The LNG Cost-to-Serve engine: what we have, and how to use it

We have a governed cost-to-serve engine for gas delivery. Give it a customer location and an annual volume, and it returns the landed cost to deliver gas there, in USD/GJ, across every realistic mode — pipeline, LNG trucking, and small-scale LNG by ship (ssLNG) — ranked cheapest-first, with the routing done for you. It started as the Mozambique (Afungi) corridor study and is now a reusable engine any volume/demand project can read from, so nobody re-implements assumptions or cost models in their own spreadsheet.

Read this first — what the number is. It is the cost to move gas, not a delivered gas price. It excludes the gas commodity (the molecule) and any margin. For an all-in delivered price you add your own molecule cost on top: delivered_price = molecule_price + landed_cost_to_serve. Saying this out loud is what keeps the output defensible.

What it answers

The questions a client or a volume project actually asks:

• What does it cost to serve this customer, and by which mode? A per-customer cost stack (USD/GJ) across pipeline / trucking / ssLNG, ranked, with the winning mode named. The engine routes everything itself — real road kilometres and inferred sea distance — from a lat/lon.
• FSRU or onshore terminal? A landing-cost breakdown at any port — lease, port, regas and storage for a floating FSRU, versus the onshore-terminal capex case — with regas and storage broken out as their own lines so they are never double-counted against a molecule price.
• Pipe or truck? The breakeven volume at which a dedicated pipeline overtakes trucking on a given leg (pipelines cheapen with volume; trucking is roughly flat per GJ).
• Under which world? Scenario levers (base / optimistic / conservative / stress), plus volume, year, WACC and FX — the same governed scenarios the rest of the model stable uses.
• New build or existing line? Greenfield pipelines are sized and costed on a full-capex NPV=0 DCF; existing regulated pipes (e.g. ROMPCO, Lily) are priced off their real NERSA tariff — a sunk asset never gets a greenfield capex.

Governance is built in

Every component carries its evidence status (regulated / study / assumption / placeholder), a confidence score, and a snapshot fingerprint. Placeholders and proxies are flagged in the output, not hidden — today that means the FSRU storage tariff (placeholder), the Richards Bay port charge (a Mozambique proxy), and the onshore-terminal capex (benchmarked, low-confidence). A consumer can auto-disclose these rather than discovering them by eye, and the fingerprint changes whenever a governed input re-anchors, so stale numbers are detectable. This is the same discipline as the rest of the stable: reproducible, vintaged, sourced.

Where to find it

• For clients & members — the interactive Project Kilimanjaro → Models hub: the LNG Cost-to-Serve calculator (pick a port, volume and scenario; see the FSRU-vs-terminal breakdown and the truck-vs-pipe breakeven, with the evidence badges), plus the landed-price map, the routing network map and the tariff report.
• For integrations / volume projects — a governed HTTP API. The main call is POST /v1/landed-price (customer + demand → ranked chains); plus /v1/landing-cost, /v1/breakeven, /v1/cost-to-serve/* and /v1/services. Interactive docs live at /docs on the running service.
• To hand to another team (or their AI agent) — the self-contained integration brief in the repo: docs/downstream_consumer_prompt.md. It is the contract: follow it and you cannot double-count or misread the numbers.

How to use it in a conversation

Open the calculator, pick the customer's nearest landing port, their annual volume and the scenario in play. You get the landed cost stack, the cheapest mode, and the volume where a pipeline would beat trucking. Lead with the winning mode and the headline $/GJ; use the badges to be honest about which lines are solid and which are still placeholders — a confident number with no provenance is not a finished answer. For an all-in delivered price, add the client's molecule cost and declare its basis so the engine charges only the tolls below it.

Status. Engine v0.3.1, governed assumptions snapshot, infra-asset WACC anchor (12%). The lightweight pricing surfaces (landing, breakeven, configured services) are ready now; the full lat/lon routing endpoint needs a larger host before it goes fully live. Repo: Africa-International-Advisors/lng_pipeline_shipping_model.

Internal field note — sanitised for cross-team learning. More field notes