Radio streaming analytics count what a station's own server can see: connections, sessions and hours delivered over the internet. They are exact for the stream and silent about anyone on FM, so they can't give a station's reach. Used for what they measure, they are very useful. Added to a survey figure, they make it worthless.
Key points:
- A streaming server reports connections. Listening hours and sessions are built on those.
- The standard metrics are session starts, total listening hours, average active sessions and cume.
- A connection is a device, which might be one person, a room or nobody.
- Cume can't be added across days or stations.
- Stream figures and survey reach overlap by an unknown amount, so they stay in separate lines.
What can a radio streaming server count?
Connections, and how long they last. Icecast's server statistics are a plain example. "Listeners" is the number currently connected to a stream, "listener_peak" is the highest number connected at once, and "listener_connections" is a running total of connections.
Those are three different numbers. A station that quotes its running total as listeners has confused visits with people.
The server knows the time, the address of the connection and the kind of player. It has no idea who is holding the phone.
What do the main streaming metrics mean?
Measurement firms have turned raw connections into a standard set. These are Triton Digital's definitions, which count a session only if it lasts at least one minute.
| Metric | What it means | Watch for |
|---|---|---|
| Session starts | Sessions begun in the period | A dropped connection that reconnects starts again |
| Total listening hours (TLH) | All hours listened in the period | Sensitive to a few long sessions left running |
| Average active sessions (AAS) | TLH divided by hours in the period | The nearest thing to an average audience |
| Cume | Unique listening agents or identifiers | Can't be summed |
| Time spent per listener (TSL) | TLH divided by cume | Differs from average time per session |
Each of these describes the stream alone. Audience metrics explained covers the broadcast versions of the same ideas.
How do you work out average active sessions?
Divide listening hours by hours in the period. Here is an invented day.
A station's stream delivers 1,800 listening hours across 24 hours. That gives 1,800 ÷ 24 = 75 average active sessions, meaning about 75 sessions were live at a typical moment.
If 600 unique identifiers connected that day, time spent per listener is 1,800 ÷ 600 = 3 hours. Those are clean, useful figures for a stream, and they describe 600 devices.
Why is a connection not a person?
Because identity on the internet is an inference. For podcasts, Triton separates listeners by a combination of IP address and user agent, which is a technical label for a device and app.
That label slips both ways. A phone moving from WiFi to mobile data can look like two listeners, and an office on one network can look like one.
Real use is messier still. A phone on an aux cable in a matatu is one connection and a vehicle full of people. A browser tab left open at a desk is one connection and no listener. Filters such as the one-minute rule remove some noise, and they can't see into the room.
What does a stream log miss?
Everything off that server. FM listeners never touch it, and for most Kenyan stations that is the great bulk of the audience.
Listeners on third-party apps, YouTube or Facebook live sit on someone else's system, so a station's own log won't show them. Measurement firms note the same limit. Triton says data from custom players, third-party delivery networks and publisher-run servers is outside its control.
Why can't you add cume figures together?
Loyal users repeat. The same device turns up in each one. Triton states that cume "cannot be accumulated or summed to create an aggregate".
A station with 3,000 unique connections on each of seven days hasn't reached 21,000. The loyal morning user is in all seven days. The same applies across stations, and across a station's app and website.
Survey reach behaves the same way, which is why daily and weekly reach are measured separately and never multiplied.
What counts as listening on a stream?
A minimum time, set by the vendor. Triton counts a session at one minute or more.
Broadcast rules differ: five minutes in a quarter hour at RAJAR, and three minutes at Nielsen's meters. A one-minute session and a diary quarter hour aren't the same unit, so the two can't be set side by side.
How much Kenyan radio listening is online?
Less can be said than stations hope. The regulator's August 2026 report shows that in April to June 2026, 79% used radio through a radio set and 27% through a mobile phone.
That 27% isn't streaming. Many phones have FM tuners, and the report doesn't separate the two. People can also use both, so the figures add to more than 100%.
The cleanest way to learn the online share is to ask. A survey question on how people mainly listen gives a share of the audience, which a server can never give.
Can streaming figures be added to survey reach?
Never by simple addition. Stream users are already inside the survey's population, so adding them counts many people twice.
Australia shows what proper integration takes. Its Radio360 system joined live streaming data to broadcast audience data from 2023, and a panel of 2,000 people wearing a watch meter is used to validate the data and remove duplicates. That is a whole measurement system built to do what a sales deck does with a plus sign.
Why does a stream rise when reach doesn't?
Stream numbers respond to things that have little to do with audience size. A link shared on social media, a sports event, a diaspora community waking up, a broken app that keeps reconnecting.
That is why publishers and vendors filter suspect events. Automated traffic counts too. A rising stream is good news about the stream, and it raises a question about the wider audience.
What are streaming analytics good for?
Plenty, when kept in their lane.
- Spotting an outage in minutes, from a sudden fall in active sessions.
- Seeing when the stream fills and empties through the day.
- Comparing app, website and smart speaker delivery.
- Reporting delivered streamed adverts to a buyer of online audio.
- Seeing rough location by network address, including listeners abroad.
For on-demand audio, the logic carries over to podcast audiences in Kenya.
How should a station report streaming to an advertiser?
On its own line, with its own label. Here is an invented example: "In our 30-day October, the stream delivered 54,000 listening hours in sessions of one minute or more, an average of 75 active sessions."
That is a precise statement, and an advertiser can use it. Below it, put measured reach from a survey, with its base and margin. Don't add them, and don't let a sales deck do it for you. See how a station proves its reach.
How do you set up streaming measurement?
Start with what you control. Put each stream through one server or provider, keep the logs, and write down the rules: minimum session length, what is filtered and how a month is counted.
Then keep those rules fixed. A change of player or provider can move the numbers on its own, so log the date of every change.
Report the same three figures each month: total listening hours, average active sessions and cume. Three steady numbers tell you more than a dashboard of forty.
What do stations get wrong?
They quote peak connections as listeners. They add stream cume to survey reach. They compare their stream with a rival's without knowing what filters each one uses.
The better habit is to use the server for what it sees exactly, and to use a survey for the actual radio listenership it can't see.
Read next
- Audience measurement software for Kenyan media: the TI pack page for this work
- How media reach is measured in Kenya: all the pages in this section
- How is radio listenership measured?
- Can a radio station know its actual listenership?
- Podcast audiences in Kenya
- Takwimu Intelligence (TI) home page
- Audience measurement in Kenya: the guide to this library
- The TI platform and how we check our numbers
Where TI fits
A server counts connections, and only people can say how they listen. Takwimu Intelligence (TI) reports the share of an audience who listened mainly over the internet or in a vehicle as survey indicators, kept apart from any stream count, with reach weighted to KNBS census totals at county level.
To see it on your own station or title, book a walk-through.