Case Study · Logistics & Freight Brokerage

How Johanson Transportation put AI agents to work for its customers

Johanson Transportation Service (JTS), the Fresno-based freight broker and 3PL, now answers customers faster than ever: freight documents delivered in seconds, freight-spend questions answered inside the customer’s own dashboard, and quicker, more personal replies from every broker. Working with AI partner FreshBI, JTS put four production AI products built on Anthropic’s Claude to work, all inside tools its customers and team already use.

Talk to us about your business See the four products

AI products live in production on one shared, governed platform

freight document types the AI document desk delivers on its own

new apps for customers or staff to learn: AI lives in Power BI, Outlook and email

100%

of AI answers grounded in computed figures, never invented ones

1 Business first Start from the customer wait: documents, answers and replies.
2 Ontology first One business ontology and one governed backend behind every answer.
3 AI ready Four production AI products, each inside a tool people already use.

About Johanson Transportation Service

Johanson Transportation Service is a Fresno, California–based freight broker and third-party logistics provider with offices across California, Oregon, Texas, Florida and Wisconsin. JTS moves dry and temperature-controlled truckload, LTL, rail intermodal and international freight for shippers in food and agriculture (dairy, produce, seafood, meat and poultry, wine and spirits) as well as forestry, paper and non-perishable goods, and runs its own cloud TMS.

JTS has always competed on service. The faster a customer gets a document, an answer or a reply, the better the relationship, and JTS saw AI as the way to raise that bar for every customer, at every hour.

Business first

The goal: faster answers for every customer

Every freight brokerage fields a steady stream of small, time-sensitive customer requests. JTS set out to answer them faster, and to free its brokers to spend more time on freight and relationships, without asking customers or staff to change how they work.

Paperwork on demand

Shippers and consignees need bills of lading, proof of delivery, invoices and rate confirmations the moment they ask, often after hours. JTS wanted those requests answered instantly, without a customer waiting on a person to look up a load.

Freight insight, self-served

Customers want to know where their freight spend is going, by mode, lane and accessorial. JTS wanted them to get that answer themselves, in plain language, instead of waiting on a custom report.

Decisions at the speed of freight

Account performance, office and agent results, receivables health: JTS leadership wanted direct, on-demand answers rather than waiting for the next reporting cycle.

Faster, more personal replies

Brokerage runs on email. JTS wanted every carrier update, customer question and quote follow-up answered faster, still in each broker’s own voice and still reviewed by a person.

Ontology first

One ontology, one governed platform

Before any AI answered a customer, the business had to agree on what its numbers and documents mean. JTS and FreshBI built that foundation once, so every product reads from it.

The house language

The leadership assistant is grounded in a business ontology: JTS’s own definitions and metrics, so it speaks the brokerage’s language and uses the right numbers.

Numbers computed, never invented

Deterministic code calculates figures from governed data and hands Claude only those results to reason over and explain. Claude does the judgment; code owns the rules.

Governance built once

All four products run on one secured backend, so identity, auditing and monitoring were built once and every product inherits them, with no weaker version of security anywhere.

The solution

Four AI products, one governed platform

Rather than one chatbot bolted onto the side of the business, JTS and FreshBI put a capable assistant exactly where each customer interaction and decision already happens. Every product follows the same rule: Claude does the reasoning, deterministic code enforces the rules, and a human is looped in the moment judgment is genuinely required.

01 · Embedded analytics · Power BI

Freight insights for customers, inside their own dashboard

JTS customers get a Claude-powered chat embedded directly in their Power BI report. They click a pre-built scenario or simply type a question (“where could I consolidate modes?”, “what’s driving accessorials this quarter?”) and get an executive-ready answer with narrative, tables and charts rendered inline, grounded only in figures computed from their own freight data.

One-click scenario reports: cost optimization, mode consolidation, accessorial review and best-method comparisons, tuned per account.
Per-client data isolation: row-level security enforced server-side by signed tokens, so each customer’s chat can only ever see that customer’s shipments.
Scoping that behaves like a dashboard: date-range and transportation-mode selectors, with automatic prior-period and year-over-year comparisons.
Deliverables in a click: export any answer to Word, CSV or PDF.

02 · Executive insights · Internal

A private analyst for leadership

A separate, single-sign-on-gated workspace where JTS leadership asks questions of confidential financial and operational data (account performance, margin trends, office and agent results, receivables health) in a split-screen view with the answer on one side and the chart it generated pinned alongside.

Grounded in a business ontology: the assistant is taught the brokerage’s own definitions and metrics, so it speaks the house language and uses the right numbers.
Internal-only by construction: restricted to named staff; this data never touches the customer-facing surface.
From question to decision: risks, opportunities and performance outliers surfaced directly, so leadership can act while it still matters.

03 · Autonomous document desk · Email

BOL, POD and invoice requests answered in seconds, around the clock

JTS customers simply email for what they need. An AI agent monitors the shared mailbox and resolves routine freight document requests on its own: bills of lading, proof of delivery, invoices, rate confirmations, claims and customs paperwork, several in a single email if that’s what the customer asked for.

1
Verify
The sender is checked against the customer roster. Unknown senders are politely declined, so no document ever leaves to a stranger.
2
Understand
Claude classifies what was asked for and extracts the reference number, even when one email asks for several documents.
3
Retrieve
The reference is resolved against live operational records and the document is built from real data.
4
Deliver
The reply goes out with the document attached plus a secure, time-limited download link.
5
Log
Every request is written to an audit trail a manager can review live.

Graceful on misses. An unmatched reference number gets a friendly request to re-check, never a wrong document.

04 · In-workflow drafting · Outlook

Faster, personal replies in each broker’s own voice

A one-click Outlook add-in that reads the open email thread and drafts a reply in the user’s voice, inserted straight into the compose window. It runs on desktop and web Outlook through Microsoft’s sanctioned add-in framework: no browser extension, no copy-paste.

Context-aware: the draft reflects the actual thread, not a generic template.
Human always in control: nothing sends automatically; the draft lands in the compose box for review.
Zero-friction rollout: deployed centrally to chosen users; the button simply appears on the ribbon.

Under the hood: governance built once, inherited everywhere

Identity & access

Microsoft single sign-on for internal tools; signed, per-client tokens for customer-facing embeds.

Governed AI

Every answer is grounded in pre-computed results. The model reasons over the numbers; it does not invent them.

Self-service administration

A point-and-click console tunes each account’s reports, questions, display names and views without a code change.

Auditability

Document-desk activity and administrative changes are logged and reviewable.

Automated data pipeline

Datasets refresh on schedule and are verified end-to-end, so a silent gap can’t reach the tools.

Resilience

Health monitoring with automated recovery, and a friendly fallback message whenever an upstream dependency hiccups.

The results

What changed for Johanson Transportation

Customers get their documents in seconds, at any hour. The document desk verifies, retrieves, delivers and logs on its own, and JTS brokers spend their time on freight and relationships, not on finding PDFs.
Customers self-serve their own freight analysis. Mode, cost and accessorial questions are answered inside their dashboard, in seconds, scoped to their own data.
Leadership gets direct answers on performance: account, office and receivables insight on demand instead of at month end.
Replies reach customers faster. Every email starts from a context-aware draft in the broker’s own voice; staff review and send.
No new habits for anyone. Every product lives in a tool JTS and its customers already used: a dashboard, Outlook, a mailbox.

AI ready

AI where the work already happens, and the first legs of a relay

JTS’s suite is AI-ready in the way that matters: every product is a specialist that picks up one piece of the customer’s journey (the question, the document, the reply) using the same governed context.

AI where the work already happens

The fastest way to kill an AI project is to make people open a new app. Every JTS product lives inside an existing workflow, so usage starts on day one.

Reasoning with guardrails

Claude handles judgment: reading a messy thread, classifying an ambiguous request, explaining a cost trend. Deterministic code owns the rules, the data access and the final say.

Built for non-technical users

Administrators configure through a console, staff work in plain language, and sensitive actions always keep a human in the loop.

Taken end to end, that same idea is what we call the Agent Relay: one AI that carries a customer from first conversation through quote, delivery and follow-up, with nothing lost in the handoffs. It’s the agent running on our own homepage right now.

If you’re a broker, 3PL, carrier or shipper, or any business whose operation runs on email, documents and dashboards, the same platform applies. Explore our AI and BI for logistics, the Empty Leg Eliminator agent, and how U.S. logistics firms are using AI and BI.

AI for logistics: common questions

How is AI used in freight brokerage and 3PL logistics?

The highest-value uses sit where brokers already lose the most time: answering inbound document requests (BOL, POD, invoices, rate confirmations), turning shipment data into answers customers can self-serve, giving leadership direct access to margin and performance data, and drafting the constant stream of email replies. At Johanson Transportation, FreshBI delivered all four as production AI products.

Can an AI agent send bills of lading and proof of delivery automatically?

Yes. The JTS document desk monitors a shared mailbox, verifies the sender against the customer roster, identifies the document type and reference number, retrieves the record from live operational data and replies with the document and a secure, time-limited link, logging every step. If a reference doesn’t match, it asks the customer to re-check rather than guessing.

Can you add an AI chat to a Power BI dashboard?

Yes. FreshBI embedded a Claude-powered chat directly inside a Power BI report. Users type a question or click a pre-built scenario and get a narrative answer with tables and charts inline. Row-level security and signed tokens mean each customer’s chat can only ever see that customer’s data. See our Power BI dashboard services.

How do you stop an AI from making up numbers?

The model never computes or retrieves figures on its own. Deterministic code calculates the numbers from governed data and hands Claude only those results to reason over and explain. Claude does the judgment; code owns the rules, the data access and the final say. More on this in AI-Ready Data.

Why Anthropic Claude for logistics AI?

Claude is strong at the reasoning-heavy parts of logistics work (reading a messy email thread, classifying an ambiguous request, explaining a cost trend in plain language) while following tight instructions about what it may and may not do. That makes it a good fit for guardrailed, production use rather than demos.

How long does an AI integration like this take?

JTS was a multi-phase engagement: each product shipped on its own once the shared, secured backend was in place, so later products inherited identity, auditing and monitoring rather than rebuilding them. FreshBI typically delivers a first production solution in as little as three weeks. See pricing.

Want AI agents working inside your operation?

We’ll map where your team loses the most time to documents, questions and email, and show you what an AI agent would take off their plate.

Book A Call