USSD revenue rarely disappears overnight.
The warning signs usually appear first: sessions begin falling, completion rates change, timeouts increase, or an application that once generated steady traffic slowly becomes less active.
The problem is that operators often notice these changes after revenue has already been affected.
AI can change that by turning USSD operational data into an early-warning system.
For a USSD service provider, the opportunity is not simply better reporting. It is identifying which applications are becoming commercially risky early enough for someone to act.
The data is already there
A USSD platform already produces much of the information needed to detect risk.
Application traffic, sessions, transaction volumes, charging activity, timeouts and usage patterns can all reveal whether a service is behaving normally.
hSenid’s USSD Gateway, for example, provisions applications using parameters including transactions per second and maximum sessions, while its session-management layer handles the initiation, management and termination of USSD sessions. The platform also supports per-request and session-based charging.
Instead of treating these as separate operational metrics, an AI model can look at how they move together.
Consider an application that normally processes 500,000 sessions every week.
Traffic drops to 450,000.
Then 410,000.
At the same time, session resumptions and failures increase.
A conventional dashboard shows the decline. An AI-driven system should go further and flag the application because its current behaviour has moved outside its normal pattern.
The important question becomes:
Which USSD applications are likely to lose revenue next?
What AI should actually detect
A useful revenue-risk model does not need to be complicated for the operator using it.
It should continuously look for signals such as:
- sustained declines in sessions or transactions
- unusual increases in failed or abandoned sessions
- applications performing differently from their historical baseline
- sharp changes after a configuration or infrastructure update
- abnormal traffic compared with similar applications
- changes in peak usage behaviour
- service degradation affecting applications that generate significant traffic
The objective is not another dashboard full of graphs.
It is a clear operational warning that tells the team which application is showing abnormal behaviour and where investigation should begin.
That gives operations and commercial teams something they can act on.
This matters more when USSD operates at scale
The scale of production USSD makes manual monitoring difficult.
hSenid’s current USSD platform documentation states that deployments can handle 45 million+ transactions per day and up to 15,000 transactions per second.
At that volume, small percentage changes can represent large numbers of sessions.
A two-percent deterioration may not look dramatic on a graph. Across millions of transactions, it can become commercially significant.
That is where AI becomes useful: not replacing the operations team, but narrowing millions of events down to the applications that deserve attention.
A real example from our own deployments
There is also a human reason to watch USSD performance carefully.
In the Democratic Republic of Congo, a leading bank used an hSenid USSD-based application to extend digital banking nationwide. The deployment reached subscribers across mobile operators and processed nearly 20,000 transactions per day through the USSD application.
Customers could check balances, transfer value, pay bills and access wallet or agent services without needing mobile data or a smartphone.
At that point, USSD availability is not simply an IT metric.
A traffic decline or repeated session problem can affect thousands of customer interactions every day.
That is why detecting deterioration before it becomes a major outage matters.
From anomaly detection to revenue-risk scoring
The next step is to combine operational and commercial signals.
Instead of giving every anomaly the same priority, an operator could calculate a risk score using factors such as traffic trend, transaction value, abnormal behaviour and how long the issue continues.
An application with a minor anomaly and low usage may receive a low-risk score.
A heavily used banking or payment service showing falling sessions for several days could be escalated much faster.
This direction is already part of hSenid’s USSD and AIOps work. The product roadmap includes ML-based revenue-risk identification, application traffic and peak-usage tracking, proactive anomaly detection and automated root-cause analysis across infrastructure and applications.
These are areas being developed as part of the platform’s AI and AIOps direction, rather than results we should claim before they are measured in production.
AI should tell you where to look
A strong USSD solution should do more than produce monitoring data.
It should help operators understand what requires action.
AI can move USSD session management from:
“What happened?”
to:
“Which application is becoming risky, what changed, and where should we investigate first?”
That is a more useful operating model for teams managing multiple USSD services across increasingly complex telecom environments.
USSD remains important because it works without an app, mobile data or a smartphone, including in low-bandwidth environments.
The next step is making the infrastructure behind that simplicity more intelligent.
