A News Feed Worth Reading
Embedded AI inside the company feed — not a separate agent surface. Draft a Must-Read in the right tone with facts preserved, extract risks / issues / ideas from discussions before they splinter, transcribe voice posts and auto-subtitle videos for multilingual workforces, and let employees listen to long posts on the go. Every capability is per-business toggleable.
Where Company Feeds Fall Apart
News Feed Agent attacks the four specific failures that turn an internal feed into a low-signal timeline — without sending the company back to email and chat.
Important Posts Read Like Slack Messages
A Must-Read about benefits changes goes out in a casual tone, lowercase, with the dates buried at the bottom. The acknowledgment number is fine; the comprehension number is the problem. Draft-then-rewrite-by-hand doesn't scale across a busy comms team.
Real Signals Drown In The Discussion Volume
A 23-comment thread on attrition has a buried risk worth escalating, an idea worth routing to People Ops, and an action item that's been sitting unowned for a week. Nobody re-reads every comment; the signal disappears with the discussion.
Voice And Video Posts Are Inaccessible To Half The Workforce
A video update from the CEO lands without captions; voice posts arrive without transcripts. Frontline staff in noisy environments, deaf and hard-of-hearing employees, and non-English speakers all hit a wall. Engagement drops; the post might as well not exist for them.
Long Posts Don't Get Read — But Get Acknowledged Anyway
A 2,000-word policy walkthrough hits 95% acknowledgment because employees click the box. Without read-aloud, summarization, or a way to revisit later, the policy goes through the compliance loop without going through the comprehension loop.
Comms Leads Don't Know What's Resonating Until The Weekly Report
Yesterday's Must-Read on the new safety protocol got 12% engagement. Today's post on vending machine choices is at 81%. The comms lead finds out at the Friday report — by which point the safety post is buried and the next Must-Read is already drafted in the same tone. A live signal would have prompted a re-cast in the right moment.
Comment Threads Splinter Across Workspaces And Lose The Thread
A single announcement generates discussions in five different team channels. Each thread surfaces a real concern; together they paint a clear picture. Without a way to pull the cross-thread signal back to the original post, the comms lead and the post author both miss the consensus that's already formed in the comments.
News Feed Agent At A Glance
News Feed AI
Embedded AI in the feed — authoring, signals, voice, video, summary.
Inside News Feed Agent — The Actual Capabilities
Every block below maps to a real service inside the News Feed app, gated by a per-business feature toggle in the marketplace app configuration. The AI is embedded — not a separate Ask AI agent — and follows the per-feature thresholds the admin sets.
AI-Enhanced Authoring — Four Tones, Facts Preserved
The composer's "Enhance with AI" button rewrites a draft for grammar, clarity, and tone via NewsFeed::AiEnhancementService. Four allowed tones — professional, friendly, urgent, concise. The system prompt enforces hard constraints: every name, date, figure, link, @mention, and #topic must survive the rewrite. Server-decided char limits (5,000 for posts, 1,000 for comments) prevent client-side gaming.
- NewsFeed::AiEnhancementService — wraps LlmConnectionManager (Claude standard tier), never direct HTTP calls.
- 4 allowed tones — professional, friendly, urgent, concise. Invalid input falls back to professional.
- Hard fact-preservation — system prompt explicitly forbids adding new info or removing existing names/dates/figures/links/@mentions/#topics.
- Per-business toggle — ai_enhanced_authoring in the News Feed app configuration; off by default until admin opts in.
Signal Extraction — Surface Risks, Issues, Ideas From Discussions
Long threads leak important signals. NewsFeed::SignalEmitter reads posts, comments, and reactions and proposes 7 signal types — risk, issue, question, idea, decision, action item, sentiment — each with a per-business confidence threshold. Above-threshold signals route into the Mango Signal app; below stay as drafts. The thresholds are an admin knob, not a model constant.
- NewsFeed::SignalEmitter — extracts signals from posts and comment threads; checks per-business toggles before emitting.
- 7 signal types — risk (0.90 default), issue (0.80), question (0.70), idea (0.70), decision (0.75), action item (0.75), sentiment (0.70).
- Per-business thresholds — every signal type has a tunable confidence threshold in the app configuration; admins move the bar to match their org's noise tolerance.
- Signal extraction master toggle — signal_extraction defaults to off; the feature must be explicitly enabled per-business.
Voice Posts, Auto-Subtitles, And Read-Aloud — Accessibility By Default
Voice posts get transcribed via Whisper (NewsFeed::VoiceTranscriptionService); video posts get WebVTT subtitles via Whisper (NewsFeed::SubtitleGenerationService); long posts can be read aloud via NewsFeed::TtsService with three voice options. Discussion summarization (ai_discussion_summarization) collapses 23-comment threads into key takeaways once a configurable comment count is hit.
- NewsFeed::VoiceTranscriptionService — voice posts uploaded to S3, transcribed via Whisper; configurable max duration (1/2/3/5 minutes).
- NewsFeed::SubtitleGenerationService — video posts get WebVTT subtitles, stored in the subtitle bucket per business.
- NewsFeed::TtsService — read-aloud button on posts; neutral / male / female voice selection in app config.
- Discussion summarization — auto-summary above a configurable comment threshold (default 10), with optional auto-trigger on discussion close.
Outcomes Teams Can Measure
The agent's job is to make the company feed a place where Must-Reads actually land, signals get caught while they still matter, and voice/video content reaches the multilingual and accessibility-needs employees who'd otherwise be excluded. Measure against your pre-AI baseline.
- AI-enhance usage on Must-Reads — share of Must-Read drafts that ran through tone-tuned enhancement before publish.
- Signal-to-action conversion — share of above-threshold signals that route to an owner and get resolved (vs sit in the Mango Signal backlog).
- Subtitle / transcript coverage — share of voice and video posts published with auto-subtitles or transcripts attached.
- Read-aloud adoption — share of long posts (>500 words) listened to via TTS, especially in the frontline workforce.
- Discussion summary adoption — share of long threads (>10 comments) where the summary is read by latecomers vs scrolling the full thread.
Embedded AI, Per-Business Toggleable, Configurable Thresholds
News Feed AI isn't a separate Ask AI agent — every capability is embedded inside the News Feed app and gated by a per-business toggle in the marketplace app configuration. Admins opt in feature-by-feature, set thresholds that match their org, and audit usage through the same AiApiLog every other LLM call lands in.
- 9 independent feature toggles — enhanced authoring, auto-topic, auto-answer, SME routing, discussion summarization, poll summarization, signal extraction, voice transcription, auto-subtitles, read-aloud. Each opts in separately.
- Per-business thresholds — signal confidence thresholds, summarization comment threshold, voice max duration, TTS voice selection — all in the app's configuration_schema.
- RubyLLM-grounded — every AI call goes through LlmConnectionManager → RubyLLM (no direct OpenAI / Anthropic HTTP) and logs to AiApiLog.
- Permission-aware — signals, summaries, and SME routing all respect the existing News Feed permission model (Audience-Post-To-Everyone, SME tab visibility, must-read permission).
WHAT TEAMS TRY INSTEAD
The four alternatives — and why none of them see your feed, your audience, or your data
Most comms and HR teams reach for one of these four before considering an embedded agent. None of them stick because none of them have the thing that makes an in-feed AI useful — cross-app context and the platform's own permission model.
Pasting posts into ChatGPT, Claude, or Copilot
General-purpose AI doing one-shot tasks on copied text
- News Feed Agent sees the live audience, acknowledgement state, and engagement signal — not a static paste
- Translation, summarization, and SME routing run inside the post composer instead of in a side tab
- Output respects your audience targeting and must-read permission rather than being generic prose
Workplace AI / Viva Engage AI
Vendor-trapped feed AI inside Meta or Microsoft's silo
- Sees your shifts, HRIS, training completion, and inspection results — vendor AI can't reach across silos
- Works for the frontline employees who never had Workplace or Microsoft 365 licenses in the first place
- Survives the vendor's strategy shifts — Workplace is being wound down and Viva's feature roadmap is moving
Custom RAG / DIY chatbot on the intranet
An engineering team's six-month build, then forever maintenance
- Shipped already. Engineering spends zero weeks plumbing identity, audit, retention, or multi-language
- Tracks live engagement signal and SME identification — most DIY builds plateau at "answer questions"
- Stays current as the platform evolves — the agent inherits new tools and capabilities without redeployment
The manual fallback — "ask the comms team"
The default when AI tools fall short
- Reads 2 million+ posts a year so the comms team doesn't have to triage every must-read read-rate
- Summarizes long announcements into a 30-second voice clip for shift workers walking to their next station
- Routes SME questions to the right person automatically instead of "let me find out and get back to you"
PLATFORM LEVERAGE
News Feed Agent inherits everything the platform already runs
A custom-built workflow agent has to plumb each of these. News Feed Agent gets them for free because the platform already does.
Cross-app data plane
Reads from HRIS, Schedule, Training, Recognition, and Surveys to rank a post and find the right SME — no separate ETL or warehouse.
Unified permission model
Honors audience targeting, must-read permission, and SME visibility from the existing News Feed model — no parallel ACL system to keep in sync.
Audit trail & retention
Every AI call lands in AiApiLog with the same retention and eDiscovery posture as the rest of the platform.
Translation in 100+ languages
The same Mango Translation service that powers Chat, SOPs, and Policies powers the agent's summaries and SME prompts.
Mobile delivery & TTS
30-second voice summaries reach frontline staff in the same mobile app they already opened for shifts and pay.
RubyLLM-grounded model tiering
Nano / small / medium / standard / very-large tier selection routes routine work to cheap models and reserves the big ones for hard problems — automatically, per call.
INDUSTRY FIT
Industries where in-feed AI moves the most weight
News Feed Agent shines where comms reach is hardest and stakes are highest.
Retail
Summarizes promo and policy posts into 30-second voice clips so store associates hear them between shifts without reading a 600-word announcement.
Healthcare
Routes clinical and patient-safety SME questions to the right hospitalist or charge nurse and confirms the right population acknowledged.
Manufacturing
Translates safety alerts inline and produces multilingual TTS audio so plant-floor crews on shared kiosks hear the same message in their own language.
Financial Services
Tags every AI-summarized post with the original source and audit trail — broker and compliance teams can prove who saw what when.
Hospitality
Surfaces low-read posts to property managers before service starts so the morning huddle covers the things housekeeping and front desk haven't acknowledged yet.
Public Sector
Runs entirely inside FedRAMP-eligible deployment options with full audit logging — no AI traffic leaving the tenant boundary.
WHY MANGOAPPS WINS
An embedded agent beats a chatbot, a vendor add-on, or a custom build on every axis
The argument finance, security, IT, and comms all share — and the one a horizontal AI or a single-vendor assistant structurally cannot answer.
Cheaper than the alternatives
No per-seat ChatGPT license, no Viva Engage AI add-on, no six-month custom RAG build, no extra comms-team headcount to triage acknowledgements.
More secure
Every prompt and response logs to AiApiLog with the same retention as the rest of the platform. Nothing leaves the tenant boundary — and no separate AI vendor has your feed data.
Easier to deploy
Already deployed if you have News Feed enabled. Turn the agent on, point it at audiences from existing HR data, and it's running the same day.
Easier to use
Lives inside the post composer and the feed itself — no separate chatbot UI, no context-switching, no copy-paste.
Easier to manage
Per-business thresholds (confidence, summarization, TTS voice) sit in the same admin console as every other app's settings. One audit log, one access model.
Easier to extend
Shares the agentic tool framework with every other MangoApps agent. New capabilities (a new SME signal, a new summarization style) ship as tools, not rewrites.
AI is actually better
A vendor-trapped or generic AI can summarize a post. Only News Feed Agent can also see who's on shift, who completed which training, and which SME has the matching certification — and act on all of it.
At our head office within HR, marketing, and finance, we use ChatGPT quite a bit, and MangoApps AI Helper is a useful tool for us. As I go around to the stores, I’ve been showing them how the AI Intranet Helper can find things and answer plain language questions about our resources, and they’re quite excited about that.
Alison van Zijl
Head of Human Resources
Full House
Customer Success
Related Customer Stories
Customer Case Studies
Video Case Studies
Customer Case Studies
Customer Case Studies
Video Case Studies
Customer Case Studies
Frequently Asked Questions About News Feed Agent
News Feed AI is embedded inside the News Feed app, not a separate Ask AI agent. 9 per-business toggles control AI behaviors — tone-tuned authoring, auto-topic, auto-answer, SME routing, discussion summarization, poll summarization, signal extraction (7 signal types), voice transcription, auto-subtitles, and TTS read-aloud.
No. NewsFeed::AiEnhancementService uses a system prompt that explicitly forbids adding new information or removing existing names, dates, figures, links, @mentions, or #topics. Char limits are server-decided (5,000 for posts, 1,000 for comments) so clients can't game the rewrite to a longer-than-allowed output.
Each of the 7 signal types (risk, issue, question, idea, decision, action item, sentiment) has a per-business confidence threshold in the app configuration. Defaults are risk 0.90, issue 0.80, decision/action 0.75, idea/question/sentiment 0.70. Above-threshold signals emit into Mango Signal; below-threshold proposals stay as drafts for admin review.
No. NewsFeed::VoiceTranscriptionService sends uploaded voice posts to Whisper; NewsFeed::SubtitleGenerationService runs the audio track of video posts through Whisper and writes WebVTT. Both go through LlmConnectionManager — no direct provider HTTP calls. Transcript / subtitle generation is per-business toggleable and configurable.
AI-enhance usage on Must-Reads, signal-to-action conversion (signals routed to owners and resolved), subtitle/transcript coverage on voice + video posts, read-aloud adoption on long posts in frontline cohorts, and discussion-summary adoption on long threads. Compare against your pre-AI baseline.
Let's Talk
Since 2008, we've been building the workforce platform — earning the trust of 2 million+ users and an NPS of 78.
Why Choose Us?
- AI-Powered Platform: The most unified workforce experience on the planet.
- Top Security: HITRUST, ISO & SOC 2 certified.
- Exceptional UX: Delightful on mobile and desktop.
- Proven Results: 98% customer retention rate.
Trusted by Legendary Companies:
