Moffatt v. Air Canada, and McKinsey's State of AI, checked June 2026.
- [Resolution Rate vs Deflection Rate vs Containment Rate: How to Read AI Support Metrics](https://owlish.bot/blog/resolution-rate-vs-deflection-rate/): Buyer-education decoder for the three automation metrics AI support vendors quote interchangeably — containment (did the conversation stay in the bot channel; contained sessions ÷ sessions entering the channel; the most inflatable because abandons and timeouts count), deflection (did any contact avoid a human across all channels), and resolution (was the issue actually solved end-to-end; Zendesk's fully-resolved ÷ handled formula). Shows why one deployment yields three numbers (an illustrative 80% / 65% / 45% walk-through), why containment needs a 24-hour re-contact adjustment (Decagon: inflates 5–15 points without it), benchmark ranges and why to distrust them, why question mix sets the ceiling, why outcome-based pricing (Intercom Fin $0.99/resolution, "fully solved without human intervention") is the market's tell that resolution is what counts, the cost-and-trust metric the headline hides (pair any number with repeat-contact and CSAT), a six-question checklist for getting the real figure out of a vendor before you buy, and where Owlish fits (grounded cited answers, honest handoff, flat session-based pricing not metered per resolution) and where it doesn't (full contact-center suite or outcome-based per-resolution pricing). Cites Zendesk, Decagon, Intercom, and CX Today, checked June 2026.
- [Tidio Alternatives for AI Customer Support (2026)](https://owlish.bot/blog/tidio-alternatives/): Honest buyer-focused comparison of six Tidio alternatives — Owlish, Intercom (Fin), Chatbase, Crisp, HubSpot, and Help Scout — with verified June 2026 pricing, why teams switch off Tidio (Lyro AI sold as a separate add-on stacked on the conversation plan, three separate quotas to track, and a steep self-serve gap from Growth at ~$49/mo to Plus at $749/mo), a short summary table, when Tidio is still the right choice (live-chat-first, low volume, fast setup), where Owlish fits as a grounded answering and handoff layer on flat session pricing with no separate AI meter (and where it doesn't — it is not a full live-chat or ticketing product), persona-based picks instead of one winner, an FAQ, sources, and a trademark note.
- [Freshdesk Alternatives for AI Customer Support (2026)](https://owlish.bot/blog/freshdesk-alternatives/): Honest buyer-focused comparison of six Freshdesk alternatives — Owlish, Zendesk, Zoho Desk, Intercom (Fin), Help Scout, and Gorgias — with verified June 2026 pricing, why teams switch off Freshdesk (the AI is a second and third meter: per-agent seats, plus the Freddy AI Agent billed in sessions — 500 included on Pro/Enterprise then $49 per 100 — plus the Freddy AI Copilot agent-assist as a separate $29/agent/mo add-on, with the good AI gated to the Pro $55 and Enterprise $89 plans and sessions hard to forecast), a short summary table, when Freshdesk is still the right choice (you need full ticketing and case management), where Owlish fits as a grounded answering and handoff layer on flat session pricing with no seat-plus-session meter (and where it doesn't — it is not a full ticketing CRM with case management or order management), persona-based picks instead of one winner (Zoho Desk for budget like-for-like, Zendesk for enterprise depth, Intercom Fin for the strongest agent, Help Scout for email-first calm, Gorgias for ecommerce), an FAQ, sources, and a trademark note.
- [AI Customer Self-Service: How to Build One That Actually Resolves (2026)](https://owlish.bot/blog/ai-customer-self-service/): Support-ops playbook for AI customer self-service that resolves instead of just deflecting — why demand is high but resolution is low (Gartner: only 14% of customer service issues are fully resolved in self-service, and ~34% of consumers have left a company that offered no self-service), why deflection is not resolution, why self-service is becoming the default front door (Gartner's self-service-and-live-chat-surpass-traditional-channels-by-2027 forecast; 74% of consumers expect 24/7), the four parts of self-service that resolves (a knowledge base good enough to answer from, answers not search results, citations so answers are verifiable, and a clean context-carrying human handoff), a what-to-self-serve vs what-to-route-to-a-human table, the mistakes that keep self-service stuck at 14% (optimizing deflection, letting the bot guess, stale content, dead-end failures, and hiding self-service — Gartner: 60% of agents fail to promote it), how to measure it (resolution rate, escalation quality, containment-with-CSAT, top unresolved questions), a rollout checklist, and where Owlish fits (grounded cited answers in a web widget, no-code ingestion, human handoff, flat session-based pricing) and where it doesn't (not a full ecommerce helpdesk or enterprise contact-center suite). Cites Gartner self-service research, Zendesk's 2026 CX Trends, and Shep Hyken's 2025 consumer study.
- [How to Add an AI Chatbot to Your WordPress Site (2026)](https://owlish.bot/blog/ai-chatbot-for-wordpress/): Practical how-to for adding an AI support chatbot to a WordPress site — the two install patterns (a dedicated plugin vs. a JavaScript embed snippet) and why most modern AI tools use the theme-proof snippet path, the WordPress.com vs. WordPress.org plan limit that blocks custom code on the free tier (custom code has traditionally required the Business plan, though WordPress.com has been opening plugin access to Personal/Premium — check your own plan), the four things a site chatbot needs (answers from real content, visible sources, clean handoff, no-code install), pulling real questions from contact forms and comments, grounding the agent in your pages, posts, and PDFs, exactly where to paste the snippet (a headers-and-footers plugin like WPCode, a theme footer-scripts field, or a page builder's custom-code panel — in the footer, never footer.php directly), restricting to your domain and clearing caches, handoff rules, WordPress-specific gotchas (caching, cookie consent, security plugins, multilingual, AMP), what to measure, and where Owlish fits (web widget on every plan including Free, domain allowlist from Starter, human handoff from Growth) with the honest no-WooCommerce-order-lookup caveat. Cites W3Techs WordPress market share, WordPress.com's 2025 plugin-plans change, and WPCode's 2M+ installs.
- [How to Find the Gaps in Your Support Knowledge Base](https://owlish.bot/blog/knowledge-base-gap-analysis/): Operations playbook for support knowledge base gap analysis — why a gap is any question you can't answer well (missing, stale, contradictory, or unfindable content), not just a missing article; where gaps concentrate (the long tail, account-specific questions, cross-team seams, recently changed things); the four data sources that reveal them (ticket/chat history clustered by intent, zero-result and no-click help-center searches, the AI agent's refusal and escalation log as highest signal, and citation spot-checks for contradiction gaps); how to rank fixes by frequency × cost of a wrong answer; closing the loop with an owner, a monthly cadence, and re-sync on every content change; why a grounded agent that refuses instead of guessing is a continuous, demand-ranked gap detector; and where Owlish fits (grounded cited answers, honest refusal and handoff, no-code ingestion) and where it doesn't (full enterprise service desk or dedicated knowledge-management suite). Cites Gartner's 80%-by-2029 agentic-AI prediction, Forrester's 2026 trust-erosion warning, and Zendesk's shift from ticket-based Content Cues to bot-conversation gap discovery.
- [AI Customer Support and Data Privacy: What to Check Before You Buy (2026)](https://owlish.bot/blog/ai-customer-support-data-privacy/): Trust and data-privacy buyer's guide for AI customer support — why privacy is now a buying criterion (consumers' confidence in companies' use of generative AI is low and most will abandon a brand over intrusive or inaccurate AI), the four-party data path a support message travels (browser, vendor, model provider, storage) and the two questions that matter (is it stored, and is it used to train models), the GDPR contract you need before turning an agent on (Article 28 DPA, sub-processor list, SCCs/Data Privacy Framework transfers, retention and deletion, data-subject requests), the EU AI Act Article 50 transparency obligation that applies to most support bots from 2 August 2026 (disclose customers are talking to AI), an eight-question vendor questionnaire, why answer accuracy (grounding plus citations) is itself a trust issue, and where Owlish fits (DPA with SCCs, published sub-processor list, US Google Cloud hosting with Vertex AI no-training default, AES-256, DSR tooling, grounded cited answers and handoff) and where it doesn't (no EU data residency; SOC 2 / ISO 27001 on the roadmap not yet certified; no HIPAA BAA today). Cites the EU AI Act, GDPR Article 28, Google Cloud Vertex AI data governance, and 2026 CX consumer-trust research.
- [Multilingual AI Customer Support: How to Answer Customers in Any Language](https://owlish.bot/blog/multilingual-ai-customer-support/): Practical guide to multilingual AI customer support — the two architectures (translate-then-answer vs reason-in-language) and why the second reads native, why one primary knowledge base is usually enough (translate only legal wording and region-specific facts), the failure modes that don't show in a demo (mixed-language/code-switching, citations pointing at source-language docs, weaker low-resource languages, formality registers, right-to-left rendering), a setup checklist, why handoff is harder across languages when you lack a language-matched operator, measuring quality per language instead of in aggregate, and where Owlish fits (answers in the customer's language via multilingual frontier models from a single ingested knowledge base, with citations and any-language handoff) and where it doesn't (no dedicated AI voice channel, no built-in per-language analytics, no parallel per-language KB management). Cites CSA Research and Intercom Fin's 45-language support.
- [Gorgias Alternatives for Ecommerce AI Support (2026)](https://owlish.bot/blog/gorgias-alternatives/): Honest buyer-focused comparison of six Gorgias alternatives for ecommerce support — Owlish, Re:amaze, Tidio, Intercom (Fin), Zendesk, and Help Scout — with verified June 2026 pricing, why teams switch off Gorgias (per-ticket helpdesk plans stacked with $0.90–$1.00 per-resolution AI that also counts as tickets, Shopify-first depth, macro-era automation), a short summary table, when Gorgias is still the right choice, where Owlish fits as a grounded answering and handoff layer with flat per-session pricing (and where it doesn't — it is not an order-management helpdesk), and persona-based picks instead of one winner.
- [AI Customer Support for SaaS: A Practical Setup Guide](https://owlish.bot/blog/ai-customer-support-saas/): Vertical setup guide for SaaS teams putting an AI agent in front of support — why SaaS support is version-sensitive and splits into trial vs paying customers, sorting tickets into four buckets (answerable from content, needs live account data, a bug/incident, product/sales feedback), grounding answers in docs that stay current with citations and scheduled re-sync, the questions an agent must never answer alone (billing disputes, security/compliance, anything during an incident) enforced by refusal, escalation rules for trials with buying-signal questions and churn signals on paying accounts, putting one agent across in-app/docs/email/Slack, the API and MCP server angle for developer products, where Owlish fits (grounded cited answers, refusal, handoff from Growth, flat per-session not per-resolution pricing) and where it doesn't (full ticketing/CRM, live account data without API wiring), and a post-launch scorecard built on verified resolution and trial-to-paid impact.
- [How to Add an AI Chatbot to Your Shopify Store (2026 Guide)](https://owlish.bot/blog/ai-chatbot-for-shopify/): Practical how-to for adding an AI customer-support chatbot to a Shopify store — what Shopify's native AI actually does (Sidekick is a merchant admin assistant; Shopify Magic drafts suggested replies in Shopify Inbox for a human to send; instant answers are static FAQs) and why none is a customer-facing autonomous agent, the four things a store chatbot needs (answers from real content, visible sources, clean handoff, no-code install), starting from real contact reasons, the live-order-data limit to plan around (content-trained chatbots can't see a specific order's live tracking unless integrated with Shopify's order APIs), grounding on shipping/returns/FAQ/product/sizing content, the theme.liquid embed steps (duplicate theme, paste before the closing body tag, restrict to your domains), handoff rules, what to measure, and where Owlish fits (web widget on every plan including Free; human handoff from the Growth plan) with the honest no-live-order-API caveat.
- [How to Reduce Customer Support Costs With AI (2026 Cost Playbook)](https://owlish.bot/blog/reduce-customer-support-costs-with-ai/): Operations playbook for cutting support costs with AI — why cost per resolution (not deflection percentage) is the number that matters, what support actually costs by channel, the four places AI genuinely reduces cost (answerable-tier deflection, after-hours coverage, shorter handle time, spike absorption) and the four costs it quietly adds (wrong answers, content upkeep, setup/maintenance, the work humans still do), a copyable spreadsheet cost model, the per-resolution pricing trap that bills you more as you succeed (with the Intercom Fin $0.99/resolution example), five moves that lower cost without lowering quality, how to measure real vs vanity savings, and where Owlish fits with grounded cited answers and flat (non per-resolution) pricing.
- [How to Choose AI Customer Support Software: A 2026 Buyer's Checklist](https://owlish.bot/blog/how-to-choose-ai-customer-support-software/): Criteria-based buyer's guide for evaluating AI customer support software — why "agent washing" makes the choice harder, and eight things to test on your own content during a trial (grounded answers with citations, knowledge ingestion and refresh, refusal behavior, human handoff context, channel coverage, real no-code setup time, the pricing unit you are metered on, and observability plus data handling), a short scoring table, how to match the tool category (helpdesk-native suites, AI-native website/document agents, full contact-center platforms) to your team and persona, and where Owlish fits as a grounded AI-native answering and handoff layer.
- [How to Reduce First Response Time in Customer Support (2026 Playbook)](https://owlish.bot/blog/reduce-first-response-time/): Operations playbook for cutting first response time (FRT) — how FRT is measured and why it is not resolution time, what counts as fast by channel in 2026 (live chat under 40s, social under 15 min, email under an hour), why FRT stalls structurally, seven ways to cut it with a grounded AI agent (front-of-queue answers, citations, after-hours coverage, routing, contextful handoff, saved replies, public SLAs), the failure modes that fake the number, and how to measure FRT honestly alongside resolution and CSAT.
- [How to Set Up AI Customer Support on WhatsApp](https://owlish.bot/blog/whatsapp-ai-customer-support/): Channel workflow guide for running an AI support agent on WhatsApp — the WhatsApp Business Platform and Cloud API basics, why the 24-hour customer service window makes reactive support the right fit (and templates the metered exception), how to train on the high-volume transactional intents customers actually send, grounding answers with checkable sources, designing fast human handoff before the window closes, measuring resolved conversations, and where Owlish (WhatsApp on the Scale plan) fits.
- [How to Train an AI Chatbot on Your Own Data (Without Fine-Tuning a Model)](https://owlish.bot/blog/train-ai-chatbot-on-your-own-data/): A practical 2026 guide to training a support chatbot on your own data — why "training" usually means retrieval (RAG) not fine-tuning, the step-by-step ingestion workflow (curate sources, structure content, crawl/upload/FAQ, chunk and embed, citations, test on real questions, handoff, re-sync), how grounding plus citations control hallucinations, and when fine-tuning is actually the right tool.
- [What Is a Good Ticket Deflection Rate? A Realistic 2026 Benchmark](https://owlish.bot/blog/ticket-deflection-rate/): How ticket deflection rate is calculated, why deflection is not the same as resolution, what a defensible 2026 benchmark looks like (and why vendor headlines run higher), how to measure it without counting timeouts and abandons as wins, and five ways to raise it honestly with grounded answers, citations, and clean handoff.
- [How to Offer 24/7 Customer Support With AI (Without Breaking Trust)](https://owlish.bot/blog/24-7-customer-support-ai/): Operations playbook for running always-on AI support — which questions an agent should answer overnight, what still needs a human, how to handle handoff at 3 a.m. with capture-and-queue or on-call, the source and refusal setup that makes after-hours answers safe, an incremental rollout, and the after-hours metrics that prove it works.
- [AI Agent vs Chatbot vs Copilot for Customer Support](https://owlish.bot/blog/ai-agent-vs-chatbot-customer-support/): A buyer-decision guide for routing each support contact to a cited customer-facing answer, a human-facing copilot, a narrowly governed action, or human escalation. Includes the Evidence, Authority, and Reversibility Support Scope Card, eight worked contact reasons, read-only-before-write integration guidance, and an honest account of where Owlish fits today.
- [RAG vs Fine-Tuning for Customer Support: Which One Your Bot Actually Needs](https://owlish.bot/blog/rag-vs-fine-tuning-customer-support/): Why RAG is the default for support bots and fine-tuning is the exception — how to tell which you need, what each costs to run, the hybrid pattern, and the mistake of fine-tuning to teach a bot facts.
- [Why AI Support Agents Get Rolled Back After Launch](https://owlish.bot/blog/why-ai-support-agents-fail/): Why most AI support agents fail weeks after launch, the five failure modes behind rollbacks, and the grounding, citation, refusal, handoff, and monitoring controls that keep an agent live.
- [6 Zendesk Alternatives for AI Customer Support Teams in 2026](https://owlish.bot/blog/zendesk-alternatives/): Buyer-focused comparison of Zendesk, Owlish, Freshdesk, Help Scout, Intercom, Gorgias, and Chatbase across AI support quality, grounded answers, human handoff, and pricing shape, checked against official public materials in May 2026.
- [AI Customer Support for Ecommerce: How to Set It Up Right](https://owlish.bot/blog/ai-customer-support-ecommerce/): Practical ecommerce guide to which support questions AI can ground and answer, where live order data and human handoff are required, and how to set it up without breaking trust.
- [How to Build a Microsoft Teams Support Bot](https://owlish.bot/blog/microsoft-teams-support-bot/): Practical Microsoft Teams support bot guide for choosing a first workflow, preparing trusted sources, controlling mentions and threads, and planning human handoff.
- [AI Customer Service for Small Business: Start Here](https://owlish.bot/blog/ai-customer-service-small-business/): Practical SMB rollout plan for choosing one AI customer service workflow, preparing trusted sources, testing handoff, and expanding safely.
- [AI Customer Service Metrics: What to Measure After Launch](https://owlish.bot/blog/ai-customer-service-metrics/): Post-launch support-ops scorecard for measuring verified resolution, citation quality, unsupported answers, handoff timing, repeat contacts, and source gaps.
- [Customer Support Chatbot Mistakes to Fix Before Launch](https://owlish.bot/blog/customer-support-chatbot-mistakes/): Pre-launch trust and guardrails checklist for avoiding weak sources, unsafe answers, poor handoff, bad metrics, and missing review loops.
- [AI Customer Service Agent POC: A Practical Test Plan](https://owlish.bot/blog/ai-customer-service-agent-poc/): Practical AI customer service agent POC plan for testing real support questions, citations, handoff quality, and post-launch improvement.
- [Support Ticket Automation: What to Automate First](https://owlish.bot/blog/support-ticket-automation/): Practical support-ops playbook for choosing AI-safe tickets, setting handoff rules, measuring quality, and avoiding bad deflection.
- [How to Build a Slack Support Bot](https://owlish.bot/blog/slack-support-bot/): Practical channel workflow guide for choosing a first Slack support use case, preparing trusted sources, configuring answer behavior, and planning human handoff.
- [How to Add an AI Chatbot to Your Website](https://owlish.bot/blog/ai-chatbot-for-website/): Practical setup guide for support teams choosing a first use case, preparing sources, testing citations, installing the widget, and planning human handoff.
- [AI Support Handoff: When Bots Should Escalate](https://owlish.bot/blog/ai-support-handoff/): Practical support-ops playbook for deciding when an AI chatbot should stop, what context to pass, and how humans take over.
- [Build an AI Knowledge Base for Customer Support](https://owlish.bot/blog/ai-knowledge-base-customer-support/): Practical guide to structuring support sources, refresh ownership, citations, and handoff so AI support answers stay trustworthy.
- [6 Intercom Alternatives for AI Customer Support Teams in 2026](https://owlish.bot/blog/intercom-alternatives/): Buyer-focused comparison of Intercom, Owlish, Help Scout, SiteGPT, Zendesk, Crisp, and Tidio, checked against official public materials in May 2026.
- [5 Chatbase Alternatives for Customer Support Teams in 2026](https://owlish.bot/blog/chatbase-alternatives/): Buyer-focused comparison of Chatbase, Owlish, SiteGPT, Intercom, Zendesk, and Tidio, checked against official public materials in May 2026.
- [6 SiteGPT Alternatives for Customer Support Teams in 2026](https://owlish.bot/blog/sitegpt-alternatives/): Buyer-focused comparison of SiteGPT, Owlish, Chatbase, Intercom, Help Scout, Zendesk, and Tidio, checked against official public materials in May 2026.
- [AI Customer Support Pricing: How to Compare Tools](https://owlish.bot/blog/ai-customer-support-pricing/): Buyer-focused guide to comparing AI support pricing models, from seats and resolutions to credits, add-ons, and handoff costs.
- [Owlish is live: AI customer support, grounded in your knowledge](https://owlish.bot/blog/launching-owlish/): Launch announcement covering what ships in v1 — source-cited answers, multi-channel deployment, operator-first handoff.
- [Why your AI support bot keeps making things up (and how to stop it)](https://owlish.bot/blog/why-grounded-answers-matter/): Playbook on AI hallucinations in customer support — how grounding, citations, and clean refusals replace guessing with verifiable replies.
- [How to Test an AI Support Agent Before You Let It Talk to Customers](https://owlish.bot/blog/test-ai-support-agent-before-launch/): Pre-launch QA playbook — build a scenario bank from real tickets, test grounding and the "I don't know" path, run adversarial checks, and keep it as a regression suite.
- [Who's Liable When Your AI Support Agent Gives a Wrong Answer?](https://owlish.bot/blog/ai-chatbot-liability/): Trust-and-accountability guide on AI chatbot liability, anchored in two rulings — Moffatt v. Air Canada (BCCRT, Feb 2024) and OLG Hamm's May 2026 Aesthetify ruling (Az. 4 UKl 3/25) — that both hold a business liable for what its chatbot says even when the AI hallucinates. Covers why "the bot did it" was never a real defense, the operational cost of an unaccountable agent (promises you have to negotiate out of, deflection numbers hiding silent wrong answers), a five-control answer-accountability stack (grounding, citations, scoped refusal, context-carrying handoff, disclosure plus logging), an eight-item vendor checklist, an FAQ, and where Owlish fits (grounded answers with citations, scoped refusal, human handoff with full context, pinned exact answers for high-stakes topics) and where it doesn't (full contact-center suites, judgment-heavy support that needs a human anyway). Sources checked June 2026; not legal advice.
- [AI Customer Support on Instagram and Messenger DMs (2026 Guide)](https://owlish.bot/blog/instagram-messenger-customer-support/): Channel playbook distinguishing real Instagram/Messenger customer support from DM "automation" marketing bots (keyword triggers, story-reply funnels). Covers Meta's two governing rules (the 24-hour standard messaging window and the 7-day Human Agent tag reserved for human-sent messages only), account prerequisites (Instagram Professional account + linked Facebook Page, messaging permissions), how to scope an agent to real DM intents instead of a whole website, grounding and citations for un-inspectable DM answers, handoff design inside Meta's windows, metrics that matter (resolution rate, handoff reason, repeat-contact rate) versus vanity "messages automated," a launch checklist, an FAQ, and where Owlish fits (Instagram and Messenger as one-click connected channels on Growth+, same agent and knowledge base as the web widget, grounded citations, human handoff, one shared inbox across channels) and where it doesn't (outbound DM marketing/lead-capture). Platform mechanics checked against Meta developer docs June 2026.
- [How to Keep Your AI Support Knowledge Base Up to Date](https://owlish.bot/blog/ai-knowledge-base-maintenance/): Knowledge-base maintenance playbook for grounded/RAG support agents — why stale content is worse with AI than with humans (a wrong answer still carries a confident citation), the Gartner finding that 61% of customer service leaders have a knowledge-article backlog and over a third have no formal revision process while 91% face pressure to ship AI anyway, the two moving parts of "fresh" (source content vs. the re-indexed copy the agent actually searches), tiering review cadence by volatility instead of a calendar, giving every source a named owner, three drift-detection signals (probe queries, handoff reasons, citation inspection), re-ingest/re-crawl/re-index discipline, an eight-step maintenance loop, an FAQ, and where Owlish fits (re-ingestable sources, inspectable citations, handoff signals as a gap map, Direct Responses for fast-moving answers) and where it doesn't (it surfaces drift but doesn't replace a human owning the source of truth). Gartner figures checked June 2026.
- [First Contact Resolution (FCR) With AI: How to Raise It Without Faking It](https://owlish.bot/blog/first-contact-resolution-ai/): Support-metrics guide to first contact resolution — SQM Group's finding that every 1% of FCR tracks to roughly 1% of CSAT and 1% of operating cost (about $286,000/year for an average mid-size center), Gross vs. Net FCR, 2026 benchmarks (industry average ~70%, 80%+ world-class, reached by only ~5% of centers), the three documented causes of low FCR (siloed knowledge, no context, no authority to act), how a grounded AI agent raises real FCR versus how careless deployments quietly wreck it by counting deflections as resolutions, why a context-carrying handoff protects Net FCR instead of hurting it, five rules for measuring FCR honestly, a six-step playbook, an FAQ, and where Owlish fits (grounded answers with citations, human handoff with full context, handoff reasons as a gap map, one agent across channels) and where it doesn't (agents lacking authority to act, issues that legitimately need a second touch). Stats attributed to SQM Group, Gartner, HDI, Zendesk, Sprinklr, Intercom, and Talkdesk, checked June 2026.
- [Salesforce Is Buying Intercom (Now Fin): What It Means for Your Support Stack](https://owlish.bot/blog/salesforce-intercom-acquisition/): News-to-evergreen analysis of Salesforce's June 15, 2026 definitive agreement to acquire Fin (the company formerly known as Intercom) for ~$3.6 billion — the confirmed facts (Intercom renamed to Fin in May 2026; deal expected to close Q4 Salesforce FY27; Fin brings 30,000+ customers; Salesforce cites Fin resolving ~76% of support volume end-to-end), why Salesforce wanted it (Fin as the packaged fast-to-deploy complement to enterprise-grade Agentforce), what actually changes for Intercom customers on a now/after-close/long-term timeline (little breaks immediately; roadmap and pricing direction shift toward Salesforce over quarters), the two durable trends the deal confirms (platform consolidation plus per-outcome AI pricing — Fin $0.99/outcome, Zendesk per automated resolution on a 72-hour window, Help Scout $0.75/resolution), a six-question checklist for choosing AI support tools that survive an acquisition or repricing (clean data export, answers grounded in content you own with visible citations, forecastable pricing, handoff you control, no-code setup, honest switching cost), an FAQ, and where Owlish fits (grounded cited answers, flat per-conversation pricing with no resolution meter, human handoff into a shared inbox, portable content) and where it doesn't (enterprise contact center, telephony, ultra-high volume). Facts checked against Salesforce Investor Relations, TechCrunch, and CNBC, July 2026.

## Optional

- [Privacy Policy](https://owlish.bot/legal/privacy/): How Chevvi Pty Ltd collects and processes personal information.
- [Terms of Service](https://owlish.bot/legal/terms/): Acceptable use and account terms.
- [Refund Policy](https://owlish.bot/legal/refunds/): How Owlish handles Free-first signup, paid billing, cancellation, case-by-case refund reviews, billing corrections, and statutory refund rights.
- [Data Processing Agreement](https://owlish.bot/legal/dpa/): GDPR Art. 28 processor terms.
- [Cookie Policy](https://owlish.bot/legal/cookies/): What cookies are set and why.