Skip to content

Supply Chain Economic Intelligence

Know what a disruption is worth before you decide.

Most supply-chain software answers what should I order? Sepavi answers the question that actually reaches an executive: what just changed, and what is that decision worth?

Go to app See a decision
What we measure
Dollars traced
Not how many suppliers are in the database.
Every figure
Decomposable
A number nobody can take apart is a number nobody will act on.
Every claim
Sourced
Observed, declared or inferred — never silently mixed.

The shift

A recommendation is not a decision.

A planning screen presents a recommendation and lets you adjust it. A decision surface presents a consequence, the alternatives, what each one is worth, and how sure we are — then takes an approval.

Sepavi does both. The planner you can use today becomes the layer underneath: the one that knows what you actually buy, hold and owe.

Today

“What should I order?”

  • Answered per item, per location.
  • Optimises the next purchase order.
  • Silent when the world changes.

Next

“What is this decision worth?”

  • Answered in revenue and margin.
  • Ranks the alternatives, including doing nothing.
  • Starts from the event, not the calendar.

Observed · Declared · Inferred

An inferred number must never look like a fact.

Supply-chain intelligence is worth nothing if you cannot tell what it knows from what it guessed. So every fact Sepavi shows carries how it was obtained — in the shape of the thing itself, not a footnote.

observed

From your systems

Your ERP, or a source document — a bill of lading, a filing. Solid fill: this is a fact.

declared

Someone asserts it

A supplier or registry says so. Outlined: somebody is accountable for it, but nobody has verified it.

inferred · 0.78

We derived it

Hatched, and it carries a confidence band and the edges it was built from. Our reasoning, labelled as ours.

Confidence has to change what the interface does. Below a threshold a recommendation stops being presented as recommended, an edge is drawn as a gap, and an automated action stays manual. A percentage that never changes an outcome is costing trust and buying nothing.

The decision surface

Every option priced. Including doing nothing.

An event arrives. Sepavi resolves which of your facilities, parts and products it touches, prices the exposure, evaluates the mitigations you actually have, and asks a human to approve one.

sepavi · decisions · DEC-2041

Typhoon warning — Kaohsiung, TW

GDACS advisory · 14 Sep · affects 3 facilities, 11 parts, 4 products

observed

Revenue at risk

$31.7M

±$4.1M · 91% confidence

Margin at risk

$8.4M

over 6 weeks

Recommended

+$7.22M

net economic benefit

7 strategies evaluated

  • Alternate source — Supplier B, Penang

    Protects $7.7M margin · $482K incremental cost · 12-day switch · declared capacity

    +$7.22M

  • Expedite existing on-order

    Protects $4.1M margin · $890K freight · 5-day gain

    +$3.21M

  • Transfer from Central DC

    Protects $2.4M margin · $110K · creates a stockout at Central in wk 4

    +$2.29M

  • Do nothing

    Baseline — $8.4M margin exposed

    $0
Approve recommended Modify Show the evidence

“Do nothing” is a listed option with a number on it. That is what makes the others comparable rather than alarming.

Network

A visible gap is a finding.

No source reliably reconstructs every deep-tier dependency. An honest interface says where it runs out, rather than rendering an empty branch that reads as “nothing there.”

A ranked tier ladder with concentration counts, not a force-directed graph. Those demo beautifully and plan terribly — nodes drift, nothing is comparable between sessions, and the question you actually have is what am I dependent on, and how badly?

sepavi · network · M8 Hex Bolt, Zinc

Dependency path

4 tiers · 2 inferred edges · concentration flagged at tier 3

  • Tier 1

    Acme Industrial Supply

    Cleveland, OH · LEI 5493001KJTIIGC8Y1R12

    observed
  • Tier 2

    Fastener Works Ltd.

    Named on 41 bills of lading, 2024–2026

    observed
  • Tier 3

    Zhen Hua Steel — Plant 2

    Sole plausible source for this grade in region

    inferred · 0.78
  • Tier 4

    Ore origin unresolved

    No source produces this edge at usable confidence

    gap

Shipping today

The planner is live, and it is the foundation.

None of the above works without a system that already knows what you buy, hold and sell. That system is running now — a replenishment planner covering the full order cycle, multi-tenant and in production.

Suggested orders

Per supplier and stocking location, with bracket building, safety stock and a 364-day projection behind every line.

Forecasting

Statistical period and daily demand, with per-period overrides where a planner knows something the history does not.

Deals & events

Forward-buy economics and demand uplift, evaluated against carrying cost rather than guessed at.

Bills of material

Kits and dependent demand, so a component shortage surfaces against the finished goods it actually blocks.

Supplier exposure

Revenue and margin at risk if a supplier stops delivering — net of the cover you already hold, over a horizon you control.

live

Entity review

Matching your suppliers to canonical entities and their LEIs, with the evidence for each candidate and a human making the call.

live

Your own export, mapped once

Upload the file your ERP already produces — its column names, its encoding, its number format. Map it once, and every drop after that lands on its own.

live

ERP connections

Connect Dynamics 365 Business Central and it keeps itself current. Every connector says what it can and cannot fill in before you connect it — a default that looks like data is worse than a gap.

live

Exposure is net of cover

observed

A supplier going dark does not put its whole forecast at risk — it puts at risk the part you cannot serve from stock and inbound orders. Everything left of the crossing is absorbed. Only the wedge is exposure.

cover runs out Cumulative demand On hand + inbound At risk 0 2 4 6 8 10 12

Horizon, in weeks · illustrative series

Without that subtraction the figure is just “revenue through this supplier”: large, alarming, and useless for ranking. Against the demo tenant it is the difference between 7 lines exposed at two weeks and 108 at twelve.

Sequence

Twelve months, in the order the data allows.

  1. Now

    Evidence language

    Chip, confidence band and evidence drawer — against the supplier identity columns that already exist, before any commercial data contract is signed.

  2. Month 1–2

    Supplier entity page

    GLEIF, SEC and OFAC for identity, ownership and sanctions, on a screen that already exists.

  3. Month 5–6

    Network and events

    Tier-2 inference and the event graph land together — the first genuinely new screens.

  4. Month 7

    Exposure, and the dashboard switch

    Revenue and margin at risk become real numbers, and the headline metrics change with them.

  5. Month 9–10

    Decision screen

    Needs the alternative network and the scenario engine. Everything before it is scaffolding for this.

  6. Month 11–12

    Approval and outcomes

    Human-approved write-back, then predicted versus actual for every executed decision. It stays human-only until the outcome record justifies otherwise.

Start with the planner.

The intelligence layer is built on your own data, so the first step is the one that pays for itself: replenishment that gets the order right.

Go to app