The short version
AI tariff loading is the process of turning unstructured carrier rate documents — PDFs, scanned tariffs, Excel rate sheets, emailed quote attachments — into structured records your business can quote against, audit, and update without hand-keying.
It is not magic, and it is not a replacement for your operations team. It is the layer that finally makes the tariff data your team already has usable the way the rest of your business expects data to be usable in 2026.
Why this is now possible
Three things changed in the last few years:
- Document AI got good enough for the messy, table-heavy reality of carrier tariffs.
- Forwarders standardized on cloud spreadsheets and shared drives, so the source documents are reachable.
- Quoting expectations collapsed from days to minutes. There is no longer time to look up rates manually.
Put together, AI tariff loading went from "interesting demo" to "competitive table stakes" inside one sales cycle.
What a good AI tariff loader actually does
A serious tariff loading product should do these four things:
1. Parse the document
Native PDFs, scanned PDFs, Excel/CSV, emailed quote attachments. Major ocean carrier templates plus messy NVOCC/forwarder rate cards.
2. Extract structured fields
At minimum: origin/destination (port pairs or zones), mode, equipment type, validity window, base rate, currency, surcharges (BAF, CAF, peak, GRI, FAF), free time, transit time. Each field tied back to the source clause it came from.
3. Surface ambiguity for review
The loader should NOT pretend it knows everything. Where a value is ambiguous, conflicting, or low-confidence, it asks a human. The output of the loader is "ready for ops review", not "shipped to production".
4. Make the result queryable
The structured record needs to land in a rate library that the rest of the business — sales, finance, the customer portal — can actually query. A clean record locked inside an internal model is useless.
What it does not do
- It does not replace carrier relationships.
- It does not invent rates that are not in the document.
- It does not autonomously commit pricing without human review.
- It does not solve customer-specific markups (that is a rules problem, not an extraction problem).
If a vendor pitches you AI tariff loading and skips over those four "does not", be careful.
How Freightools.ai approaches this
Inside Freightools.ai, the product is called Tari. Tari ingests the document, extracts the structured fields, links every value to the source clause, and surfaces ambiguous fields for ops review in a side-by-side viewer. Once approved, the agreement is live in Miles for instant quoting and in your customer portal.
The reviewer can confirm a clean tariff in under ten minutes. Complex bundles take longer — appropriately, because that is operations work, not data entry.
The honest payoff
Forwarders who load tariffs with AI typically see two things:
- The time-to-quote for repetitive RFQs collapses from hours to seconds.
- The margin leakage from missed surcharges drops, because the system surfaces them every time.
You do not need 10x productivity claims to justify this. You just need to look at how much time your operations team currently spends as the bottleneck between sales and a number on a customer's screen.
Where teams get the rollout wrong
The most common failure in adopting AI tariff loading is not the technology — it is treating it as a one-time import project. Carrier tariffs are not static; they change every GRI cycle, every peak season, every contract renewal. A team that loads everything once and walks away ends up back where it started within a quarter, except now the stale data looks authoritative because it is in a system.
The teams that get value treat loading as a continuous flow: new tariffs and updates land, get extracted, get reviewed, and go live on a rolling basis. The reviewer queue becomes a small daily habit, not a heroic annual migration.
How to judge the output quality
Not all extraction is equal, and the difference only shows up under pressure. Before you trust a loader, pull five of your messiest real tariffs — the scanned NVOCC rate card, the email with rates in the body, the spreadsheet with merged cells — and check three things:
- Field completeness. Did it capture validity windows and every surcharge, or just the base rate?
- Source traceability. Can you click any value and see the exact clause it came from?
- Honest uncertainty. Does it flag the fields it was unsure about, or silently guess?
A loader that scores well on clean carrier templates but falls apart on your real document mix will not survive contact with a live sales floor. The messy cases are the test.
Where AI tariff loading fits in the wider workflow
It helps to see tariff loading not as a feature but as the first stage of a pipeline. The document arrives, gets parsed and structured, gets reviewed by a human, and then becomes a live rate the rest of the business can act on. Each stage hands off cleanly to the next, and the value compounds the further down the pipeline you go.
Loading on its own saves the data-entry hours. But the real payoff appears downstream: because the rate is now structured, the quoting engine can return an instant number, the customer portal can show a self-service price, and finance can validate the invoice against the agreed buying cost. A rate that started life as a scanned PDF ends up powering four different workflows — and it only had to be read once.
This is also why loading quality matters so much. An error introduced at the extraction stage does not stay contained; it propagates into every quote and every invoice check downstream. A missed surcharge in the tariff becomes a missed surcharge in a hundred quotes. That is the case for the human-review step: it is cheaper to catch an extraction error once, at the source, than to chase its consequences across the business for a quarter.
Seen this way, the question "should we use AI to load tariffs?" is really "do we want our rate data to be structured at the point of entry, or do we want to keep re-deriving it by hand at every stage?" Once the data is structured early, everything built on top of it gets faster and more reliable. Kept unstructured, every downstream team pays the re-keying tax again and again.
The forwarders who get the most from this do not treat loading as an IT project with an end date. They treat it as the front door to their rate data — the place where messy supplier documents become clean, queryable records that the whole business can finally trust.
Want to see it on your own tariffs?
Book a demo and bring a tariff PDF you actually use. We will load it live on the call.