Solutions · by industry · telecommunications

Telecom: churn, plan migration, outage minutes and recharge value, read every morning — on separate stars, without an invented bridge

Q: What if a metric returns an error?

It is marked broken in the model note, the agent stops spending queries on it, and every sentence that needed it is held with the reason.

CIO and head of data (billing, CRM, provisioning and the network operations system are separate stars; the page says what a join needs) · chief commercial officer and marketing director (churn, ARPU, plan migration, gross adds) · network operations director (outage minutes, tickets, time to restore) · channel sales director and prepaid product manager (dealer vs digital adds, recharge value). The buyer is the CIO with the chief commercial officer; the daily readers are regional commercial managers and the network operations centre lead.

Questions

The three questions a telecom board asks every month

Where did churn rise — which region, which plan, which tenure — and was it port-out?
Did outage minutes and tickets move in the same region, and how long did restoration take?
Are the gross adds we pay dealers for the ones that recharge?

In conversation

Sample questions people type into the conversation

deepen path — the finding's evidence is carried frozen; a new number needs a new query, and the agent says so

"Break Region East's postpaid churn down by plan and tenure band."
"How many of the downgrades came from the two unlimited plans?"
"Show East's outage site-minutes by cause code and week."
"Which dealer tier's adds recharge least within 30 days?"
"How current is the billing data?"

Personas

Three example agents

The agents below are examples for this industry, not a fixed set: Chief Commercial Officer Agent · Network Service Performance Agent · Channel & Recharge Agent. Each one’s scheduled run is shown in the example sets that follow.

Set 1

Chief Commercial Officer Agent

Audience: the chief commercial officer, regional commercial managers, the pricing manager.

Set 2

Network Service Performance Agent

Audience: the network operations director, regional network operations managers, the field operations lead.

Set 3

Channel & Recharge Agent

Audience: the prepaid product manager, the channel sales director, dealer managers.

An agent's name says which measures it reads and for whom. It holds none of the role's authority: it does not decide, approve or act.

Agents for the topics you choose

An agent is a job description, a bound semantic model and a schedule — not code — so the topics it watches are yours to set: a payer, a service line, a campaign, a supplier. We write the first ones with you.

Build your agents with us →

Example sets

Three example sets, one run each

Each set is one agent persona and one scheduled run’s output — the brief, the recommendations and the findings they rest on — in the product’s own record envelope. The first set is open; the other two are collapsed. The model and the measures the sets assume are listed at the foot of the page.

IllustrativeSet 1Chief Commercial Officer Agent

Persona: Reads churn, ARPU, plan migration and fibre growth for a mobile and fibre operator with about nine million subscribers in one country (4.86 M postpaid, 3.9 M prepaid, 1.21 M fibre); separates "the base shrank" from "the base moved down-plan", and "one region" from "the country". · Audience: the chief commercial officer, regional commercial managers, the pricing manager. · Tone: executive-summary language, number-led, no adjectives. · Output language: en
Signals it watches: (1) Postpaid churn by region and plan, with the port-out share of disconnects. (2) Plan migration — downgrades against upgrades, and the ARPU of the migrated cohort. (3) ARPU read beside churn, never alone. (4) Fibre gross adds against fibre churn. (5) Net adds turning sign.
Thresholds it was given: finding 15 % · critical 40 % · ratio metrics (churn rate, ARPU, penetration) 8 % · minimum share 2 % of the segment base · severity rule: medium when a threshold is crossed, high when a ratio metric moves more than 1.25× its threshold or a count metric more than 1.5× the finding threshold, critical above the critical threshold; a finding with a breakdown cell takes the severity of the cell's own move; an ongoing topic keeps the severity it was given

Run header

Scheduled monthly on the 3rd, 06:00, Europe/Berlin
Period: August 2026 vs July 2026 (month over month; churn is a monthly rate on the opening base, unaffected by the 31/31-day match)
Data complete through 2026-09-02, measured via max(BILLING_MONTH) (billing close)
Scope: the agent runs under the commercial scope (all regions, no network metrics)
Model: the workspace's approved model, written onto the run
Cost: recorded on the run
Amounts in EUR

Brief

August 2026: Postpaid Churn 1.12 % → 1.41 % Led by Region East at 2.06 %, Downgrades +63.85 % With the Cohort's ARPU −26.97 %, Fibre Gross Adds +21.41 % on Flat Churn — Net Mobile Adds Turned Negative

Assumptions strip

Month-over-month comparison (August vs July 2026) was used; churn is a monthly rate on the opening base. Network metrics are outside this agent's scope and do not appear; nothing in this brief links a churn event to a network event. The price list is not in this model.

The postpaid base lost more subscribers in August than it gained — net adds +8,200 in July, −11,400 in August: 68,530 disconnects against 54,430 in July took monthly churn from 1.12 % to 1.41 %, with Region East at 2.06 % and carrying 26.88 % of the disconnects on 18.40 % of the base; 47.00 % of the disconnects are port-outs. Inside the base, plan downgrades rose 63.85 % while upgrades fell, and the downgrading cohort now pays 17.60 € against 24.10 €. Postpaid ARPU still rose 1.90 %, so the revenue line has not shown the loss yet. Fibre moved the other way — gross adds up 21.41 % on unchanged churn — and stands at 38.41 % penetration. A sentence attributing the downgrades to a price rise was returned once by the claim check; the rewrite states the migration without a cause and passed, because the price list is not in the model.

covers: Finding 1, Finding 2, Finding 3

— Run: monthly, the workspace's approved model, data complete through 2026-09-02. Guard Ledger: held 0 after one coached rewrite — candidate f-02 draft carried after the July price rise, gate 4 returned it as unevidenced causality, the rewrite dropped the clause and passed; not_selected 0; duplicate 0.

Recommendation 1

Break Region East's August postpaid disconnects down by plan, tenure band and port-out flag, against July, and hand the record to the retention team as a segment list, not a subscriber list

based_on: Finding 1 · owner: retention manager, with the East regional commercial manager

If East's port-outs sit in one tenure band or one plan, the retention question is narrow; if they are spread, it is regional. All three cuts are in this model. The destination operator is not; the retention manager reads the porting clearing-house report beside the record.

The findings this rests on

Finding 1

Postpaid Churn 1.12 % → 1.41 % of the Opening Base (+25.90 %) — Region East at 2.06 %, 26.88 % of Disconnects on 18.40 % of the Base; Net Adds Turned From +8,200 to −11,400

severity highdirection upnovelty newtrust Abreakdown REGION = East

Of 4.86 M postpaid subscribers at the start of August 2026, 68,530 disconnected during the month (1.41 %) against 54,430 in July (1.12 %): a +25.90 % change in a ratio metric, 3.24× the 8 % ratio threshold — severity high; the 40 % critical threshold is not reached. Region East disconnected 18,420 of its 894,000 subscribers (2.06 %); it holds 18.40 % of the base and 26.88 % of the month's disconnects. Postpaid ARPU moved from 21.60 € to 22.01 € (+1.90 %, below threshold), so the revenue line does not show the base loss yet; net adds turned from +8,200 in July to −11,400 in August. The disconnect reason code is in the model: 32,209 of August's 68,530 disconnects (47.00 %) carry port-out, and the destination operator is not a field. Network quality is not a metric in this star; whether East's churn moves with East's outages is a question for the network agent's own record, not a bridge this data has.

Recommendation 2

Put the downgrade count and the downgrading cohort's ARPU change on a weekly watch by prior plan, and give the pricing manager the plan cut for August

based_on: Finding 2 · owner: pricing manager

A migration that costs 26.97 % of the cohort's ARPU is a revenue fact before it is a pricing question. The weekly watch will show whether August was one month or a run; the agent stamps a repeat ongoing. The price list is not in the model, and the record makes no claim about a price change.

The findings this rests on

Finding 2

Plan Migration Turned Negative: Downgrades 21,300 → 34,900 (+63.85 %) While Upgrades 28,100 → 26,400 (−6.05 %) — the Downgrading Cohort's ARPU 24.10 € → 17.60 € (−26.97 %)

severity criticaldirection upnovelty newtrust Abreakdown MIGRATION_TYPE = Downgrade

Plan downgrades rose from 21,300 in July 2026 to 34,900 in August (+63.85 %, above the 40 % critical threshold — severity critical) while upgrades fell from 28,100 to 26,400 (−6.05 %, below threshold); net migration moved from +6,800 to −8,500. The subscribers who downgraded in August paid 24.10 € on their prior plan and 17.60 € on their new one (−26.97 %). Two unlimited plans account for 22,100 of the 34,900 downgrades (63.32 %); the plan cut is in this model. The price list and its change dates are not in this model, and the sentence the agent drafted about a price rise was returned once by the claim check and rewritten without the cause — the record states the migration and nothing about why.

Also published in this run, with no recommendation resting on it

Finding 3

Fibre Gross Adds 18,400 → 22,340 (+21.41 %) While Fibre Churn Held at 0.64 % — Penetration 38.41 % of 3.15 M Homes Passed

severity mediumdirection upnovelty newtrust Abreakdown SEGMENT = Fibre

Fibre gross adds rose from 18,400 in July 2026 to 22,340 in August (+21.41 %, 1.43× the 15 % finding threshold — severity medium) while fibre monthly churn moved from 0.62 % to 0.64 % (+3.23 %, below threshold). The fibre base of 1.21 M subscribers stands at 38.41 % of 3.15 M homes passed (1.21 of 3.15 M). The mobile base lost subscribers in the same month; the fibre base gained them — the two segments are stated separately. The campaign flag is not on the activation row in this model, so whether the added fibre subscribers came on a promotion cannot be read here.

What this run could not see.

The destination operator of port-outs, the price list and its change dates, the campaign flag on fibre activations, and anything about the network. The network star is another agent's scope and another star.

IllustrativeSet 2Network Service Performance Agent

Persona: Reads outage site-minutes, trouble tickets per 1,000 subscribers and time to restore for the same operator's network star; separates "the network was worse" from "one region's sites were down", and never writes a sentence that links an outage to a churn figure — the two stars have no bridge. · Audience: the network operations director, regional network operations managers, the field operations lead. · Tone: operations-report language, exact. · Output language: en
Signals it watches: (1) Outage site-minutes by region and cause code, month over month. (2) Trouble tickets per 1,000 subscribers by region, beside the national rate. (3) Mean time to restore by fault type. (4) Site availability as the ratio behind the minutes, stated but not used as the threshold metric — availability moves in hundredths of a point. (5) Data states on decommissioned technology layers.
Thresholds it was given: finding 15 % · critical 40 % · ratio metrics (tickets per 1,000, hours to restore) 8 % · minimum share 2 % of network sites · severity rule: medium when a threshold is crossed, high when a ratio metric moves more than 1.25× its threshold or a count metric more than 1.5× the finding threshold, critical above the critical threshold; a finding with a breakdown cell takes the severity of the cell's own move; an ongoing topic keeps the severity it was given

Run header

Scheduled weekly, Monday 05:30, Europe/Berlin
Period: August 2026 vs July 2026 (31 days each; outage minutes are per calendar day and the match is equal)
Data complete through 2026-09-20, measured via max(EVENT_DATE) on the ticket star, 5 hours ago; the outage star loads daily and shares the edge
Scope: the agent runs under the network operations scope (all regions, no commercial metrics)
Model: the workspace's approved model
Cost: recorded on the run

Brief

August 2026: Region East Outage Site-Minutes +112.14 % on Power Faults, Tickets per 1,000 3.10 → 4.70 Against a National 3.10, Time to Restore 7.4 → 11.2 Hours — a Sentence Linking This to Churn Was Held

Assumptions strip

Month-over-month comparison (August vs July 2026) was used; both months have 31 days. The network star and the commercial star are not joined; the subscriber count is used only as a denominator. The 2G dropped-call metric returned no rows for August and is recorded as a data state.

Region East carried August's outage rise: site-minutes without service more than doubled to 87,400 on 1,840 sites, with the power cause code at 58.00 % of the month's minutes, while the other five regions moved little. The region's subscribers raised 4,202 tickets, 4.70 per 1,000 against a national 3.10, and each took 11.2 hours on average to restore against 7.4 in July. The agent drafted a sentence saying this was why East's churn rose; it was held after one rewrite, because no bridge between the two stars exists in this model and the record makes no such claim.

covers: Finding 1, Finding 2

— Run: weekly, the workspace's approved model, data complete through 2026-09-20. Guard Ledger: held 1 — candidate b-01 draft, reason unevidenced_causality (gate 4): clause which is why East's churn rose referenced the commercial star, which this agent cannot open; one coached rewrite kept the link, then held; data state 1 — envelope e4 (dropped_call_rate, 2G layer) returned 0 rows for 2026-08; not_selected 0; duplicate 0.

Recommendation 1

Break Region East's August outage site-minutes down by cause code and by week, with the mean time to restore for the same fault type beside them

based_on: Finding 1, Finding 2 · owner: regional network operations manager, East

If the power minutes sit in one or two weeks, it is an event; if they run through the month, it is the sites' power condition. Both cuts are in this model. The record ends there: which sites to visit first is the field lead's call, and the served-subscriber count that would rank them is not in the star.

The findings this rests on

Finding 1

Outage Site-Minutes in Region East 41,200 → 87,400 (+112.14 %) on 1,840 Sites (18.93 % of the Network) — Availability 99.95 % → 99.89 %

severity criticaldirection upnovelty newtrust Abreakdown REGION = East

Region East's 1,840 sites recorded 87,400 site-minutes without service in August 2026 against 41,200 in July (+112.14 %, above the 40 % critical threshold — severity critical). Expressed as availability the move is 99.95 % → 99.89 %, which is why this agent's threshold metric is the minutes, not the ratio. The cause-code cut is in this model: power carries 50,692 of August's 87,400 East minutes (58.00 %), up from 12,772 of 41,200 (31.00 %). The other five regions moved inside ±12 %. This data cannot say how many subscribers were under a site when it was down — the served-subscriber count per site is not in the network star, and no subscriber-level statement is made.

Finding 2

Trouble Tickets per 1,000 Subscribers in Region East 3.10 → 4.70 (+51.61 %) While the National Rate Moved 2.90 → 3.10 (+6.90 %) — Mean Time to Restore in East 7.4 → 11.2 Hours (+51.35 %)

severity criticaldirection upnovelty newtrust Abreakdown REGION = East

Tickets raised by Region East's 894,000 subscribers rose from 2,771 in July 2026 to 4,202 in August, taking the rate from 3.10 to 4.70 per 1,000 (+51.61 %, above the 40 % critical threshold — severity critical), while the national rate moved from 2.90 to 3.10 (+6.90 %, below threshold). Mean time to restore for East tickets rose from 7.4 to 11.2 hours (+51.35 %); the fault-type cut places the added hours under site power. The subscriber count used as the denominator is the base from the commercial star as of month-end, read as a plain number; no commercial metric beyond the count is opened by this agent. Whether the ticket rise and the outage rise are the same subscribers is not knowable here — tickets are not keyed to sites in this model.

Data state

Data state (not a finding): the dropped_call_rate metric on the 2G layer returned no rows for August 2026 — the layer was decommissioned in the region during the month. No severity; the model owner is notified to retire the metric from the layer.

Recommendation 2

Put mean time to restore on a weekly watch by fault type and region, and ask the model owner to retire the dropped-call metric from the decommissioned layer

based_on: Finding 2 · owner: field operations lead, with the data-model owner

The restore time rose by half in one region; a weekly watch by fault type will show whether it is power alone. The empty 2G metric will otherwise produce a data state every run and spend a query on nothing.

What this run could not see.

Subscribers served per site, the site key on tickets, and anything on the commercial star beyond the subscriber count. Whether East's churn and East's outages are one story is a question this model cannot answer, and the page says so below.

IllustrativeSet 3Channel & Recharge Agent

Persona: Reads prepaid recharge revenue and count, gross adds by channel and dealer tier, and the first-recharge rate of new prepaid subscribers for the same operator; separates "fewer recharges" from "smaller recharges", and "more adds" from "adds that stay". · Audience: the prepaid product manager, the channel sales director, dealer managers. · Tone: plain and commercial. · Output language: en
Signals it watches: (1) Average recharge value — recharge revenue over recharge count, by denomination and channel. (2) Gross adds by channel, with each channel's share. (3) First recharge within 30 days by channel and dealer tier. (4) Dealer commission cost where the metric works. (5) Recharge frequency per active prepaid subscriber.
Thresholds it was given: finding 15 % · critical 40 % · ratio metrics (average recharge value, first-recharge rate, channel share) 8 % · minimum share 2 % of gross adds · severity rule: medium when a threshold is crossed, high when a ratio metric moves more than 1.25× its threshold or a count metric more than 1.5× the finding threshold, critical above the critical threshold; a finding with a breakdown cell takes the severity of the cell's own move; an ongoing topic keeps the severity it was given

Run header

Scheduled weekly, Tuesday 06:00, Europe/Berlin
Period: August 2026 vs July 2026 (31 days each)
Data complete through 2026-09-19, measured via max(EVENT_DATE) on recharges and activations, 7 hours ago
Scope: the agent runs under the channels scope (all regions, prepaid and activation stars only)
Model: the workspace's approved model
Cost: recorded on the run
Amounts in EUR

Brief

August 2026: Average Recharge Value 6.84 € → 6.27 € on More but Smaller Recharges, Dealer Adds −20.10 % While Digital Adds +41.96 %, Dealer Adds' 30-Day First Recharge 71.20 % → 63.40 % — Commission Cost Unreadable

Assumptions strip

Month-over-month comparison (August vs July 2026) was used; both months have 31 days. The dealer commission metric returns a source error and is marked broken; every sentence that needed it was held. This agent reads the prepaid and activation stars only.

Prepaid subscribers recharged slightly more often in August but for less: 3.98 M recharges worth 24.96 M €, an average of 6.27 € against 6.84 € in July, with the two smallest denominations now 49.00 % of recharges. Gross adds held near July's level, but the mix moved from dealers to digital — dealer adds fell 20.10 % to 54.15 % of the total while digital adds rose 41.96 % — and the dealer adds that did come recharged less often within 30 days, 63.40 % against 71.20 %, while digital adds held at 81.60 %. What the dealers were paid for those adds could not be read: the commission metric is broken.

covers: Finding 1, Finding 2

— Run: weekly, the workspace's approved model, data complete through 2026-09-19. Guard Ledger: held 2 — candidates f-03 and r-02 (commission cost per add), reason source_execution_error on metric dealer_commission (envelope e6 errored; metric marked broken in the model note); not_selected 0; duplicate 0.

Recommendation 1

Break August recharges down by denomination and channel against July, and put the average recharge value of each channel beside it

based_on: Finding 1 · owner: prepaid product manager

If the shift to small denominations sits in one channel, it is that channel's assortment; if it is in every channel, it is the subscriber. Both cuts are in this model. Denomination availability by channel is not, and the product manager reads the channel contracts beside the record.

The findings this rests on

Finding 1

Prepaid Recharge Revenue 26.67 M € → 24.96 M € (−6.41 %) While Recharge Count 3.90 M → 3.98 M (+2.05 %) — Average Recharge Value 6.84 € → 6.27 € (−8.29 %)

severity mediumdirection downnovelty newtrust Abreakdown SEGMENT = Prepaid

Prepaid subscribers made 3.98 M recharges in August 2026 against 3.90 M in July (+2.05 %, below threshold) but paid 24.96 M € against 26.67 M € (−6.41 %, below the finding threshold on its own). The average recharge value fell from 6.84 € to 6.27 € (−8.29 %, 1.04× the 8 % ratio threshold — severity medium), and the denomination cut shows the shift: the two smallest denominations rose from 1,599,000 of 3.90 M recharges (41.00 %) to 1,950,200 of 3.98 M (49.00 %). The base of 3.9 M prepaid subscribers was flat. This data cannot say whether subscribers recharge smaller amounts more often by choice or because a larger denomination was withdrawn from a channel — the denomination availability by channel is not in this model.

Recommendation 2

Put dealer gross adds and their 30-day first-recharge rate on one weekly watch by dealer tier and region, and ask the model owner to repair the commission metric before any cost per add is read

based_on: Finding 2 · owner: channel sales director, with the data-model owner

Fewer dealer adds that recharge less often is two facts on one channel; the tier cut will show whether the lowest tiers carry both. The commission metric is broken and every sentence that needed it this run was held; a cost-per-add figure will exist when the metric does.

The findings this rests on

Finding 2

Dealer Gross Adds 61,200 → 48,900 (−20.10 %) While Digital Gross Adds 22,400 → 31,800 (+41.96 %) — Dealer Share of Adds 66.02 % → 54.15 %; Dealer Adds' First Recharge Within 30 Days 71.20 % → 63.40 %

severity highdirection downnovelty newtrust Abreakdown CHANNEL = Dealer

Prepaid gross adds totalled 90,300 in August 2026 against 92,700 in July (−2.59 %, below threshold); the channels moved apart. Dealer adds fell from 61,200 to 48,900 (−20.10 %, 1.34× the 15 % finding threshold) and digital adds rose from 22,400 to 31,800 (+41.96 %, above the 40 % critical threshold on that cell); own-store adds were 9,600 against 9,100. The dealer share of adds fell from 66.02 % to 54.15 % (48,900 of 90,300 against 61,200 of 92,700). The adds that stay moved too: of dealer adds, 31,003 of 48,900 (63.40 %) recharged within 30 days against 43,574 of 61,200 (71.20 %) in July — −10.96 %, 1.37× the 8 % ratio threshold, and the breakdown cell is the dealer channel, so the finding reads severity high on the cell's own moves — while digital adds held at 25,949 of 31,800 (81.60 %) against 18,390 of 22,400 (82.10 %). The dealer-tier cut is in this model and places the fall in the two lowest tiers. This data cannot say what the dealers were paid — the commission metric returns a source error in this model and is marked broken in the model note.

What this run could not see.

Dealer commissions, denomination availability by channel, and the reason a dealer add did not recharge. Postpaid and network metrics are outside this agent's scope by design.

Every number carries an address

In the product each figure binds to an evidence address — envelope, row, cell, check digit — and a number without one is held, not published. The anatomy of a record →

Derived figures show their operation

A change, a share or a gap is computed by the engine and carries its operands; the model never divides. How every number is proved →

AI governance is the publication layer

Eight deterministic gates, the trust tier, badged interpretation, an assumptions strip that cannot be switched off, and the Guard Ledger. Where the model can and cannot reach →

FAQ

Four questions this page is usually asked

01Our billing, CRM and network data are three different systems. Can one agent read all three?

An agent reads what the imported semantic model holds. Where the stars are not joined, the agent reads each on its own and never invents the bridge — the page's three personas are split along those stars for that reason.

02Does the agent see subscribers?

No. Subscriber and MSISDN are marked personal; the finest cuts are region, plan, tenure band, channel and dealer tier.

03Can it say why churn rose?

It can say where churn rose, in which plan and tenure, and how much was port-out. A cause needs evidence the model holds; a guessed cause is held with its reason, visibly.

04What if a metric returns an error?

It is marked broken in the model note, the agent stops spending queries on it, and every sentence that needed it is held with the reason.

Free data discovery study

See which of these signals your own warehouse can carry.

Send us the name of your data platform and the domain you argue about most. We read your semantic model with you and answer in writing: which of the signals on this page your model carries today, which need a change in Axoria Data Studio, and what a four-week pilot would measure. No cost, no sequence — a person replies within two working days.

Request a free data discovery study → Start a 4-week pilot

What the examples assume

The data, the model and the measures behind the three sets

Nothing in a set rests on a source outside the model described here. Open what you want to reconcile.

01Assumed data sources — data a telecom operator already has

Billing and rating (revenue by subscriber segment, plan, recharge events), CRM (activations, disconnections with reason codes, plan changes, port-out flags), provisioning and order management (fibre orders, homes passed), the network operations system (site availability minutes, faults by cause code), the trouble-ticket system (tickets by region, time to restore), the dealer and digital channel systems (gross adds by channel and dealer tier). Typical warehouse grain: one row per subscriber per month for base metrics, one per event for activations, disconnections, recharges and tickets, one per site per day for availability.

02Assumed semantic model — the minimum for the three personas

Measures: subscriber_base (by segment) · gross_adds · disconnects · churn_rate (engine-derived: disconnects / opening base) · net_adds (engine-derived) · service_revenue · arpu (engine-derived) · plan_upgrades · plan_downgrades · arpu_of_migrated_cohort · homes_passed · fibre_connected · outage_site_minutes · sites · tickets · tickets_per_1000 (engine-derived) · mean_time_to_restore_hours · recharge_revenue · recharge_count · average_recharge_value (engine-derived) · first_recharge_within_30_days · gross_adds_by_channel.
Dimensions: month · region · segment (postpaid / prepaid / fibre) · plan · tenure band · disconnect reason code · port-out flag · channel (dealer / digital / own store) · dealer tier · site cluster · fault cause code · technology layer · recharge denomination. Marked personal and never broken down by: subscriber, MSISDN, dealer owner.
Data edge: measured by the engine via max(BILLING_MONTH) for the commercial star, max(EVENT_DATE) for tickets and outages; the commercial star and the network star have no bridge — the agents never claim that a network event caused a churn event, and the page says so.

03Assumed KPIs — what telecom executives track

KPI	Definition	Unit	Typical band	Usual breakdown
ARPU / ARPA	service revenue / average subscribers (accounts)	currency/month	US postpaid $48–55, prepaid $28–38, fibre $65–75 [vendor]	segment · plan
Monthly churn	disconnects / average subscribers	%/month	US postpaid phone ≈ 0.9–1.0 % [carrier filings]	segment · region
Net adds	gross adds − disconnects	count	company-specific	product · channel
EBITDA margin	EBITDA / revenue	%	mature operators 30–45 % [vendor]	business unit
Subscriber acquisition cost	sales & marketing + subsidies / gross adds	currency	company-specific	channel
Capex intensity	capex / revenue	%	company-specific	network domain
Network availability	available site-minutes / total site-minutes	%	no universal band	region · technology
Outage site-minutes	site-minutes without service	minutes	company-specific	site cluster · cause
Trouble tickets per 1,000 subscribers	tickets / subscribers × 1,000	count	company-specific	region · fault type
Mean time to restore	mean hours from ticket to closure	hours	company-specific	region · fault type
Time to provision	order to active	days	company-specific	product · region
Fixed broadband penetration	subscribers / homes passed	%	company-specific	footprint
Average recharge value	recharge revenue / recharge count	currency	company-specific	denomination · channel
First recharge within 30 days	prepaid adds recharging within 30 days / prepaid adds	%	company-specific	channel · dealer tier

The departments these personas belong to

Sales and revenue operations

Telecom: churn, plan migration, outage minutes and recharge value, read every morning — on separate stars, without an invented bridge

The three questions a telecom board asks every month

Sample questions people type into the conversation

Three example agents

Chief Commercial Officer Agent

Network Service Performance Agent

Channel & Recharge Agent

Agents for the topics you choose

Three example sets, one run each

August 2026: Postpaid Churn 1.12 % → 1.41 % Led by Region East at 2.06 %, Downgrades +63.85 % With the Cohort's ARPU −26.97 %, Fibre Gross Adds +21.41 % on Flat Churn — Net Mobile Adds Turned Negative

Break Region East's August postpaid disconnects down by plan, tenure band and port-out flag, against July, and hand the record to the retention team as a segment list, not a subscriber list

Postpaid Churn 1.12 % → 1.41 % of the Opening Base (+25.90 %) — Region East at 2.06 %, 26.88 % of Disconnects on 18.40 % of the Base; Net Adds Turned From +8,200 to −11,400

Put the downgrade count and the downgrading cohort's ARPU change on a weekly watch by prior plan, and give the pricing manager the plan cut for August

Plan Migration Turned Negative: Downgrades 21,300 → 34,900 (+63.85 %) While Upgrades 28,100 → 26,400 (−6.05 %) — the Downgrading Cohort's ARPU 24.10 € → 17.60 € (−26.97 %)

Fibre Gross Adds 18,400 → 22,340 (+21.41 %) While Fibre Churn Held at 0.64 % — Penetration 38.41 % of 3.15 M Homes Passed

August 2026: Region East Outage Site-Minutes +112.14 % on Power Faults, Tickets per 1,000 3.10 → 4.70 Against a National 3.10, Time to Restore 7.4 → 11.2 Hours — a Sentence Linking This to Churn Was Held

Break Region East's August outage site-minutes down by cause code and by week, with the mean time to restore for the same fault type beside them

Outage Site-Minutes in Region East 41,200 → 87,400 (+112.14 %) on 1,840 Sites (18.93 % of the Network) — Availability 99.95 % → 99.89 %

Trouble Tickets per 1,000 Subscribers in Region East 3.10 → 4.70 (+51.61 %) While the National Rate Moved 2.90 → 3.10 (+6.90 %) — Mean Time to Restore in East 7.4 → 11.2 Hours (+51.35 %)

Put mean time to restore on a weekly watch by fault type and region, and ask the model owner to retire the dropped-call metric from the decommissioned layer

August 2026: Average Recharge Value 6.84 € → 6.27 € on More but Smaller Recharges, Dealer Adds −20.10 % While Digital Adds +41.96 %, Dealer Adds' 30-Day First Recharge 71.20 % → 63.40 % — Commission Cost Unreadable

Break August recharges down by denomination and channel against July, and put the average recharge value of each channel beside it

Prepaid Recharge Revenue 26.67 M € → 24.96 M € (−6.41 %) While Recharge Count 3.90 M → 3.98 M (+2.05 %) — Average Recharge Value 6.84 € → 6.27 € (−8.29 %)

Put dealer gross adds and their 30-day first-recharge rate on one weekly watch by dealer tier and region, and ask the model owner to repair the commission metric before any cost per add is read

Dealer Gross Adds 61,200 → 48,900 (−20.10 %) While Digital Gross Adds 22,400 → 31,800 (+41.96 %) — Dealer Share of Adds 66.02 % → 54.15 %; Dealer Adds' First Recharge Within 30 Days 71.20 % → 63.40 %

Every number carries an address

Derived figures show their operation

AI governance is the publication layer

Four questions this page is usually asked

See which of these signals your own warehouse can carry.

The data, the model and the measures behind the three sets

Where these pages lead

The departments these personas belong to

On this site