By Luis Lopez, AI transportation consultant, CEO of Go Hub.io Holdings Corp and subsidiaries, and host of the Freight Guru Podcast
Most freight companies that want to use AI already have plenty of data. The trouble is that it lives in too many places and none of them agree. The TMS says a load delivered on Tuesday, the accounting system shows it invoiced on Friday with a different customer name, the tracking platform has a third set of timestamps, and the real story is in someone’s email. An AI tool connected to that environment will not fix the disagreement. It will inherit it.
That is why the most valuable AI preparation work is often not AI at all. It is building a single source of truth for freight data.
What a Single Source of Truth Means in Freight
A single source of truth does not have to mean one giant system. It means that for every important data element, there is one agreed-upon place where the official value lives, and every other system either reads from it or syncs to it. If someone asks what the delivery date was, which customer this load belongs to, or what rate was agreed, there is one answer and everyone knows where to find it.
In freight, the core records usually include:
- Customers and locations: legal names, billing entities, ship-from and ship-to facilities, and their addresses and hours.
- Carriers and assets: carrier identities, authority, insurance, equipment, trucks, trailers, and drivers.
- Loads or shipments: the central record linking customer, carrier, lane, dates, equipment, commodity, and references.
- Events: pickup, delivery, arrival, departure, and exception timestamps.
- Money: rates, accessorials, invoices, payables, and payments.
- Documents: rate confirmations, bills of lading, proofs of delivery, and invoices tied to the right load.
Why AI Exposes Data Problems
As an AI transportation consultant, the first thing I look at is not the model a company wants to buy. It is whether the company can answer simple questions consistently from its own data. If two reports produce two different on-time percentages, an AI dashboard will produce a third.
AI amplifies the data it is given in a few specific ways:
- Duplicates distort patterns. If the same customer exists under three spellings, a pricing or credit model sees three small customers instead of one large one.
- Missing timestamps break predictions. ETA and detention models need reliable arrival and departure times. Gaps or manual backfills teach the model the wrong behavior.
- Unlinked documents defeat automation. If a proof of delivery cannot be matched to its load, invoice automation stalls.
- Free text hides meaning. Critical facts typed into notes fields are hard for any system to use reliably.
For a deeper look at what AI cannot fix on its own, see what AI cannot do about freight data quality.
How to Build It, Step by Step
1. Inventory your systems and data flows
List every system that holds freight data: TMS, accounting, ELD and telematics, tracking, load boards, email, spreadsheets, and customer portals. For each, note what data it creates, what it receives, and who maintains it. The spreadsheets matter most, because they are usually where workarounds hide.
2. Decide the system of record for each data element
For every core field, assign one owner system. Customer master data might live in accounting, load data in the TMS, and event timestamps in the tracking platform with a sync into the TMS. Write it down. Arguments about which number is right become much shorter when this decision has already been made.
3. Define the fields and their rules
Create a short data dictionary: what each field means, its format, whether it is required, and who can change it. Small definitions matter. Does “delivered” mean arrival at the facility, completion of unloading, or a signed proof of delivery? AI models trained on inconsistent definitions produce inconsistent answers.
4. Clean the master data first
Start with customers, locations, and carriers. Merge duplicates, standardize names and addresses, and retire records that are no longer used. This is tedious, but it pays off in every downstream use. The fields worth prioritizing are covered in top 10 data fields to clean before buying AI.
5. Use stable identifiers everywhere
Every load, customer, carrier, and location should have one internal identifier that travels with it across systems. Customer reference numbers, PO numbers, and container numbers should be stored in defined fields, not buried in notes.
6. Automate the connections
Manual re-keying between systems is where most data drift begins. Integrations through APIs, EDI, or scheduled syncs keep systems aligned. Where a manual step is unavoidable, add validation so obvious errors are caught at entry.
7. Capture events at the source
Timestamps should come from the most reliable source available, such as telematics, geofences, or facility check-in systems, rather than a dispatcher typing them in later. When manual entry is necessary, record who entered it and when.
8. Assign data ownership
Someone must own each core data domain. Without an owner, quality decays. Data ownership does not need a new department; it needs a named person who reviews exceptions and is accountable for fixing them.
9. Measure data quality
Track a few simple indicators: the share of loads with complete timestamps, the share of invoices matched to proof of delivery, the number of duplicate customer records. Review them regularly. What gets measured gets fixed.
Signs You Are Ready for AI
You do not need perfect data to start using AI, but you do need data you trust for the specific use case. A company is usually in a good position when:
- Core reports produce the same numbers regardless of who runs them.
- Master records for customers, carriers, and locations are deduplicated and maintained.
- Load records link reliably to events, documents, and invoices.
- Key definitions are written down and used consistently.
- Someone owns data quality and reviews it on a schedule.
If several of these are missing, the warning signs in top 10 signs a trucking company is not ready for AI will probably look familiar.
Start Narrow
The single source of truth does not have to be finished before any AI project begins. A practical approach is to choose one high-value use case, such as invoice matching or ETA prediction, identify the data it depends on, and get that slice right first. Each project then improves the foundation for the next one. When you are ready to evaluate tools, ask vendors directly how they handle duplicate records, missing fields, and conflicting sources; the questions in how to evaluate an AI vendor for trucking and logistics are a useful starting point.
The Bottom Line
AI is only as reliable as the data underneath it. A single source of truth for freight data means agreed systems of record, clean master data, stable identifiers, automated connections, and clear ownership. It is not glamorous work, but it determines whether an AI investment produces trustworthy answers or just faster confusion.
For more conversations on freight and AI, subscribe to the Freight Guru Podcast.
About the author: Luis Lopez is a Miami-based AI transportation consultant and logistics entrepreneur, the CEO of Go Hub.io Holdings Corp and subsidiaries, and host of the Freight Guru Podcast.
