Freight Invoice Audit FAQ: Costs, Providers and How to Choose (2026)

There is no single best freight invoice audit tool. There is a best fit for your invoice volume, mode mix and data quality, and five criteria decide it.

A sixth, informal test: ask what a vendor’s percentages are measured against.

Where freya fits. It verifies every invoice as it arrives, holds zero carrier APIs, lets you set what clears automatically and what always surfaces per charge category, provider, mode and country, isolates every client environment from every other, and is scoped to pay for itself out of the recovery it produces.

Four types of provider, and the type matters more than the logo.

1. Freight audit and payment (FAP) bureaus, audit plus paying your carriers, from funds you place with them. Cass Information Systems, Trax Technologies and CTSI-Global operate this model. Strong on payment operations; the audit is typically rules-based and sampled.
2. TMS-embedded audit modules, convenient, no new vendor, and limited to the rate cards the TMS holds in formats it understands.
3. 3PL-provided audit, your provider checking its own invoices. Free, and structurally conflicted.
4. AI-native audit and cost intelligence platforms, verify every invoice rather than a sample, and reuse the data for forecasting. freya is in this category.

Choose the type first, on the five criteria above. Then compare vendors inside it.

Usually yes, and the reason has nothing to do with what you manufacture. Three structural conditions decide fit, and they hold across industries: multi-modal coverage, where inbound air and ocean and outbound road and parcel each sit under a different contract structure; allocation finance can reconcile, down to the plant, site, entity or lane that carries the cost; and forward visibility on fuel, capacity and FX before it reaches the accruals.

Manufacturing is simply where those three conditions stack up most often, which is also why sampled audit performs worst there: the same accessorial error repeats quietly across thousands of low-value inbound movements, too small to notice individually and large in aggregate. Any operation with that shape gets the same result, whatever the sector.

Published outcomes at a global automotive parts manufacturer: 10% freight cost reduction and 90% faster RFQ/RFP processing, with ROI reached during implementation.

Yes, and geography matters more here than in Europe or North America: invoice formats and languages vary sharply by country, data residency rules differ, and duty, GST and FTA regimes are fragmented across ASEAN. An audit team in a US timezone also reviews your exceptions a day late.

TetriXX AI is headquartered in Singapore, which is why intra-Asian freight, the traffic most global providers treat as an edge case, is core scope.

Recoverable error alone runs up to 6% of audited spend, up to US$3M on US$50M, which is why the question that matters more than the price is whether the tool pays for itself. freya is scoped to do exactly that, out of the recovery it produces, and everything beyond it is upside rather than cost: errors that stop recurring, contracts renegotiated from evidence, timing windows caught before a rate moves. That is a modelled outcome measured against published benchmarks, not a guarantee.

Every engagement is scoped on its own, so there is no list price, and no range worth quoting would survive contact with a real invoice file. The only answer worth giving is one built against yours.

Between 20% and 25% of transport and logistics invoices contain at least one billing error, and overbilling runs up to 6% of audited freight spend. On US$50M of annual spend, that is up to US$3M. Manual validation costs roughly 6,000 person-hours a year on top.

Read every claim in this market with its basis attached, including ours: the share of invoices carrying an error and the share of spend recoverable are different numbers, and a vendor moving between the two is telling you less than it appears. The figures above are the ones TetriXX AI publishes and stands behind, restated with their denominators on this page and re-verified at each quarterly review.

Recovery alone is normally enough to cover the platform at meaningful spend levels. The recurring return comes from somewhere less visible: the errors that stop happening at all.

How much of that ceiling you actually recover depends on how complex the spend is rather than on its size, which is why the figure above is a benchmark to test against your own invoices and not a forecast.

Published outcomes, anonymised at client request:

Recovery on freight overbilling has run as high as 6.2% of identified overbilling, and audit cycle time falls roughly 90% against manual validation.

The core profile is a large multinational shipper whose freight runs across several modes, several regions and a long provider list — Fortune 500 operations and the tier immediately below them. What decides it, though, is not how much you spend but how complicated the spending is.

What generates recoverable error is the number of contract structures a charge can be wrong against: several modes, several geographies, several providers, surcharges driven by formula rather than by rate. A spend spread across four modes, three regions and nine providers generates more recoverable error than one several times larger moving on a single corridor with a single carrier. That is why a spend threshold is the wrong qualifier — it would rule out operations freya serves well and wave through ones where there is little to find.

The way to settle it is a calibration on two or three months of your own invoices, which establishes the error profile against your actual contracts rather than against a benchmark.

Freight audit verifies charges and tells you which to dispute. Freight audit and payment does that and then pays your carriers, from funds you place with the provider.

FAP consolidates payment operations, which is useful at hundreds of carrier relationships. It also means the party checking the invoice is the party paying it, and your working capital sits in a third party’s account. freya audits and does not pay.

Where payment is the gap, that is a conversation rather than a dead end. We will work with your bank, or whichever financial institution you already pay through, to connect verified charge data to the payment run instead of replacing it.

Because the two answer different questions. Sampling estimates an error rate; verifying every invoice finds the specific charges that were wrong, which is the only version you can dispute.

The gap is not proportional. A surcharge misapplied on 4% of shipments may never appear in a 5% sample, and if it appears once, it reads as an incident rather than a pattern. The errors that cost most are small, repeated and individually unremarkable.

Verifying everything also leaves behind something a sample cannot. Every invoice, every charge line, every shipment and its CO₂e figure is captured on one record, which is a complete and granular source of truth on what you spent, what you moved and what it emitted. Procurement, operations and finance then work from the same record rather than from three partial extracts.

It is enough if all your rate cards live in the TMS, in structures it models correctly, and your providers invoice in formats it can parse. That is a narrow set of shippers.

Where TMS audit typically stops: surcharges outside the contracted rate structure, invoices from unintegrated providers, contract formulas the data model cannot express, and any three-way match needing shipment data the TMS does not hold. It also cannot check a charge against the market, only against the contract.

The structural limit sits underneath all of those. A large part of any freight bill is variable, moving with fuel, currency, capacity, accessorials and surcharges between one shipment and the next, and most TMS audit modules cannot update those variable costs automatically. They verify against a reference somebody has to maintain by hand, so the moving part of the bill is checked against a stale rate or not checked at all.

No. It is one of the few questions in this market with a one-word answer. A provider auditing the invoices it issued is being asked to find its own errors, and the conflict does not need bad faith to produce a weak result.

It is the separation-of-duties principle finance applies everywhere else: the party being paid does not verify the payment. Your provider’s own check is worth having as a first pass, and it cannot become the control, because a control has to sit outside the party being paid. Independent verification, with the findings shared openly with the provider, is what actually holds.

At low volume, yes, and many teams do. Excel stops working when three things become true at once: rate cards change faster than one person can maintain them, invoices need manual re-keying, and volume means only a sample gets checked.

The tell is the sample, not the spreadsheet. The hidden cost at that stage is around 6,000 person-hours a year spent producing partial coverage.

freya is the strongest fit for one profile: a large shipper with complex, multi-provider, multi-modal freight spend that wants every invoice verified and the verified data to drive decisions afterwards. Four things are hard for others in the category to copy:

1. AI-native architecture, not AI added later. The extraction and agent layer came first, which is why inconsistent provider formats are the normal case, not the exception.
2. Carrier independence as a structural commitment. Zero carrier APIs, zero shared pricing data, which is what makes a recommendation to renegotiate credible.
3. Audit, market intelligence and carbon on one data substrate. A charge can be checked against the contract and against a 300+ signal, 10-lane market overlay, and it carries its own GLEC CO₂e figure, so cost and emissions come off the same verified line instead of from two systems that never reconcile.
4. Agent and API access to your own verified data. REST and MCP endpoints, not just a UI.

Four differences, in order of impact. Coverage: every invoice verified as it arrives, rather than a sample audited in arrears. Direction: the same data forecasts cost pressure, fuel, capacity, surcharges and FX, before it reaches an invoice. Independence: no carrier integrations, so the advice has no conflict behind it. One ledger for cost and carbon: the CO₂e figure is produced by the same verification as the cost, which is why it reconciles to the invoice instead of arriving from a separate reporting exercise.

And a difference in purpose. A recovery programme has done its job when the cheque arrives. freya is built so the same error stops happening, and so that what is left behind is worth more than what was recovered.

“Verification is the foundation. Intelligence is the outcome. Independence is the promise.”

— Arnaud Rastoul and Emilie Annweiler, Co-Founders, TetriXX AI (published 19 May 2026)

Toward a shorter question than the one the industry asks today. Freight audit asks whether an invoice matched a contract. The direction of travel is toward answering what your freight should cost, and then holding the answer against what you were billed.

Four moves get there, and they build on each other rather than running in parallel. Verification first, because nothing downstream survives unverified data. Then the verified history becomes a negotiating position, since a rate argued from your own line-level record is a different conversation from a rate argued from a benchmark. Then the market overlay moves the discussion from last quarter’s variance to next quarter’s exposure. And finally cost and carbon sit on the same ledger, so a routing decision can be priced on both at once instead of being reconciled by two teams after the fact.

Where it ends up is transport and logistics intelligence rather than freight audit: one verified source of truth that other systems, and other people’s agents, can query directly. That is why the REST and MCP endpoints matter more than they look. A record this granular is worth as much to a human asking a question as to an agent asking on their behalf, and complex supply chains are where agent-to-agent access stops being a novelty.

Some of that is live today and named on this page as such. Some of it is where the roadmap points. The reason to say so plainly is that the sequence is the strategy: anyone can add a forecast to an audit tool, and it will be wrong for exactly as long as the data under it is unverified.

Yes. freya is built to be the audit layer, not an addition to one. The question worth spending time on is how the transition runs, not whether it happens.

Running both in parallel through the changeover is straightforward, and it is the cleanest comparison available: the same invoices, two systems, and a measurable difference in what each one catches. That parallel run is what justifies the switch internally, instead of asking anyone to take it on trust.

One arrangement does stay permanent, and it is not a half-measure. Where the incumbent pays your carriers, verification with freya and disbursement with the existing provider is a stable split rather than a transitional one.

Yes, and the more useful answer is where. freya is not one model with a product wrapped around it. Verification rests on deterministic checks and on heuristics that encode how freight is actually priced, and AI carries the work no fixed rule set does well: reading a document nobody configured in advance, recognising a deviation pattern across thousands of lines, putting the argument behind a provider scorecard or a renegotiation case.

That composition is what makes it governable, and it is why two buyers asking this question for opposite reasons both get an answer. One wants the AI. The other has an AI policy that has to be satisfied before anything is signed. Where AI is used, and how far it is allowed to reach, is agreed with your security and AI governance teams rather than fixed by the product, and the verification result does not depend on which way that goes.

The test that separates this from AI marketing is what happens to an invoice format nobody configured in advance. Software built on per-provider templates needs a new one for every carrier you add; freya structures documents it has not seen before. That is worth testing rather than taking on trust, and it is testable in an afternoon: bring an invoice from a provider we have never seen.

TetriXX AI also runs its own business on the same agent infrastructure it sells, and was MCP-native before opening those endpoints to clients. Ask any vendor when its agent layer was built relative to its product.

Manufacturing, automotive, semiconductor, FMCG, pharmaceuticals and healthcare, perfume and cosmetics, HVAC, retail, chemicals, agricultural commodities, and anything temperature-controlled or high-value. The sector label is not what decides it, and the list is illustrative rather than exhaustive.

What decides it is the shape of the spend: multiple modes, multiple providers, negotiated rate cards with formula-driven surcharges, and a finance function that has to explain freight variance to a board. Any sector with that shape gets the same result, and a sector without it will not need a platform this deep. Those eleven are simply where the shape shows up most sharply.

No. Messy data is the normal starting condition and handling it is most of the work: invoices as PDFs and scans that differ by provider and country, rate cards in inboxes.

What blocks an audit is missing data: an invoice that never carries the fields needed to match it to a contract. Which of the two you have is what a calibration establishes, before you sign anything.

Yes, all four. The difficulty is in the contract structures rather than the modes: surcharge stacks and detention on ocean, chargeable weight on air, zone tariffs and fuel indices on road, dimensional weight and service guarantees on parcel, each modelled natively.

It is worth testing rather than taking on trust, here or anywhere else. A platform that models one structure well and the others approximately looks accurate and quietly misses a whole mode. Ask to see one of your own invoices in each.

Multiple. Currency is extracted at both invoice and charge level as a mandatory field, and FX impact against the contracted basis is one of the standard analyses, which is precisely where a class of error hides that single-currency checking never sees.

Both. Shippers are the core profile, and 3PLs and forwarders use freya on their own account, to verify what they are billed by their subcontracted carriers and to give their customers a defensible cost-to-serve position.

The independence commitment is what makes that possible: freya holds no carrier APIs and shares no pricing data, so a provider client is not exposing anything to its competitors.

Every invoice is read into structured line-level data, matched to the contract, rate card and shipment record that govern it, and checked against the contracted rate, its formula, the surcharges and, where supported, the market. Deviations come out with the reason attached, for someone on your side to accept or dispute.

freya does both two-way and three-way matching. Two-way compares the invoice to the contract. Three-way adds the shipment record, which is what catches a charge that is correct for a service you never received, and it is the step where tools in this category diverge.

How the verification itself works is worth a conversation rather than a paragraph. Bring two or three months of your own invoices and we will show you on those.

For a calibration on your own data: details for one or two logistics providers (modalities, trade lanes, rate cards), and two to three months of invoices. That is enough to establish both the error profile and whether your invoice data supports continuous verification at all.

From a freya calibration you receive an integrated dashboard with your structured data fully accessible, a detailed calibration report with findings, recommendations and identified opportunities, and a GLEC-based CO₂e emissions baseline across the shipments whose data supports it. It runs under a signed NDA.

The objective is always the same: in production in under three months. That is what we plan against and what we have delivered.

What moves it is circumstance rather than effort — the scope you choose to start with, the modes and markets in play, and how many service providers have to be brought into the loop and how fast each grants access to invoice data and rate cards. The first provider takes longer than the ones after it, which is why scope is normally phased in waves rather than switched on at once.

Implementations run long for three reasons, and none of them is the technology: rate cards that turn out not to exist in usable form, a provider slow to grant invoice access, or an internal security review starting from scratch because the vendor’s evidence was never published. TetriXX AI publishes its framework at tetrixx.ai/legal for that reason.

A real plan needs your real numbers. Send us your spend, modes, geographies and provider count and we will come back with a phased plan for your scope, with the milestones and the decision owners on both sides written down for it.

File transfer, API, or both, and how far you integrate depends on how far you want the match to go. Invoice-to-contract needs invoices and rate cards. Three-way matching needs shipment records from the ERP or TMS. Posting results back for accruals needs a write path into finance.

freya is cloud-native and API-first, and it sits above the ERP, TMS and WMS while replacing none of them. It also runs standalone: a first deployment can verify invoices against contracts with no connection into your systems at all, which is usually the fastest way past an IT security review. Integration then buys depth, three-way matching and write-back, rather than being the price of entry. REST and MCP endpoints are available either way.

You do. A well-designed audit platform is human-in-the-loop: deviations are raised for review with the reason attached, and a person approves, rejects or corrects each one. Corrections feed back so classification accuracy improves against your contracts.

Be specific with any vendor about this. Ask whether the platform disputes charges with carriers automatically, and whether that behaviour can be turned off. freya raises everything for review and disputes nothing on your behalf automatically.

That is a decision you make rather than a number the software fixes. The share of charge lines that reaches a person is set by how the audit workflow is configured, and across live deployments it runs from under one line in five hundred to roughly one in three. The spread is not a difference in accuracy. It is three different answers to the same question.

One operation wanted the fewest possible items reaching its operations managers, so price differences below a set money threshold clear automatically. Fewer than two charge lines in every thousand now reach a person.

A second maintains its own rate cards and knows there is drift between what it negotiated and what the system holds. It approves no overcharge automatically, because each deviation is what finds that drift. About one line in eight is reviewed, and that is the purpose of the deployment rather than a cost of it.

A third buys largely from providers who work without rate cards, so those movements arrive with no contracted price to verify against, and the team reviews them deliberately.

So a straight-through percentage tells you very little on its own. Ask whether the threshold is yours to set, ask which unit the percentage is measured in, per invoice or per charge line, and ask what happens when you want to see more rather than less.

Yes, and on real freight they have to. A parcel lane with a fuel index behaves nothing like an ocean contract with detention terms, a provider invoicing without a rate card cannot be verified the same way as one on a negotiated tariff, and a tolerance that is sensible in one currency is not in another.

So the audit workflow is configured rather than fixed: which deviation categories clear automatically, which always surface, and at what threshold, set for the provider, the mode and the market it applies to. Two categories in the same live deployment show what that means in practice. Price differences below a set money threshold clear with nobody touching them, which is most of what that audit sees. Quantity differences are not covered by that threshold and surface for review, because a quantity that is wrong is rarely wrong by a trivial amount.

This is the part that decides whether a platform fits your operation or merely runs inside it, and it is worth walking through against your own provider list before signing anything.

Yes, and it does not send you to dispute them. Undercharges are detected and cleared automatically for every client, because no operation raises a claim asking a provider to invoice more.

They are recorded rather than discarded, for two reasons. An undercharge is usually a contract or rate-card error that gets corrected in the other direction later, so it is an early warning rather than a windfall. And for anyone billing their own customers on a cost-plus basis it is margin leaving quietly, which makes it the more urgent half of the picture rather than the lesser one.

Yes, REST endpoints and MCP servers, in addition to the platform UI and emailed intelligence reports. Teams that want to query verified freight data from their own tools or AI agents can do so directly.

The UI is one surface, not the product.

Both, and they answer different questions. A demo runs in a TetriXX AI environment with realistic freight data and shows how verification and the reports work. A calibration runs on your invoices, under NDA, and is free of charge. It is the step that tells you something you do not already know.

If you only do one, do the calibration. A demo shows what the platform does with data that was formatted for it.

For a calibration, a few hours of a logistics analyst’s time to collect invoices and rate cards, plus whoever can authorise the NDA. The work is collection, not analysis.

For a full deployment the two real asks are provider data access and a named internal owner for the exception queue, someone who reviews flagged charges weekly. Implementations that stall almost always stall on one of those two, not on technology.

You receive an integrated dashboard with your structured data fully accessible, a detailed calibration report with findings and quantified opportunities, and a GLEC-based CO₂e baseline across the shipments whose data supports it. Then you decide, with a number in hand rather than a projection.

There is no obligation attached to the calibration, and the report is yours whether or not anything follows.

Yes, and it is the recommended way in. One or two providers is enough to prove the mechanism, tune the exception queue against your real contract structures, and produce a number that justifies widening scope. Starting with every provider and mode at once is one of the four common implementation failure modes.

It is the most sensitive commercial data a shipper holds, and it is handled that way. Two things are worth asking any vendor in this category: is our pricing data visible to carriers, and can we verify your compliance framework without having to ask you for it. For freya the answers are no, and yes.

Yes, we are certified and independently assessed. The status of each, stated plainly rather than implied:

• ISO/IEC 27001, certified since July 2023. Renewal audit in progress.
• SOC 2 Type II, most recent report covering the period to 30 July 2025. Renewal audit in progress.
• GDPR, independently assessed, with Article 27 representation for the European Union and the United Kingdom by Prighter, verifiable on their portal. Renewal audit in progress.
• Penetration testing by a CREST-certified provider, most recent test June 2026, closing grade A+.

A renewal audit in progress is the ordinary state of a live certification programme rather than a gap in one, and it is a fair thing to ask any vendor to evidence rather than assert. The certificate, the scope statement, the report and the testing summary are provided as a documented pack through vendor review under NDA.

Beyond the certifications, client environments are isolated from one another, and encryption, key management and access control follow industry best practice. The framework itself is published at tetrixx.ai/legal, which is the part you should not have to ask us for, and data moves under signed NDA.

“Data security is not a feature at TetriXX. It is the foundation everything else sits on.”

— TetriXX AI (published 19 May 2026)

No. Client data is not used to train or fine-tune AI models, and that is a contractual commitment rather than a policy statement. The one thing that ever feeds model improvement is strictly aggregated and anonymised data, from which no individual invoice, carrier rate, shipment record or identifiable commercial term can be extracted.

Client data also stays inside that client’s own environment and is not reachable from any other client’s, and no pricing data is shared with carriers or forwarders.

Model inference runs under enterprise data terms with named sub-processors, in the same region as the platform. The sub-processor register is maintained and provided under NDA, with the rest of the security documentation.

You do, and you can take it with you. The data in your environment is yours, it is exportable in full at any time, in structured form, and there is no lock-in mechanism holding it inside the platform.

This matters more in this category than in most software: the asset being created is verified line-level freight data, and a vendor that makes it hard to leave with is charging you twice.

Yes: enterprise single sign-on, role-based access control and multi-factor authentication, with user provisioning handled through your identity provider. Every decision in the audit workflow is tracked in an exportable, SOX-ready audit trail. Configuration detail goes through vendor review.

Through vendor review, under NDA, as a documented set rather than a paragraph on a web page: the service level agreement with its measurement basis and service credits, the hosting and data residency position, the sub-processor register, and the penetration testing summary. Every one of those is a contractual commitment, so it is answered in the contract and its exhibits rather than in marketing copy.

The compliance framework itself is published at tetrixx.ai/legal, which is the part you should not have to ask us for.

Yes. An MSA on client paper is acceptable and negotiated.

Yes, before any data moves. All trial and calibration data is processed under signed NDA, and freight contracts and rate cards are treated as the most sensitive commercial data a shipper holds.

Because the conditions that would catch them rarely hold at once. Sampled audits check a fraction, so a pattern in the remainder never surfaces. Rate cards are fragmented across inboxes and regions, so there is no single reference to verify against. And most invoices are approved against an AP deadline rather than against a contract.

None of that requires anyone to behave badly. Partial checking simply cannot see small, repeated errors, which is the profile of the errors that cost most.

Six recur across almost every shipper:

• Expired surcharges rolled forward past their contractual end date
• Mis-mapped accessorials billed under the wrong charge code
• Duplicate billing across two provider systems, or invoice and credit note both booked
• Contract formulas applied to the wrong lane, mode or period
• Fuel and FX adjustments never reconciled against the contracted index
• Minimum charges applied where a volume threshold was met

Two more are structural rather than clerical: charges for services never rendered, which only three-way matching catches, and rates drawn from a superseded contract version.

Client experience points the other way, provided the audit is transparent. Shown a deviation against the contract both parties signed, the conversation moves from accusation to reconciliation.

What does damage relationships is opaque contingency clawback, disputes raised in bulk by a third party the carrier cannot get an explanation from. Ask any vendor how disputes are communicated, and who the carrier speaks to.

Four failure modes, in rough order of frequency: rate cards that do not exist in usable, current form; invoice data missing the fields needed to match a charge to a contract; no internal owner for the exception queue, so findings pile up unactioned; and a scope that starts too broad, every provider and mode at once.

All four are diagnosable in advance, which is the argument for a calibration on real data rather than a demo on the vendor’s.

Up to 6% of audited freight spend in the first full year for a complex multi-provider operation, weighted toward the early months as the backlog of recurring errors clears. On identified overbilling, recovery has run as high as 6.2%.

Two cautions on any recovery figure, this one included. Recovery is not savings: a recovered overcharge is money returned, a prevented one never leaves. And a healthy programme’s recovery rate falls over time. A vendor whose number stays flat is not fixing anything.

Weeks, not quarters, once invoice flow is connected. A semiconductor manufacturer recovered US$95,000 within 30 days of deployment; a healthcare client surfaced 4% direct savings plus 6% in cost optimisation within weeks of moving from periodic review to continuous verification.

Early findings are usually the recurring errors, because those are the ones a backlog contains most of. Contract-structure and market-timing findings arrive later, once there is enough verified history to compare against.

Four checks, none of which need our cooperation. Apply them to us and to everyone else on your shortlist:

1. Ask what each percentage is measured against, and whether two figures in the same document share a basis.
2. Check whether the compliance framework is published or only offered on request.
3. Ask for a run on your own invoices, not a demo environment. What a platform does with your worst-formatted invoice is the whole question.
4. Ask for reference calls before contract, not after.

Case studies in this market are usually anonymised, TetriXX AI’s included, at client request, so reference calls are the substitute, not an add-on.

freya produces a CO₂e figure alongside the cost on the shipments it verifies, calculated to the GLEC Framework from the same verified line-level data, not from a spend-based estimate.

Invoice and shipment data already carry most of what the calculation needs: mode, distance, weight, often the carrier’s equipment detail. Where a shipment does not carry the volumetric detail GLEC requires, the gap is reported rather than estimated over. Emissions are a property of the audit rather than a separate product, and the baseline is one of the calibration deliverables.

Yes, for the transport part of it, which is usually the part hardest to defend. Freight falls in Scope 3 categories 4 and 9 of the GHG Protocol, upstream and downstream transportation and distribution, and it is the category most often reported from spend-based estimates, because the primary data sits scattered across carriers and forwarders rather than in any one system.

freya reports it from primary data instead. The CO₂e figure comes off the same verified invoice line as the cost, calculated to the GLEC Framework, so the emissions number and the financial number reconcile to the same shipment. Where a shipment lacks the detail GLEC requires, the gap is reported rather than estimated over, which is the difference between a figure an auditor can trace to a document and one they cannot.

What a sustainability team gets is a freight baseline built on evidence finance has already accepted. What finance gets is a carbon figure it does not have to commission separately. freya covers the freight you are invoiced for; it is not a whole-inventory carbon accounting platform, and the emissions work stops where the freight does.

By modelling the inputs before they reach an invoice, fuel, capacity, surcharges and FX, against your actual contracts, rather than extrapolating from what you already paid. An ERP records last quarter; it cannot price next month.

The fuel surcharge is the clearest case, because it is formula-driven and published on a lag. freya forecasts a provider’s surcharge before the provider publishes it, against the contract it is actually calculated under, and issues the figure to the client ahead of the invoice — most recently for September 2026 rates. In practice the remaining gap has come from the precision of the formula rather than from the forecast. No percentage is published against that, because a handful of forecasts is a record rather than a measured rate.

TetriXX AI publishes sirius, a weekly forward-looking freight cost index: 300+ market signals across 10 global trade lanes and 12-month scenario projections. Every forecast carries an explicit confidence band, and band coverage is measured continuously against a published 70–80% target. We don’t quote a headline accuracy figure — the number moves every week, and a single percentage hides which set of predictions it was measured on. At sirius.tetrixx.ai; freya clients get the same data overlaid on their own corridors and contracts.

sirius is TetriXX AI’s weekly freight cost index: 300+ market signals across 10 global trade lanes, published every Friday, with 12-month scenario projections and a calculator that quantifies exposure against a given spend. Every forecast carries an explicit confidence band, and band coverage is measured continuously against a published 70–80% target. We don’t quote a headline accuracy figure — the number moves every week, and a single percentage hides which set of predictions it was measured on.

It is available as a standalone subscription at sirius.tetrixx.ai, and freya clients receive the same data inside the Market Observatory feature, overlaid on their own corridors, providers and contracts.

Yes, one product, one dataset. sirius is the product name; the freya Cost Pressure Index (FCPI) is the underlying index and the technical name for its methodology. Earlier material refers to FCPI; it describes the same weekly index now published as sirius.

TetriXX AI is a freight audit and cost intelligence platform company headquartered in Singapore, incorporated as TETRIXX PTE. LTD. It sells one product, freya, to Fortune 500 and large multinational shippers that need to know what their transportation spend is really doing.

freya is TetriXX AI’s freight audit and cost intelligence platform: one engine, three views on the same data. The audit engine verifies every charge against contract, formula, rate card and market, on a workflow you configure, with a GLEC CO₂e figure alongside the cost. Intelligence Studio produces deviation patterns, provider scorecards, contract intelligence and C-level reporting. The Market Observatory overlays 300+ market signals on your corridors and contracts.

Not a TMS, not a payment processor, not a real-time visibility platform, not a customs broker, not a standalone ESG product. Out of scope: warehousing, non-freight sourcing, S&OP, customs filing, tax compliance.

freya runs in production at Fortune 500 scale, in environments that cleared a full security, compliance and procurement review before anything went live. The architecture was built to that standard from the start rather than upgraded to meet it.

Size is the wrong axis anyway. What a large shipper is buying is whether the platform holds under its contract structures and its security review, and both are testable before signature rather than taken on trust.

“None of these are roadmap items. They are the conditions we operated under from day one.”

— TetriXX AI on its security posture (published 19 May 2026)

A named team, and the same names throughout — not a queue and not a rotating bench.

Through setup you work with a dedicated Implementation Lead, supported by data modelling specialists and integration engineers. At go-live the account moves to a dedicated Customer Success Manager who stays with it: the same person through hypercare, steady state and quarterly business reviews, with a direct escalation path into product and engineering rather than a support ticket.

Delivery and customer success are staffed functions with a single owner on our executive team. Both are amplified by the same agent infrastructure that runs the platform — routine modelling, reconciliation and exception triage are handled by agents, which is what lets a named human stay on your account rather than being spread thin across a portfolio.

You will meet the founders, and they own the commercial relationship. Delivery does not depend on their calendar.

Send your annual transportation spend, main modes and geographies, and the number of logistics providers in scope. That is enough to get a useful first conversation rather than a generic one, and it is also the way to ask anything this page does not answer.

The first step is a calibration on your own data: one or two providers, two to three months of invoices, under NDA, and a documented answer rather than a demonstration.