SMS Routing Best Practices: Route Selection and Monitoring
An ineligible route can cause problems regardless of price: traffic sent down a route that was never eligible to carry it may fail whatever the rate. A workable routing policy therefore answers eligibility questions first and price questions second, and it writes both answers down so that the next person can see why a route was chosen.
Decide eligibility before you compare price
Before a rate matters, seven things need to be true for a candidate route. Treat them as gates rather than preferences.
- Destination and network: the route serves the country and the specific mobile networks in your traffic mix, not just the country as a headline.
- Traffic type: authentication, transactional alerting, two-way conversation and promotional traffic are not automatically interchangeable on the same route.
- Sender arrangements: the alphanumeric sender ID, long number or short code you intend to use is permitted, registered where required, and behaves consistently.
- Permitted content and conventions: character set, URL handling, opt-out wording and template rules for that market.
- Receipt semantics: what statuses the route returns, how quickly, and what each one is understood to mean.
- Capacity: the throughput the route is expected to sustain, and how it behaves when you exceed that.
- Cost: the price and the billable units — only now, once the route is eligible at all.
Route labels are shorthand, not specifications. Describing a connection as direct or one-hop tells you something about the transit path, but it does not by itself prove suitability, sender behaviour, receipt quality or delivery speed for your destinations. Those need confirming per destination and network — which is the first thing to work through with an SMSC gateway provider for carriers and SMS aggregators. Equally, routes that work by disguising traffic origin, or by using consumer SIM hardware, tend to fail on sender consistency, receipt reliability and market rules; they are not a shortcut worth building a service on.
A short route review record
Route decisions decay. A compact review record per destination keeps them honest and makes handovers possible.
| Field | Example entry | Why it is recorded |
|---|---|---|
| Destination and networks | Market A, networks 1-3; network 4 excluded | Prevents a country-level assumption hiding a network gap |
| Approved traffic types | Authentication and transactional only | Stops promotional traffic drifting onto a sensitive route |
| Sender arrangement | Registered alphanumeric sender, template-controlled | Sender rules are a common cause of silent rejection |
| Receipt behaviour | Accepted immediately; final status usually within minutes | Sets the window before an absent receipt means anything |
| Assumed throughput | Steady rate agreed; behaviour above it documented | Makes peak planning a calculation rather than a guess |
| Last reviewed / owner | Reviewed on a set cadence by a named owner | Someone has to own the decision to keep or change it |
Accepted is not the same as delivered
The single most useful distinction in routing monitoring is between acceptance and confirmed delivery. Acceptance means your submission was taken for onward handling. Confirmed delivery means a downstream status event has reported the message as delivered. Between the two sit messages that are in progress, messages that failed after acceptance, and messages whose final status never arrives. Public provider documentation such as the Twilio message resource statuses illustrates how many intermediate states can exist — but each platform defines its own vocabulary, so map statuses explicitly instead of assuming a shared meaning.
A missing or late receipt is not proof that a handset did not receive the message. Report those messages as unresolved rather than folding them into a failure count, and state the receipt window you used whenever you quote a delivery figure.
Sample by destination and sender, not in aggregate
An aggregate delivery percentage hides the problems that matter. Break monitoring down by destination, mobile network, sender, traffic type and time of day, and watch error-code distributions rather than a single ratio. The reasoning behind the signals in the Google SRE book transfers well here: watch traffic, errors, saturation and latency together, and prefer percentiles over averages when you look at timing. Averages hide the tail, and in messaging the tail is where authentication traffic fails its purpose.
Set a sampling cadence you can actually sustain — a small, regular review beats an exhaustive review that never happens — and keep a comparable baseline period so that a change has something to be compared against.
Own the change, not just the decision
Every route change should have a named owner, a stated reason, an effective time in UTC, a rollback path and a note of who was told. Where a change affects sender behaviour, receipt semantics or cost, the customer-facing and billing consequences need to move at the same time as the routing configuration.
Two habits are worth building into normal operations: reviewing routes on a fixed cadence rather than only after an escalation, and confirming eligibility before volume moves rather than after a customer notices.
Discuss your routing requirements
Bring your destination and network list, the traffic types you send, your sender arrangements and the delivery evidence you need. We can then talk through how route selection and monitoring would work for your service.
Sources and further reading
Independent references are general context, not evidence of MessageTrade capabilities, partnerships or performance.
Related guides
- SMS Routing Optimisation: Balancing Cost and Delivery Quality
Once a routing policy exists, the ongoing work is trade-off management: comparable cohorts, honest unresolved counts and a cost measure that includes every attempt you paid for.
- SMS Failover and Redundancy: Designing Safer Retries
Redundancy in SMS is less about uptime slogans and more about knowing which failures may be retried, which must be reconciled, and which must simply stop.