Carrier Interconnect Settlement: Why Your Minutes and Theirs Will Never Match, and What to Build

Why interconnect settlement is a legal reconciliation, not a billing run

It is the eighth working day after period close at a wholesale voice carrier with sixty something bilateral partners. An invoice has arrived from a partner in another region claiming a figure for terminated minutes. The carrier's own call detail records say the volume was materially lower. The dispute window in that agreement is thirty days from invoice date. Nobody in the building can produce, today, a defensible statement of exactly where the two views diverge, so the wholesale manager exports both sides to a spreadsheet, pivots by destination and by day, and starts guessing at which prefix group is causing it.

This is not a billing problem in the retail sense. In retail billing you are the authority on what happened. In interconnect you have a counterparty with their own switches, their own mediation and their own rating, and both of you are right about your own data. Settlement is the process of agreeing a number between two parties who measured the same traffic with different instruments, under a contract that specifies how disagreements get resolved and by when. The system you build is evidence infrastructure first and an invoicing tool second.

Problem 1: your minutes and their minutes will never match, and that is normal

The variance has ordinary causes and they are all worth enumerating before anyone accuses anyone of anything. Billing increments differ, so a partner rating in sixty second increments and you rating in one second increments produce different billed minutes from identical calls. Rounding conventions differ at the call level and at the aggregate level. Answered versus attempted treatment differs when a call is answered by an announcement. Short duration calls may be subject to a grace period in one direction and not the other. Day boundaries depend on whose time zone and whose clock, and a mismatch there moves traffic between periods rather than losing it. Records go missing at a mediation feed on either side.

The operational fix is to stop treating variance as a scandal and start treating it as a measurement with a tolerance. Set a policy threshold, reconcile every partner every period automatically, and only escalate what breaches it. The system should be able to say, within minutes, that the gap is concentrated in one destination group on three days and is consistent with an increment mismatch, which turns a four week argument into a phone call. That capability is the entire value proposition of the build.

Problem 2: rating a partner requires their agreement, not your tariff

Every bilateral agreement is a small legal document that encodes rates by destination prefix, effective dates, currency, rounding and increment rules, minimum durations, volume commitments and tiers, and surcharges. None of it generalises. The operational reality is worse than the legal one: a partner emails a spreadsheet with several thousand rows containing rate changes, dial code additions and dial code deletions, effective in seven days, and if you do not load it correctly by the effective date you rate a week of traffic wrong and find out at settlement.

The hard engineering here is effective dated longest prefix matching at volume. A destination is matched to the most specific prefix in force on the date and time of the call, and the same call must rate identically if re-rated two years later during a dispute, which means the rate table needs full history rather than an updated current state. Rate sheet ingestion needs to be a routine operation your commercial team performs, with a difference report showing exactly which codes moved before anything is committed, rather than a change request to an engineer. Operators who get this right run a rate sheet load in an hour. Operators who do not are the ones whose disputes are really about rate loading errors rather than minutes.

Problem 3: jurisdiction and regulated rates are their own subsystem

In the United States, access settlement depends on determining jurisdiction from the calling and called numbers, using industry reference data to establish where numbers sit, because interstate and intrastate treatment differ. Access billing exchanges its own record formats, and the regulatory regime for terminating switched access has been on a long transition toward bill and keep since the FCC's intercarrier compensation reform, with the practical result that your rate logic has to carry historic rules by period rather than a single current answer.

Mobile operators face regulated termination rates that change on a regulator's calendar rather than a commercial one, which means a rate change you did not negotiate can land mid period. Roaming settlement is a separate discipline again, exchanging usage through the GSMA's file formats and settling against inter operator tariffs, with the industry's billing and charging evolution work moving that exchange forward. An operator that does voice interconnect, access and roaming has three settlement models under one roof, and treating them as one product is how a build fails.

Problem 4: netting and disputes settled from screenshots

Two carriers who exchange traffic in both directions owe each other, and the number that actually moves is the net. In most operators the net position is known several weeks after close, which means treasury cannot plan and the wholesale team cannot make routing decisions with margin in view. If you know on day three that a partner's inbound volume has collapsed while your outbound cost to them is unchanged, you can act. On day twenty five you can only file it.

Disputes deserve a first class object. A dispute has a partner, a period, an amount, a stated cause, an evidence set, a deadline from the agreement, an owner and an outcome. The evidence must be a saved query that re-runs, not a screenshot pasted into an email, because the counterparty will come back three weeks later with a partial acceptance and you need to reproduce the position exactly. Partial acceptances, credit notes and the effect on the next netting statement all have to flow through automatically or the dispute becomes a second reconciliation problem on top of the first.

Problem 5: the traffic that arrives disguised

Settlement records are where you see interconnect fraud, because fraud is a commercial pattern before it is a technical one. Traffic refiling and re-origination, calling line identity manipulation that changes the jurisdiction or the rate that applies, and artificial inflation of traffic toward high payout destinations all show up as a shape in the settlement data: a destination group whose volume moves in a way the commercial relationship does not explain, or a sudden change in average call duration on one route. Build the alarms into the same pipeline that produces the settlement, because a separate fraud tool looking at separate data will always be arguing with a different number.

What TEOCO, Subex, Comarch and Hansen actually do, and where they stop

These are serious products and for a large operator they are a reasonable answer. TEOCO and Subex both bring mature interconnect and settlement capability with deep control libraries and long operator track records. Comarch and Hansen come at it from full business support system estates, which is an advantage if you are buying the whole stack and a disadvantage if you want settlement alone and do not intend to replace billing.

The friction for a mid sized voice carrier, transit provider or growing mobile operator is consistent across all four. Onboarding is a project of its own before the first reconciliation runs. Commercial models are scaled to operator sizes above yours. Most importantly, partner file format onboarding tends to route through the vendor, and partner formats are the one thing that changes constantly and unpredictably, because every counterparty invents their own. If adding a new partner's invoice format takes six weeks through a support queue, the system is slower than the business it serves, and the wholesale team will keep a spreadsheet alongside it. Once that spreadsheet exists, you own two versions of the truth and have paid for one of them.

What a custom build has to include

• An effective dated rate engine with full history and longest prefix matching, capable of re-rating any past period at the rules that were in force.
• Self service rate sheet ingestion with a difference report per load, so commercial staff see which codes were added, changed or deleted before committing.
• A partner format layer where a new counterparty file type is a configuration and a mapping, done in days by your own team, not a vendor change request.
• Bilateral reconciliation with tolerance policies, variance decomposition by destination group and by day, and automatic classification of the likely cause.
• Dispute case management with agreement deadlines, evidence stored as reproducible queries, partial acceptance handling and credit note flow through.
• Netting statements with currency and foreign exchange date conventions taken from each agreement rather than a global setting.
• A near real time margin view per route and per partner, feeding routing decisions instead of arriving after them.
• Fraud pattern alarms on the same records that produce settlement, so the fraud conversation and the settlement conversation share one number.

What it costs and how long it takes

A first release covering the rate engine with effective dating, call record ingestion, partner invoice ingestion for your top counterparties, bilateral reconciliation with variance decomposition and dispute cases runs $90,000 to $200,000 and ships in 14 to 20 weeks. A full platform adding netting statements, multi currency, access jurisdiction determination, roaming settlement, the routing margin feed and fraud alarms runs $250,000 to $600,000 phased over 9 to 15 months.

Price moves on specific things. Record volume, because rating a few million calls a day and rating a few hundred million are different engineering problems. The number of settlement models you carry, since voice interconnect, access and roaming are effectively three products. Jurisdiction determination, if you settle access in the United States, because that brings reference data licensing and its own update cadence. Currency and foreign exchange handling, which sounds small and is not once agreements specify different rate dates. And the number of distinct partner file formats you must ingest on day one, which is the item most often underestimated in scoping.

Build versus buy, and the honest threshold

Buy or stay put if you run a small number of partners on a single softswitch whose bundled wholesale billing already rates them acceptably. At that scale the settlement spreadsheet is annoying rather than expensive, and the money belongs in traffic acquisition.

Our position on the threshold: build when the value of a single disputed month exceeds the cost of a month of engineering, which for most wholesale voice operators happens somewhere past fifteen to twenty active bilateral relationships, and earlier if any single partner represents a large share of your traffic. The second trigger is time. If your netting position is known four weeks after close, you are making routing and credit decisions on stale margin, and the loss from that is invisible because it never appears as a line item anywhere. The third is disputes you are losing not on the merits but because you cannot assemble evidence inside the contractual window. That last one is pure, recoverable money and it is usually the argument that gets the project funded.

How to choose a developer for an interconnect settlement build

Ask them to explain how they will re-rate a call from two years ago at the rates in force on the day it was made. If the answer involves updating a current rate table, stop the conversation. Effective dated rating with full history is the foundation and retrofitting it later means rebuilding everything above it.

Ask how a new partner's invoice format gets onboarded and who does it. The right answer puts your own commercial or operations team in control with a mapping interface, because counterparty formats change without notice and a vendor queue in that path guarantees a shadow spreadsheet.

Ask what they will do about variance classification, not just variance detection. Any competent developer can show you that two numbers differ. The valuable behaviour is telling you that the gap sits in one destination group across three days and matches an increment convention mismatch. Ask them to describe how they would build that.

Ask who owns the code and get repository and infrastructure ownership in the contract before kickoff. At Digital Heroes the client owns the code from the first commit, which matters here because your rate history and settlement records are the evidence base for disputes that may run for quarters. Bring us two things to start: one partner agreement with its rate sheet, and one month where your number and theirs disagreed. We will scope from those rather than from a feature list.