
One ranked, permitted action per customer moment — including no action — with the reason recorded.
Several campaigns want the same customer in the same week; contact fatigue and margin loss follow.
Demonstration availableLevel 2 of 5 — A journey runs end to end on sample data, including its refusal and failure states. Bought by CVM and Retention lead. Nothing here has run against an operator system.
Open the demonstrationA customer at a mobile-money counter — the moment a decision is about. The customers here are the sample's; no offer was sent.
Public nowAnyone can do this from this page, without signing in, and the page keeps no record of it. This page, its sample workbook and the sample's review need no sign-in; the steps below say which need more.
5 of 8 customers get a permitted action worth 359 KES in expected value; 3 get no action; 20 action-customer pairs refused by a gate or ineligible.
HeuriTel default synthetic sample — Kenya, in KES, Kiswahili; sample next-best-action-ranking-sample v1. Synthetic rows invented by HeuriTel, sized for Kenya in KES. Not an operator's figures; nobody here is real. Configure a market to see the same sample recomputed for it. Configure a market and this page, the workbook and the review recompute for it.
- 5 customers — Actions chosenbest permitted action with expected value above zero
- 3 customers — No actionnothing permitted, or nothing worth more than doing nothing
- 359 KES — Expected valuesum of response rate × value − cost over chosen actions
- 20 pairs — Gate refusalsan action refused for a customer by a policy or by eligibility
- 3 customers — Contacts avoidedcustomers the ranking left alone
- 62.5 % — Share actionedchosen over customers
Approve the 5 ranked actions (2 × Loyalty call with tailored plan, 1 × Bonus data on recharge, 2 × Recharge reminder) and leave 3 customers alone this week. If nobody acts: 8 customers receive whatever each campaign sends them, 4 of which the contact cap would have refused. No offer was sent. No customer was contacted. No campaign system was updated.
10 steps: 3 public now, 1 after sign-in, 3 in an entitled workspace, 1 a request form, 1 needing an authorised sandbox, 1 not implemented.
Each step says who can take it today. A route existing is not the same as an action being available: the state beside each step is decided by the same rules that refuse a request.
- 01 · What you can buyPublic nowOne ranked, permitted action per customer moment — including no action — with the reason recorded.All ten offers
- 02 · A working demonstrationAvailable after sign-inThe offer's journey on sample data.Open the demonstration
- 03 · The sample, as a workbookPublic now3 data sheets, a manifest and the generated guide, dictionary, validation lists and expected outputs — downloadable without signing in, in the sample's market and currency.Download the sample workbook
- 04 · Validation, on the samplePublic nowcustomers, actions and policies as three sheets: every action has a cost and an expected value per eligible customer; every policy gate names a field and a rule; the no-action alternative is always present. The public review checks a returned workbook against the sample it was issued from, answers with the result before and after, and stores nothing.
POST /api/sample/next-best-action-ranking/review - 05 · Acceptance and the stored simulationEntitled workspaceYour organisation's own workbook, reviewed, accepted as an immutable version with its original retained, and the engine's result stored whole — never recomputed on reopening.Open the workbench
- 06 · Decision and approvalEntitled workspaceA second person decides the recommendation; the maker cannot. Approval applies to a simulated recommendation.The approvals queue
- 07 · EvidenceEntitled workspaceranked permitted action per customer with the reason trace; the refused actions and which gate refused them; the explicit no-action alternative; run register export. The register export, the retained original, the run history, the dry-run plan and the pilot package are downloadable from the workbench.Evidence on the workbench
- 08 · A controlled pilotPilot request availableThe request form is available here; the pilot is scoped in writing before anything runs on your systems: shadow-mode recommendations on operator data compared against what the operator actually did. No pilot has been requested or agreed, and none is demonstrated.Request a controlled pilot
- 09 · Execution against your sandboxRequires an authorised sandboxThe dry-run plan names every connector call an approved decision would make across CRM, campaign management, billing; nothing is called and no receipt exists, because no authorised sandbox is registered.
- 10 · Live operationNot implementedUnder a separate production authorisation, with outcomes reconciled against your own records: response against a held-out group; contacts avoided against the send log. Not built; no connector exists.
- Public now
- Anyone can do this from this page, without signing in, and the page keeps no record of it.
- Available after sign-in
- A verified sign-in sees it on sample data; it reads nothing private and stores nothing.
- Entitled workspace
- A member of an organisation that holds the owning product and has the journey configured; the role decides which actions.
- Pilot request available
- The request form is available. A pilot is scoped in writing afterwards; the pilot itself is not demonstrated.
- Requires an authorised sandbox
- Needs an authorised operator sandbox, which no organisation has registered. The permission that would reach one is refused to everyone.
- Requires production authorisation
- Needs a production authorisation from the operator and a connected operator system. No such authorisation exists.
- Not implemented
- No code performs this.
The contract, the data, the pilot and the products beneath.
Four things a technical or commercial evaluation asks for, each in the same terms as the other nine offers.
The offer in twelve lines
The offer in the same terms as the other nine.
Read from the offer contract and the engine's own run; the same twelve fields on every offer page.
- Commercial offer
- Next-best-action and Retention Decisioning
- Buyer
- CVM and Retention lead
- Problem
- Several campaigns want the same customer in the same week; contact fatigue and margin loss follow.
- Outcome a pilot would count
- Response against a held-out group, and contacts avoided.
- Current availability
- Demonstration available — level 2 of 5: A journey runs end to end on sample data, including its refusal and failure states. Of 10 steps, 3 are public now, 3 need an entitled workspace, 2 need what does not exist yet. Public now Entitled workspace Not implemented
- Public sample
- HeuriTel default synthetic sample: Kenya, in KES and Kiswahili (sample next-best-action-ranking-sample v1). Download the workbook. Synthetic rows invented by HeuriTel, sized for Kenya in KES. Not an operator's figures; nobody here is real. Configure a market to see the same sample recomputed for it.
- Data requirements
- Customers (subscriber_ref, tenure_months, monthly_spend, days_since_recharge, risk, contacts_7d, consent); Actions (action_id, name, cost, response_rate_pct, value_if_taken, minimum_risk, minimum_spend); Policies (policy_id, field, operator, value, reason). customers, actions and policies as three sheets: every action has a cost and an expected value per eligible customer; every policy gate names a field and a rule; the no-action alternative is always present. A subject column holds your own pseudonymous reference; a phone number, IMSI, IMEI or account number in it is refused and never echoed back.
- Executable result
- Engine ranked-decision v1, on the public sample: 5 of 8 customers get a permitted action worth 359 KES in expected value; 3 get no action; 20 action-customer pairs refused by a gate or ineligible. 6 measures with their basis, 3 sections, 0 exceptions with owners. Recommendation: Approve the 5 ranked actions (2 × Loyalty call with tailored plan, 1 × Bonus data on recharge, 2 × Recharge reminder) and leave 3 customers alone this week. If nobody acts: 8 customers receive whatever each campaign sends them, 4 of which the contact cap would have refused. Simulated; nothing ran against an operator system.
- Approval model
- Maker-checker: the person who accepted the data or requested the run cannot decide its recommendation; a client administrator does. Approval applies to a simulated recommendation.
- Evidence
- ranked permitted action per customer with the reason trace; the refused actions and which gate refused them; the explicit no-action alternative; run register export; the retained original with its inspection verdict; the run history; the per-record dry-run plan; the pilot package.
- Pilot requirements
- shadow-mode recommendations on operator data compared against what the operator actually did. None is evidenced; no pilot is requested or agreed.
- Live requirements
- Connected under a separate authorisation: CRM, campaign management, billing — none connected; reconciled before any outcome is claimed: response against a held-out group; contacts avoided against the send log. Not built.
Beneath the offer: Decisioning and Next Best Action, Customer Intelligence, Journey Activation, Experimentation and Value, Offer, Pack and Pricing Intelligence, Customer Growth & CVM.
Your data — 3 sheets, checked before anything is read
3 sheets, checked before anything is read from them.
customers, actions and policies as three sheets: every action has a cost and an expected value per eligible customer; every policy gate names a field and a rule; the no-action alternative is always present. A subject column holds your own pseudonymous reference; a phone number, IMSI, IMEI or account number in it is refused and never echoed back.
- Customersthe customers a decision is made for, with the signals the gates and the values read. Columns: subscriber_ref, tenure_months, monthly_spend, days_since_recharge, risk, contacts_7d, consent.
- Actionswhat could be offered: what it costs, how often it is taken, what it is worth, and who is eligible. Columns: action_id, name, cost, response_rate_pct, value_if_taken, minimum_risk, minimum_spend.
- Policiesthe gates every action passes: a field, an operator and a value. Columns: policy_id, field, operator, value, reason.
A pilot’s measure and prerequisites — 4 things that would have to be true
Response against a held-out group, and contacts avoided.
Nothing is counted until it is reconciled against your own records: response against a held-out group; contacts avoided against the send log.
- shadow-mode recommendations on operator data compared against what the operator actually did
- Connected under authorisation: CRM. Not connected today.
- Connected under authorisation: campaign management. Not connected today.
- Connected under authorisation: billing. Not connected today.
Built on 5 governed products and 3 catalogue streams
5 governed products and 3 catalogue streams.
The offer is a commercial view onto the same platform every other offer runs on. Its workbench opens with Decisioning and Next Best Action.
- Customer IntelligenceCanonical product 1.
- Decisioning and Next Best ActionCanonical product 2 — the entitlement that opens the workbench.
- Journey ActivationCanonical product 3.
- Experimentation and ValueCanonical product 4.
- Offer, Pack and Pricing IntelligenceCanonical product 5.
- Customer Growth & CVMCatalogue stream.
- Campaigns & PersonalisationCatalogue stream.
- Loyalty & RewardsCatalogue stream.
