Can you run a telecom commercial function on public data?
We rebuilt as much of a telecom operator's commercial analytics function as the public record allows — market sizing, operator economics, product portfolio, pricing, forecasting and decision support — and then counted what was missing. About a third of what a commercial team watches can be reconstructed from outside. The third that cannot includes churn, which is why the answer is a qualified no, and why the qualification is the useful part.
MEDIUM-HIGH CONFIDENCE high — the assessment of what is and is not available is itself close to certain, because it rests on exhaustively working through the published record. The commercial conclusions drawn from the data that does exist inherit that data's quality, which is high for the market and financial layers and absent below them
This is not client work. No party commissioned it, no client is named or implied, and the scenario is hypothetical. It is published to show how Nyra applies its public research to a specific commercial decision — the method is real, the project is not.
A commercial analytics function needs to be stood up for a Maldivian telecom operator, or an outside party needs to assess one. How much of that function can be built from the public record, what does it cost to build, and which decisions can it actually support?
Data & analytics
Series construction, modelling and indices built from evidence that does not yet exist publicly.
The public record supports a genuine market intelligence and operator-economics capability: fifteen years of monthly regulator data, thirty operator-quarters of filed income statements that reconcile to published annual figures, and both operators' complete published price lists. It does not support anything at customer level. Churn, product volumes, recharge behaviour and usage are disclosed nowhere, which means retention, elasticity, cannibalisation and lifetime value can be framed and demonstrated but not measured. A function built only on public data can tell management what is happening and what it implies; it cannot tell them which customer to call.
The regulator's series is monthly, official and fifteen years long, and the operator financials are filed statements whose annual sums reproduce both companies' separately published full-year revenue to the rounded million. Product data is exact but is a single point-in-time observation with no history.
The forecasting is properly backtested and the winning model beats the seasonal-naive baseline on every series. The commercial models — elasticity, lifetime value, churn scoring — are structurally sound and empirically unanchored, because the data that would anchor them is not published.
The conclusion about what can and cannot be built is a direct consequence of the inventory, not a judgement call. The commercial recommendation that follows — monetisation before acquisition — rests on an accounting identity in the published data and is unusually firm for this kind of work.
- How much of a telecom commercial analytics function can be built from public data alone?
- Which commercial decisions can that function actually support?
- What is the single most valuable dataset that is missing?
The decision
Someone has to stand up commercial analytics for a telecom operator in a small market — or assess one from outside, as an investor, a lender or a regulator would. The first question is not which model to build. It is what the available data can support, because a model built on data that does not exist is a slide, not a capability.
This case study answers that question by doing the work: building as much of the function as the public record allows, and then counting what is left over. The full apparatus — dashboards, product explorer, scenario engines, forecasts — is at the telecom programme. What follows is what it cost, what it proved, and where it stopped.
What was built
Six layers, in the order a commercial function actually needs them.
Market. The regulator publishes one HTML page per year of monthly subscription statistics. Scraped, parsed on row labels rather than row positions, and validated: 173 consecutive months from January 2012, covering mobile total, prepaid, postpaid, fixed and mobile broadband, data-with-voice, data-only, LTE fixed-wireless, satellite and fixed lines split between Malé and everywhere else.
Operators. Sixty-three filings harvested from the capital-market regulator’s disclosure portal. Income statements parsed out of the PDFs to produce thirty operator-quarters of revenue split by line of business, cost, depreciation, operating result and profit.
Products. Both operators’ published catalogues — 204 products with price, allowance and validity, normalised onto price per day and price per gigabyte.
Quality. Automated identity checks, completeness checks and outlier detection across every dataset, with the findings published rather than filed.
Forecast. Five candidate models, rolling-origin backtests, MASE-scored against a seasonal-naive baseline.
Decision support. Revenue decomposition, elasticity, lifetime value, churn scoring, promotion incrementality and an opportunity ranking — each anchored on observed data where that exists and explicitly labelled where it does not.
What it proved
Three results are worth reporting because they were not obvious before the work.
The extraction is externally validated. Summed by year, the quarterly income statements parsed out of PDFs reproduce both operators’ separately published full-year revenue — MVR 2,216m and MVR 2,838m for 2025, MVR 2,203m and MVR 2,787m for 2024 — to the rounded million [OWN-CALC against OFFICIAL]. Four independent checks, four matches. That is the difference between a dataset and a claim about a dataset.
The two operators’ reported EBITDA are not built the same way. Adding depreciation back to Dhiraagu’s reported operating result reproduces its stated EBITDA to within 0.6%. The same arithmetic on Ooredoo lands 7.6% below its stated figure — MVR 1,167m against MVR 1,263m [OWN-CALC]. Something above Ooredoo’s operating line is excluded from its EBITDA definition, and it is not disclosed. Comparing the two headline margins without saying so, as most published comparisons of these companies do, is a like-for-unlike comparison.
The sophisticated forecasting models lose. Over six rolling origins at a twelve-month horizon, a naive drift model beat Holt-Winters and SARIMA on every series tested. On total mobile subscriptions it achieved a MASE of 0.188 and a mean absolute percentage error of 0.81%; SARIMA managed 0.758 and 3.29%, and on postpaid it diverged outright. Monthly subscription counts move slowly and administratively. There is very little seasonal structure left for a seasonal model to find, and the extra parameters buy nothing but confidence. That result is published rather than quietly replaced with the model a reader would expect to see.
The commercial finding
One finding survived every check and drives everything else. In 2025 the Maldivian mobile market grew its subscription base by 5.2% and its mobile service revenue by 0.1%. Revenue per subscription therefore fell 4.9%, from MVR 342.26 to MVR 325.61 a month [OWN-CALC].
This is an accounting identity, not an inference: the three numbers cannot all be true any other way. It says nothing about cause — low-value additions, migrant composition, price erosion and product mix are all consistent with it, and the public record cannot separate them — but it says something decisive about priority. The denominator is already growing. Acquisition is not the binding constraint in this market; monetisation is.
Where it stopped
Of the metrics a telecom commercial function watches, roughly a third can be reconstructed from outside. The distribution of what is missing is not even, and it is not random.
The financial branch is well covered — quarterly filings with a revenue split. The market growth branch gives the base and its direction but no gross adds and no channel. The product branch gives the entire catalogue and none of the volumes. The usage branch counts subscriptions and no traffic at all.
The retention branch is empty. Neither operator publishes churn in any filing, annual report or results release; neither the regulator nor the exchange requires it. Everything that depends on churn — lifetime value, retention targeting, the economics of an acquisition campaign, the value of an annual prepaid plan — is therefore a framework with a stated assumption rather than a measurement. This programme builds those frameworks, demonstrates them on an explicitly simulated base, and labels them on every screen. It does not present them as findings about anyone’s customers, because a simulated result described as real would be worth less than no result at all.
What we would recommend
Build the market and financial layers from public data — they are genuinely good. For an investor, lender or new commercial hire, the public record supports a defensible view of market size, direction, operator economics and competitive product positioning, and it can be rebuilt and refreshed by one analyst.
Do not attempt the customer layer from outside. Anyone publishing Maldivian telecom market shares, churn rates or elasticities without operator data is estimating and should say so. This programme’s most useful contribution may be the list of things it declines to claim.
Instrument churn first, inside the business. A defined churn event, a monthly cohort survival curve and a value-weighted at-risk population. Every other customer-level model depends on it, and its absence is the largest single gap in this market’s evidence base.
Then re-ladder the mid-tier packs. Effective price per gigabyte varies forty-six-fold inside a single duration band and does not fall monotonically with price. Fixing a ladder is cheap, entirely within an operator’s control, and acts directly on the one number that is falling.
The full decision set, separated into what is observed, what is inferred and what is recommended, is at what should management do.
Sources
- Communications Authority of Maldives — monthly telecom statistics, 173 months
- CMDA InformInvestor — 63 Ooredoo Maldives and Dhiraagu filings, 2023–2026
- Both operators' published product catalogues, observed 18 August 2026
- Nyra — Maldives telecom programme, methodology and data-quality report
- Disclosure of churn or product volumes by either operator, which would move retention and pricing from framework to finding
- Evidence that the subscription growth of 2023–25 is concentrated in genuinely new customers rather than low-value additions, which would weaken the monetisation-first conclusion
- A regulator move to publish traffic or per-operator statistics, which would make market share and usage analysis possible for the first time
- Sustained recovery in revenue per subscription across 2026, which would make the 2025 fall a one-year event rather than a trend
This analysis is built entirely from the public record. To convert it from a preliminary assessment into investment-grade due diligence, Nyra would need evidence that only the party making the decision can open:
- Subscriber-level recharge and usage records, which would convert the churn framework from a demonstration into a working model
- Product-level subscription counts across price changes, without which elasticity cannot be estimated at all
- A defined churn event and a monthly cohort survival curve — the single most valuable missing series in this market
- Acquisition cost by channel, to close the lifetime-value loop
- Operator-side definitions of 'customer', which would make the two operators' published customer counts comparable and market share computable
- Data traffic volumes, which no Maldivian source publishes at any level of aggregation
Nyra publishes independent market and economic research. This material is general information and commercial analysis only. It is not investment, legal, tax or accounting advice, is not a recommendation to buy or sell any security or asset, and does not take account of any reader's objectives or circumstances. Nyra is not a licensed investment adviser. Figures are sourced and tiered; estimates and scenarios are labelled as such and may change as new data is published.