Skip to main content

​
Read the Insights dashboards

The Insights API reference documents the aggregate endpoints — this guide is the prose walkthrough for the richer dashboards that sit on top of them in the Orbit console, and explains what each one actually tells you when you open it. Each section names the console surface so you can cross-check what you’re reading against the live page. Example requests in this guide assume you are using the Insights API directly; most operators read the dashboards in the console and use the API for export or integration.

​
What an Insights surface is

Every Insights page reads from a read-replica of your tenant’s event and conversation data with buckets pre-aggregated at hourly or daily granularity. That keeps the analytics load off the write path that serves your live conversations, and it means the numbers update on the bucket cadence, not in real time — expect “today” (or the current hour) to lag slightly. Access is role-scoped per surface. The cost-family surfaces — LLM spend, Agent ROI, the cost governor — restrict to the owner, admin, developer, and billing roles, because they expose aggregate cost and revenue analysis. The analytics-family surfaces (retention, journeys, funnels, topic intelligence, deliverability, SMS click-throughs, containment) are visible to the owner, admin, developer, and viewer roles. Readers without the right role simply don’t see the page in the sidebar, and API calls from those roles are rejected.

​
Cohort retention — Insights → Retention

Console route: /insights/retention, page title “Cohort Retention”. What a cohort-retention grid answers: of the subjects (contacts, users, devices) who first did a thing in period X, what share keeps doing it in each later period? A cohort is a group of subjects bucketed by when they first fired an event — every subject belongs to exactly one cohort (their first-period bucket), and the grid then follows each cohort across subsequent periods. The dashboard renders both weekly and daily cohort curves over your collected CDP (Customer Data Platform) events, with a heatmap view where each row is one cohort and each column is one trailing period. Pick the event to measure (any event you emit via the events/CDP endpoints) and the dashboard groups first-fires, then charts what percentage of each cohort returns in period 1, period 2, period 3, and so on. How to read it as an operator: scan down a column to compare cohorts’ same-period retention (is the Tuesday campaign week holding better than Thursday?), and scan across a row to see how fast one cohort decays. The surface also exposes curated broadcasts and flows behind it — the broadcasts you run and the flows subscribers enter all emit events, so cohort grids tie directly to them.

​
Journey paths — Insights → Journey Paths

Console route: /insights/journey-paths, page title “Customer Journey Paths”. Journey paths visualize the touchpoint sequences customers actually take across channels — a cross-channel path/Sankey (a directed-weight graph where edge thickness encodes traffic share) over web, chat, voice, email, and messaging, with drop-off between stages. The dashboard attributes multi-step conversions to the ordered path and renders starting/ending nodes — pick the node where held journeys commonly start (e.g. “website visit”) or end (e.g. “order confirmed”) and the surface re-reweights the whole graph around that pivot. Use it to spot the dominant drop-off edges: if 80 % of your voice-answerers abandon at the “verification step” node, that is the failure to fix before any funnel level.

​
Funnels & Retention — Insights → Funnels

Console route: /insights/funnels, page title “Funnels & Retention”. The funnel builder writes ordered steps — each step is a CDP event (“contact created”, “first message sent”, “flow completed”) — and the dashboard measures step-to-step conversion: of subjects that reached step N, what share reached step N+1 within the conversion window? Add steps, set the conversion window, run the funnel, and the Step conversion chart shows Entered (subjects who reached a step) and Converted (subjects who moved on) per step. The page also embeds the same cohort grid as the Retention surface, so you can move between “did the campaign convert within 7 days?” (funnel) and “does the campaign cohort still convert after 30 days?” (retention) in one pass.

​
Topic intelligence — Insights → Topic Intelligence

Console route: /insights/topic-intelligence, page title “Topic Intelligence”. Topic intelligence clusters the contact reasons across voice, chat, email, and WhatsApp into auto-tagged topics, then trends them. For each topic you see volume, a per-channel split, and outcome signals: CSAT, deflection (how often the topic resolved without a human), and average handle time. Use it to find the topics routing badly to AI containment (high volume + low deflection) or the topics absorbing human-handle time (high handle time + high deflection — expensive if your flow could have resolved them).

​
Deliverability — Insights → Deliverability

Console route: /insights/deliverability. Deliverability aggregates per-channel delivery KPIs and diagnostics across SMS, WhatsApp, RCS, Email, Voice, Instagram, Messenger, and Viber — one tab per channel with a 24 h / 7 d / 30 d window selector. The score at the top of the page is the composite of route-level delivery, latency SLOs (percentage of deliveries that met your latency window), delivery-waterfall diagnosis (which delivery step leaked), and DLR anomaly detection on SMS routes — the “score” and “waterfall” give you the diagnosis, not a bare “good/bad” verdict. Use the export-to-workbook button on the page to pull the current channel’s series out as a spreadsheet. For the glossary-style scoring terminology your account admins should align on, see the deliverability score definitions in the glossary (the dashboard reuses the same terms).

​
SMS click-throughs — Insights → SMS CTR

Console route: /insights/sms-ctr, page title “SMS click-throughs”. SMS click-through reporting is the link tracking counterpart of delivery reporting: per campaign and per sending queue, it shows sends, tracked links, clicks, and the computed click-through ratio (CTR). The dashboard reads Orbit’s shorten-links pipeline, so campaigns that send shortened URLs (as Orbit campaigns always do) produce CTR by default — use this page when you need to know whether the message actually pulled recipients to your link, not just whether it delivered.

​
AI containment & resolution — Insights → Containment

Console route: /insights/containment, page title “AI Containment”. Containment is the share of AI-agent conversations that resolved without a human handoff; the surface additionally reports the share that reached a resolution (regardless of who resolved it), a daily trend, and a per-agent breakdown. Use it to separate “the AI contained the customer” from “the AI resolved the customer”: an agent with weak containment but strong resolution is escalating unnecessarily; an agent with strong containment but weak resolution is sitting in a loop. The companion Agent ROI dashboard prices those outcomes against the LLM cost of the agent.

​
LLM spend — Insights → LLM spend

Console route: /insights/llm-spend, page title “LLM spend”. LLM spend reports tokens and cents per agent, per model, and per conversation, in both overview (org-wide aggregate) and timeseries (daily buckets) form. It also tracks daily-cap progress and projects month-to-date spend. Use per-model and per-agent breakdowns when rebalancing routing — a cheaper model on a route with the same resolution rate reads directly as lower spend with no product change. For the full endpoint shapes — overview, summary, timeseries, top conversations, by-feature — see the LLM spend section of the Insights API reference.

​
AI cost governor — Insights → Cost Governor

Console route: /insights/cost-governor, page title “AI Cost Governor”. The cost governor is the enforcing counterpart of the dashboards above: per-tenant budget caps for LLM tokens (and ASR, the speech-to-text leg of voice-agent calls), automatic model downshift when spend nears its cap, and cost-per-resolution margin analytics per agent. This is the surface that makes the AI agent’s cost behaviour bounded rather than just observable — set a cap here and traffic continues at a cheaper model tier instead of runaway billing. Together with the AI agent cost controls page it is where agent ops teams live.

​
Marketing mix modeling — Insights → Media Mix Model

Console route: /insights/media-mix-model, page title “Marketing Mix Model”, backed by POST /api/v1/cdp/analytics/marketing-mix-model if you want the same fit in your own tooling. Where the attribution surface shares credit across the touchpoints each contact actually hit, and holdout/lift asks whether one campaign caused a lift, the marketing mix model is top-down: it regresses daily channel spend against daily revenue over time and answers the budgeting question — given what I spend per channel and what that spend returns on average, how should I split a fixed budget? To run a fit, point the model at your data: the revenue event whose value carries conversion or order value (any event you emit via the events/CDP endpoints, with a monetary property on it — value by default), the channel and spend properties on your spend events, a lookback window (14 to 180 days), and an adstock decay — how much of yesterday’s spend still works today. Set the decay higher for brand channels whose effect lands over weeks, lower for intent channels that convert the same day. Hit Run model and the surface regresses spend (with carryover and saturation applied) against revenue, then reports per channel: incremental revenue contribution and its share, ROI, current average daily spend, and a recommended daily spend, with the delta next to it. Read the response curve (the sparkline per channel) as the diminishing-returns shape: it is the expected revenue as spend on that channel rises, flattening as the channel saturates. The recommended budget split shifts money out of channels already on the flat part of their curve and into channels still on the steep part — reallocate roughly the budget the curve says, not the budget history says. The summary tiles above the table report the fit quality (R²), total and channel-attributed revenue, and the baseline revenue per day that no channel explains. If the page reports “not enough history yet,” the window has fewer than two weeks of aligned daily spend and revenue — widen the lookback or confirm the revenue event and spend/channel properties are actually being recorded. Access matches the cost-family dashboards because marketing mix modeling (MMM) reads the tenant-wide event stream: the owner, admin, and developer roles see the page, and api access needs the contacts:read scope.

​
Other Insights surfaces on the console

Beyond the walkthroughs above, the Insights section of the console also ships Agent ROI, agent comparison, high-friction sessions, KB auto-draft, language quality, sentiment, account scores, attribution, contact reasons, logs, links, custom dashboards, and alert/report surfaces. For the endpoint shapes behind all of them see the Insights API reference. This guide now carries the walkthrough for every formerly un-narrated Insights surface — most recently the marketing mix model above.

​
See also