Skip to content
Preptima
hardScenarioDesignMidSeniorStaff

A subscriber goes abroad and their roaming charges only reach you days later. How does charging cope with that?

Roaming splits usage from its record: the visited operator meters it and sends files later, so postpaid rates on records that may miss the bill and prepaid cannot wait at all. The design turns on which usage you allow before you have proof of it, and what the delay costs when it is fraud.

6 min readUpdated 2026-07-29Target archetype: Enterprise Captive
Practice answering out loud

What the interviewer is scoring

  • Does the candidate separate the visited operator metering the usage from the home operator charging for it
  • Whether prepaid roaming is recognised as needing real-time authorisation rather than file-based rating
  • That the exposure window is quantified as spend a subscriber can incur before any record reaches you
  • Whether records arriving after a bill has closed are given a defined destination rather than being dropped
  • Does the answer name the wholesale side and the possibility of disputing the other operator's records

Answer

Two operators, two records, one subscriber

Domestically, the network that carries the traffic and the business that bills for it are the same organisation, so mediation gets its records within minutes. Roaming breaks that. The subscriber attaches to a visited operator's network abroad, that operator meters the usage, and the record has to travel from their systems to yours before you can charge anybody. It travels as files, in batches, on an inter-operator schedule — the Transferred Account Procedure format, exchanged directly or through a clearing house that many operators use to avoid maintaining hundreds of bilateral connections.

That gives you two distinct charging problems, and separating them is most of the answer. There is the wholesale problem, which is what you owe the visited operator under your roaming agreement for what they carried. And there is the retail problem, which is what you charge your subscriber, based on your own tariff and their bundle. The two are rated from the same underlying events but by different rules, in different currencies, on different timescales, and they will not agree. Any answer that treats the incoming file as the invoice line for the subscriber has collapsed a distinction that entire teams exist to maintain.

sequenceDiagram
  participant S as Subscriber abroad
  participant V as Visited network
  participant C as Clearing house
  participant H as Home charging
  participant B as Bill
  S->>V: Call or data session
  V->>H: Real time authorisation for prepaid
  V->>C: Usage files on the agreed cycle
  C->>H: Records delivered later
  H->>B: Retail rating and bundle counters

Look at the gap between the second arrow and the fourth: real-time authorisation exists for prepaid because the file path is far too slow, and everything difficult about roaming lives in the space between those two paths.

Prepaid cannot wait, so the network asks first

A prepaid subscriber with no credit must not be able to run up usage abroad, and a file arriving in two days cannot enforce that. So prepaid roaming is authorised online: the visited network signals the home operator's charging system during setup, obtains a quota, and comes back for more as it is consumed. In legacy mobile networks that dialogue was carried by the intelligent-network mechanism designed for it; in packet networks it is the same credit-control pattern used domestically, with the visited network's gateway asking the home charging function for permission.

The practical consequences are the interesting part. The authorisation only works if a roaming agreement supporting it is in place with that operator and the signalling path is available, which is not universal, so an operator has to decide what happens when a prepaid subscriber attaches somewhere that cannot ask. Allowing them on means metering they cannot control; refusing means a customer abroad with credit and no service. Reservation granularity matters too: grant too much and a subscriber can overspend a small balance, grant too little and you generate signalling for every few seconds of a video call across an international link. And a reservation held by a session that dies has to be released, or the balance is quietly unavailable.

Postpaid: rating a record that may arrive after the bill

Postpaid is where the delay bites. The rule that keeps it defensible is the same rule that governs any late usage: rate against the event time, not the arrival time, and against the tariff and bundle state that applied when the usage happened. The counters a bundle maintains therefore have to remain open to amendment after the period has ended, because a record for the twenty-eighth will turn up on the third of the following month and it belongs to the earlier period's allowance.

Then decide, deliberately, what happens to a record that arrives after the bill for its period has been issued. There are only three honest options and each has a cost. You can hold the bill for a defined window, which delays revenue for every customer to accommodate a few. You can bill it in the next cycle as prior-period usage, which is the usual answer and requires the invoice to show clearly that the charge is for an earlier month or you will generate disputes. Or you can absorb records older than a cut-off, which is a decision about revenue that belongs to finance rather than to an engineer, and which needs to be measured so that nobody discovers a large write-off by accident.

What you must not do is let a late record fall into a reject queue nobody works, because that is indistinguishable from the third option except that no one chose it.

The delay is a fraud window as well as a billing one

The reason the industry pushed usage information onto a faster path than the settlement files is that the gap is exploitable. A stolen or fraudulently obtained subscription used abroad can accumulate very large charges before the home operator has any record that anything is happening, and it is the home operator that owes the visited operator for that traffic regardless of whether the subscriber ever pays. The near-real-time exchange between operators exists precisely to shorten that window, so that the home network learns about a roamer's usage soon enough to act rather than at settlement.

So a competent answer states the exposure rather than assuming it away. Ask how much a single subscription can spend before any signal reaches you, per destination and per usage type, and treat that number as the risk being carried. Then put controls on it: velocity checks on newly activated subscriptions that immediately roam, alerts on a subscriber appearing in two distant networks in a short period, thresholds per subscription rather than a single global one, and a defined action when a threshold is crossed. The action is where judgement shows. Barring a genuine customer abroad is a serious thing to do to them, so the response ladder usually runs from notification, through restricting expensive usage types, to a bar only for the clearest cases.

Alongside fraud sits the customer-protection side. Several markets require the operator to notify a subscriber as data spend crosses a threshold and to stop usage at a cap unless the subscriber opts out, and those obligations have to be enforced on information that arrives late by design, which is another argument for the faster usage feed rather than the settlement files.

Where the losses actually appear

The failures that cost real money in this area are quiet ones, so name them as monitoring rather than as design.

Files stop arriving from one partner and nobody notices, because the absence of a file is not an event. The control is an expected-arrival schedule per partner with an alert on a missed window, and a reconciliation of record counts and totals against what the partner says they sent. Records are rejected for format or for referring to a subscription you cannot identify, and the reject queue grows unattended. Rounding and rating rules differ between the wholesale agreement and the retail tariff — different increments, different minimum durations, different treatment of a call that never connected — so the wholesale cost of a subscriber's trip can exceed what you charged them without anyone comparing the two. Currency conversion adds another layer, since settlement happens in an agreed currency and the retail charge is in the subscriber's, and the rate and the date it was applied both have to be recorded to defend the number later.

The reconciliation that closes the loop is per partner and per period: usage you were charged for wholesale against usage you rated retail, with the difference explained. An unexplained divergence is either a rating fault, a fraud, or a partner sending records you should dispute — and you have a contractual window in which to dispute them, so finding out months later is the same as accepting the charge.

Roaming separates the usage from the evidence of it. Decide explicitly how much a subscriber may spend before you have any record, where a record that arrives after the bill will go, and how you would know that a partner's files stopped.

Likely follow-ups

  • Roaming records for one visited network stop arriving entirely. How long before you notice, and how?
  • A subscriber's data allowance was exhausted while abroad but the records arrived after the period closed. What do they owe?
  • How would you decide a spend threshold at which you cut a roaming subscriber off, and what does that risk?
  • Which of the wholesale charges you receive would you challenge, and what evidence would you need?

Related questions

roamingcharginginter-operator-settlementprepaidrevenue-assurance