Established or absent
A number appears on the site only when it is defined here. We do not publish averages we cannot show, and we do not round a single result up into a claim about all customers.
[ 01 | THE RULES ]
Three rules every figure on this site follows
Worked example, labelled
One dataset, run through the agents, is used across the site so every page shows the same figures. It illustrates the mechanics. It is not an average across customers and is never presented as one.
Corrected in the open
When a figure changes, this page changes first and says so. Vendor comparisons carry the date they were last checked and are corrected within a week of being told.
[ 02 | THE WORKED EXAMPLE ]
One dataset, used everywhere on the site
These figures appear on the home page, the agent pages, pricing and the solution pages. They all come from the same run.
100 pieces of feedback become 8 themes and 35 sub-themes
Measures: how Chorus clusters a batch of feedback: one input set of 100 reviews, tickets and call excerpts, grouped by intent into 8 themes, each split into sub-themes (35 in total), each carrying a mention count, the quotes behind it and the ARR of the accounts that raised it.
Does not claim: that every 100 pieces of feedback produce 8 themes. Theme counts depend on the input. The ratio illustrates the compression from records to meaning.
The RingCentral free-tier example
Measures: one cause-and-effect pairing in the worked example. Recon detected a competitor free-tier launch on 22 May. Chorus recorded SMB pricing complaints in calls and tickets rising by 31 mentions, with the rise beginning 8 days after the move. Atlas priced the accounts behind those mentions at $214K of ARR. The link between move and reaction carried a 68% confidence score with the records behind it.
Definitions: mentions are individual records matching the theme; 8 days is the lag from the move to the first sustained rise; ARR touched is the annual recurring revenue of the accounts whose records make up the reaction, priced from the CRM; 68% is the agent’s own confidence in the causal link.
Does not claim: that RingCentral caused every complaint, or that lags are typically 8 days. It shows the view: move above the line, reaction below, lag measured rather than assumed.
15 COMPETITORSRecon in the worked example tracks 15 competitors across sites, changelogs, pricing pages and reviews, daily. Fifteen is the size of that watchlist, not a limit; Enterprise plans include up to five competitors with win/loss and battlecards as standard, and more on request.
[ 03 | COUNTS ]
Figures that are simple counts, and how we count them
76 connectors
The number of sources listed on the integrations page on the date shown there. One connector is one system that connects once over OAuth and is read continuously. Custom sources ship in two to three weeks and are added to the count when live.
7 comparisons
The number of tools compared on the alternatives pages. Every statement about another vendor comes from that vendor’s own public documentation, carries the date it was last checked, and is corrected within a week of being told it is stale.
Starter units: signals and insights
Starter is free and counts two things a month: signals read (50 from one feedback source, 10 from one competitor source) and insights produced (10 and 3). A signal is one record read; an insight is one ranked finding with its evidence. The definitions are the ones in the glossary.
[ 04 | TIME TO ACTION ]
The 8 days, 36 hours and −81% on the home page
DEFINITIONMeasures: the time from a signal landing in a connected source to an action with a named owner. By hand is the worked example done manually: exports reconciled, tickets read one by one, the competitor move found late, 8 days. With HyperOrbit is the same example with all three agents running: read overnight, themed and priced, the play drafted with an owner, 36 hours. −81% is the difference between the two.
Does not claim: a customer average. These are modelled figures from the worked example. They will be replaced with production medians, and this page will say so, once enough customer workspaces have run for a full quarter.
Does not claim: a customer average. These are modelled figures from the worked example. They will be replaced with production medians, and this page will say so, once enough customer workspaces have run for a full quarter.
[ 05 | CONFIDENCE ]
What a confidence score is, and is not
What it is
The agent’s own estimate, as a percentage, that a finding is right: that a theme is one theme, that a reaction follows a move, that an account is at risk. The evidence behind the score is one click away so a person can judge it.
How it is used
Autopilot acts only above the confidence floor you set. Below it, the agent suggests (Advisory) or waits for a click (Supervised). Anything touching pricing or an account inside its renewal window stops and asks regardless of score.
What it is not
Not an accuracy rate measured across customers, and not a probability that has been calibrated against outcomes. We publish it because an agent that cannot say how sure it is cannot be trusted with anything that matters.
[ 06 | GO DEEPER ]
Related on hyperorbit.ai
CAUSE & EFFECTSee the RingCentral viewMove above the line, reaction below, the lag measured.
GLOSSARYWhat is agentic customer intelligence?The definition and the 25 terms used across this site.
COMPARISONSHyperOrbit alternativesSeven tools compared on the same dimensions, dated and correctable.
TRUSTTrust and complianceSOC 2 Type II, ISO 27001:2022, GDPR, DPDPA and the sub-processor list.
[ 07 | FAQ ]
Citing and correcting
Can I quote these figures?
Yes, with this page as the source and the qualifier that they come from a worked example unless this page says otherwise. Link to the anchor for the figure you quote, for example /resources/methodology#ringcentral.
Why publish a worked example instead of customer averages?
Because we can show every record behind it. Averages across customers arrive once enough workspaces have run for a full quarter; until then a labelled example is more honest than a rounded claim.
How do I report a wrong or stale figure?
Write to agent@hyperorbit.ai with the page and the figure. Corrections land within a week and this page records what changed.
Are the vendor comparisons independent?
No. They are written by HyperOrbit from each vendor’s public documentation and say so on every page. Each states what the other tool does well and who should choose it, and carries the date it was last checked.
Every signal, on one orbit
Connect your first source in an afternoon. The first pass lands before your next standup.
Book a demo