Three systems, one bill.
The gap telecom expense management leaves open costs a multi-entity group the most, and your own files prove most of it.
Request a pilot
- Type
- Position
- Updated
- 29 September 2026
- Reading time
- 9 minutes
Riopex is Reduction in OpEx, a telecom expense management platform for groups that hold several legal entities, usually across several countries. It serves the group finance team that consolidates the spend and the regional IT lead who owns the services behind it.
The three disconnections
The first sits between what was ordered and what is delivered. A 100 Mbps port is contracted, a 50 Mbps port is installed, and the invoice carries on as though the order held. The evidence lives in the monitoring system, and the monitoring system knows what runs rather than what it costs.
The second sits between what is delivered and what is still billed. Service status and billing status are different facts, so a service can be operationally disconnected and commercially live at the same time. That is the ghost circuit, and it becomes representable the moment the two are modelled as separate axes.
The third sits between one operating company and the next. Each entity negotiates alone, in its own currency, against its own carrier. Somebody has to hold the comparison, or one operating company pays several times what another pays for the same delivered megabit.
One join, in two steps
Step one needs only files. The bills say what you pay, the contracts say what you signed, and the inventory says what exists. Reconciled per operating company, in each company's own currency, they show on day one what is billed but missing from the inventory, what is held but never billed, and what is priced outside its contract or its term. Cost per megabit subscribed works from here, from the bills alone.
Step two needs the monitoring system. Once a read-level key is connected, utilisation joins the same rows, and cost per megabit delivered compares the same megabit across operating companies. PRTG is the first connector and is in progress; the others are planned. Until one is connected, step one stands on its own.
The three bets
Group hierarchy first. Multi-legal-entity rather than multi-site, with functional currency per entity under IAS 21 rather than a display toggle on a single-currency ledger. Permission boundaries mirror the legal structure, and intercompany allocation and cross-entity comparison sit in the schema. That is a different data architecture, and it is laid down at the start rather than added to a flat one later.
Regional carrier depth. The invoices that matter to a group of this shape are bilingual, carry Arabic-Indic numerals and Hijri dates, are denominated in three-decimal currencies, and are taxed under rules that diverge per jurisdiction. Building for those first is a deliberate choice about which groups we are useful to.
Cost per megabit delivered. Bills joined to utilisation, per service, per site, per operating company, and compared across the group. Both halves of that join exist in the market already, and joining them at group level is the work this product exists to do. It is the second step, and it waits on your monitoring system; the first step does not.
Our commitments
Riopex is bound by the six commitments listed below, and three of them carry reasoning worth reading first. We read the monitoring system that is already there, read-only. Running polling infrastructure inside fifteen countries' networks is, in our estimate, six to twelve months of infosec review, penetration testing and data-processing agreements per customer, for data the customer already collects. Reading the existing system is a smaller ask, it starts sooner, and it leaves your network exactly as your team built it.
The MACD workflow is internal and ends at the carrier boundary. The ordering relationship stays with your team, so what crosses the boundary is a handoff whose reference and acknowledgement we keep as evidence. The MACD capability sets out why the boundary sits exactly there.
Comparison stays inside your own group, where you already hold the right to compare. Pricing-confidentiality clauses make anything wider a consent question rather than a data question.
- Monitoring: We read the monitoring system you already run, read-only, over a read-level key, and put a cost against its numbers.
- Carrier boundary: The MACD workflow is internal and ends at the carrier boundary, with the handoff reference and the acknowledgement kept as evidence.
- Quiet links: Backup, DR, contractual minimums, burst headroom and seasonal capacity are legitimately billed, so a low utilisation figure is a question rather than a verdict.
- Comparison: Comparison stays inside your group, where you already hold the right to make it. Pricing-confidentiality clauses make anything wider a consent question.
- Independence: Our revenue comes from you alone, and we earn zero carrier commission. A group evidencing arm's-length carrier selection needs software with no stake in which carrier wins.
- Provenance: Every field carries its provenance and every correction is reversible. The review queue is a first-class surface, and it is where the work happens.
A quiet link is a working link
Backup and DR circuits are supposed to be quiet. Contractual minimums, burst headroom and seasonal capacity are all legitimately billed. A utilisation figure becomes useful the moment the circuit's role and its contract sit beside it.
So the product earns the word before it uses it. A finding clears six gates first. It needs enough billing history, high enough coverage and a window of at least thirty days. It needs a known role outside backup, DR and burst. It needs a contract with no minimum this service helps meet and an early-termination exposure smaller than the saving. And it needs a join it can show its working for.
A failed gate keeps the candidate, marks it suppressed and records the reason. An unknown role becomes a data-quality task that improves the inventory, which is worth more than a red number that loses you the room.
Onboarding is the product
The schema is easy. The data is the product. Per customer the sources are, in descending reliability: invoices, carrier portals, contracts, the CMDB, router configs, and an engineer's memory. Reconciling them is months of human effort per group, so we budget it openly and build the machinery that makes it cheaper.
That machinery is product rather than tooling. Raw bytes for everything ingested, kept forever. Imports with a dry run and an undo that rolls back to the pre-batch state. Provenance as a row rather than a column, so a conflict between the CMDB and the invoice stays visible. And one queue for every unresolved thing, because an implementation team lives in one inbox for weeks.
Three mechanisms make month two cheaper than month one, and each is built. Saved mappings are keyed to the shape of a carrier's header row, so its next CSV or XLSX invoice reads without remapping. Opt-in import cleanup fixes placeholder IDs, cells holding several IDs, bandwidth written as a 100M/20M pair and the company's own words for circuit roles, once, in the import rather than by hand. And carrier confirmation sends each carrier what you hold for it, so the carrier checks your inventory for you and its answers come back as evidence, field by field, without overwriting a record.
5–10% of spend typically surfaces as unjustified services once an inventory exists, and that happens before a single utilisation reading is joined.
Built in the open
The domain model is locked: four levels of hierarchy, two of them legal, money held in integer minor units, and FX rates that are append-only. The platform is built against it and runs today on a fictional demo group, and the first pilot groups will shape the order the next capabilities land in.
Every named integration carries its status label, so a page says what it is today rather than what it will be. The first monitoring connector, PRTG, is tested against recorded fixtures, and its units are asserted in a fixture test.
We are asking for one thing: a group with this problem that would like it counted properly. Describe how the entities are arranged, and the reply asks for three invoices.
Two numbers worth knowing
- 5–10%
- of spend typically surfaces as unjustified services once an inventory exists
- 40–80
- invoice formats across 15 countries for a group of this shape: our estimate, with the working in the article on invoice formats
