Throughput economics
Cost per message or per minute decides the business. An architecture that is fine at current volume can be fatal at ten times it.
Telecommunications
Volume changes what correctness means.
At telecom scale a failure rate of one in ten thousand is not a rounding error, it is thousands of customers a day. Systems that mediate messages, calls and data have no quiet period to recover in, and the difference between a well-built platform and a merely working one only becomes visible at a volume most software never sees. That difference is what telecom software development is really about.
The question to answer first
“What happens to throughput and cost when we are ten times bigger?”
24/7
Support coverage, for platforms that never get a quiet hour
The pressure
Four forces that only appear at volume.
Regulatory obligation
Number portability, lawful intercept, emergency services and data retention are not optional and not negotiable.
Legacy protocols
SS7, SMPP, Diameter and SIP are not going away, and they sit next to whatever modern stack you are adding.
Always on
There is no maintenance window that customers accept. Deployment has to be safe while the system is carrying live traffic.
The work
What we build for operators
Four systems, all of which have to be correct at a volume where one in ten thousand is thousands of customers.
Messaging and routing platforms
High-throughput gateways with delivery guarantees, retry semantics and the reporting that proves both.
Provisioning and OSS/BSS
Ordering, activation, rating and billing, where an error is a customer-visible charge and a regulatory question.
Network-adjacent tooling
Monitoring, capacity planning and fault correlation across equipment from vendors who do not agree on anything.
Self-service portals
Customer-facing account, usage and support, which is where most of the cost-to-serve actually sits.
Unit economics
What a single message costs, and where that cost goes.
At two hundred million messages a month, a tenth of a penny is two hundred thousand pounds. This is the arithmetic that decides the architecture, and it is often done after the architecture is chosen. The figures below are illustrative, not a client’s numbers; the proportions are the point.
Ingress and validationParse, authenticate, rate-limit. Cheap per unit, and the first thing to fall over under a flood.
0.004pper message
Routing decisionChoosing a carrier by price, quality and availability. Where the margin is actually made.
0.011pper message
Carrier costThe part you do not control. Everything else exists to spend this well.
1.900pper message
Delivery receipt handlingAsynchronous, out of order, sometimes twice. Cheap to process, expensive to get wrong.
0.006pper message
Storage and retentionWhat regulation requires you keep, for as long as it requires. Compounds quietly.
0.002pper message per year
The number that mattersEverything above carrier cost. This is the whole engineering budget, per message.
0.023pcontrollable
Non-negotiable
What has to be true at any volume
- Backpressure by design
- Every queue has a limit and a defined behaviour when it reaches it. Systems that grow queues without bound fail in the worst way at the worst time.
- Measured per unit
- Cost and latency tracked per message rather than per month, because that is the number that decides the architecture.
- Deploy under load
- Rolling, reversible, and safe while traffic flows. There is no window in which you can stop.
- Retention and lawful access
- What is kept, for how long, and who may request it, implemented as designed policy rather than as an afterthought.
Where AI actually lands
Four places it earns its cost, and one line we do not cross.
Worth doing
- Fault correlation, turning ten thousand alerts into the three underlying causes.
- Fraud and abuse detection on traffic patterns, which is the same problem as payments fraud with different units.
- Customer support deflection, where volume is high and most contacts are the same handful of questions.
- Capacity forecasting against real historic traffic rather than a vendor’s spreadsheet.
Where we stop
Automated changes to live network configuration. The blast radius is the whole customer base, and nobody has an appetite for a model that can take the network down faster than a person can.
Evidence
LOX24, a messaging business, is on our client wall. We have no telecom case study to publish yet, so this page sets out how we would approach the work.
How we deliver it
The services this work draws on.
Backend Development
APIs, data models, and the boring reliability that everything above depends on.
DevOps and Cloud
Pipelines, infrastructure as code, and deploys that stop being an event.
Enterprise Software
Platforms that hold up under real load, real compliance, and real headcount.
Other sectors we work in
Next step
Tell us what your platform has to carry at ten times today.
A senior engineer reads every enquiry and replies within one business day with an honest view of what it will take, including where the cost per message goes as you grow.
