ADOPT AI PARTNERS
Insights / logistics

AI for Employee Onboarding in Logistics: Depot Ready Before a Six O'Clock Start

6 October 2026  ·  8 minute read  ·  Adopt AI Partners
cargo ships docked at a pier
Photo via Unsplash

It is ten to six at a depot. A new driver is standing by the transport office with a bag and a printed offer letter, and the shift planner has a run allocated to him. The handheld has not been assigned, his login to the transport system does not exist, there is no high visibility jacket in his size, and the only person who could fix all three is on the yard chasing a trailer that did not arrive.

Nobody there was careless. The work did not get finished in time because it is spread across six systems and one administrator who also has a day job.

A coordination problem wearing a compliance costume

In most industries onboarding is complete when the accounts work. In logistics it is complete when a named person can legally and safely perform a specific task on a specific site, and that fact exists as a record you can produce months later.

For a driver that means the licence is valid for the category of vehicle they will drive, endorsements have been seen and judged, the driver qualification card is in date and periodic training hours are accounted for, the tachograph card has been issued or applied for, the medical and eyesight requirements are met where they apply, right to work has been checked properly by a person, the site induction has happened and been recorded, the kit has been issued, and the logins work. For a warehouse starter the list is different but no shorter: plant certification with a real expiry date, the right induction for the right site, and access to the warehouse system and the handheld.

Every item has an owner, a system and a lead time. Onboarding here is a coordination problem wearing a compliance costume.

Peak hiring and the arithmetic of volume

What makes this distinctive is not any single check. It is doing all of them for a large intake inside a few weeks, twice a year, across sites that each run their own version of the process, with high churn underneath so the same depot is also offboarding, reissuing kit and recovering devices.

A process that works at two starters a week quietly collapses at thirty, and the collapse looks like a message in a group chat asking whether anyone has seen a licence check. An agent helps because this work is repetitive, deadline driven and spread across systems. It holds hundreds of open cases without losing one and never gets pulled onto the yard.

What an onboarding agent actually does

In practice it runs the pre start checklist per starter and per site, and it chases. It knows which documents have not arrived, who was asked, when, and what the start date is, and it follows up with the candidate, the agency or the subcontractor in the channel they actually read. Chasing is most of the job and the part humans do worst, because it is interruptible work with no satisfaction in it.

It also verifies what software can honestly verify: that a document is present, legible, in date, and consistent with the other records. A name that differs between the licence and the right to work document, a certificate expiring nine days after the start date, a licence category that does not cover the allocated vehicle. Those are found by comparison. Then it flags, separately and clearly, what a human must look at in person.

discussion in a modern office
Photo via Unsplash

Kit, vehicle and system access before the first shift

The failure at the top of this article is a provisioning failure, the most common one, and it happens because the starter exists in the HR record and nowhere else. System access here means the transport management system, the warehouse management system, a handheld or scanner with the right profile, a telematics or driver app account, a fuel card, a tachograph driver card, a site access card and sometimes a customer's own portal. None of them share an identity. Each has a different owner and lead time, and some take days.

An agent can raise each request with the role profile derived from the job and the site, track all of them against one start date, and escalate what will not land in time while there is still time to act. The same run covers the physical items: uniform in the right sizes, protective equipment, keys, a locker, and the vehicle or round allocation the planner needs before the morning.

It should not grant its own access: provisioning agents holding standing credentials across your transport and warehouse systems are a security problem that outlives the one they solved.

Agency, subcontractor and self employed labour

Most networks run a mixed workforce, and the record is split. The agency holds identity and some training history. The subcontractor holds their own drivers' documents and may regard them as none of your business. Owner drivers and self employed couriers sit on a different compliance path with different paperwork and a different legal footing, and treating them like employees in your onboarding flow creates both a status risk and a gap in the records.

So the first useful mechanic is to look for a prior record before anything starts, matching on several signals rather than a name, and put a candidate match to a human rather than merging automatically. The courier who left in January comes back in October under a different agency reference, and most depots either induct them from scratch or assume they are already inducted because someone recognises them.

The second is to maintain one answer to a question most operators cannot answer today: of the people working on this site this week, which hold a valid induction, a valid licence check and in date certification, whoever employs them. Where that data may sit and what a supplier is obliged to hand over is a commercial and data protection question to settle before any build.

The expiring certification register

Licence checks, driver qualification cards, medicals, forklift and plant tickets and periodic training all expire. The register that tracks them is usually a spreadsheet, owned by one person, and behind.

Nothing breaks on the day a check lapses. The driver keeps driving, the operator keeps operating, and you find out during an audit or after an incident.

An agent can keep the register honest by reconciling every source holding a fragment of it, flagging contradictions to a named person instead of choosing a winner, and surfacing lapses far enough ahead that a recheck can be booked on a shift the person is rostered to work. On a rotating pattern that constraint matters more than the reminder.

tall shelving units in a distribution warehouse
Photo via Unsplash

The induction evidence pack per site

A customer audits a site and asks for proof that everyone on their contract was inducted to that site, on what date, by whom, against which version of the briefing. Today that is folder archaeology across several systems and a filing cabinet, under time pressure, by the same administrator.

Retrieval and assembly are what agents are genuinely good at. If induction produces its evidence as a by product, with sources linked rather than retyped, the pack comes together in minutes rather than days. Capture the briefing version, because when a procedure is revised the useful output is the list of people briefed against the superseded one. The agent tracks completion against your records, and must never mark an induction as delivered.

The first week, and answering the starter

A new starter's first week happens at a depot at six in the morning, not in an HR office. They are standing in a yard, on a phone, in cold weather, and often reading your policies in a second language.

An agent can answer their questions from your actual documents: how breaks and driving hours work in your operation, the defect reporting process, how to record a delivery exception on the handheld, how pay for an owner driver is calculated, who to call when the vehicle will not start. In their own language, with the source named, and with a tested answer of not knowing when the question falls outside what you have written down. That beats a handbook nobody opens, and it takes real volume off the depot administrator.

Where AI fails here

Statutory checks stay with accountable humans. A right to work check must be carried out by a person through the prescribed process, and the defence it establishes depends on that process being followed and recorded properly. An agent can collect documents, prompt the check, compare details and hold the record. It cannot perform the check, and software claiming otherwise is selling you a liability.

The decision to let someone drive a vehicle or operate plant also stays with a named human. Entitlement data, endorsements and expiry dates can be gathered and presented, but whether this driver goes out in this vehicle today is a transport manager's judgement, and an endorsement history needs a person to read it rather than a rule to score it.

An agent cannot observe a person. It cannot tell you whether a site induction was understood or whether the new counterbalance operator is safe in a busy aisle. Proving a briefing happened and proving someone is competent are different claims, and blurring them produces a comfortable fiction that fails at the only moment it is tested. A confidently wrong compliance record is worse than no record, because people stop checking, so the agent must answer unknown rather than infer and must never derive an authorisation from a training completion.

The biggest limit is not technical. An onboarding agent built on a broken process just automates the mess faster. If each depot runs its own induction, the certification register has never been reliable, and nobody can say who owns device issue, the agent has nothing coherent to orchestrate. It will chase the wrong people for the wrong documents with great persistence, and that fix is yours before it is anyone's software.

Where to start

Take one depot and one intake. Start read only: reconcile the certification register from every source holding part of it, and let the accountable person see the contradictions and the lapses. That alone tells you whether your data supports anything more. Then add the pre start checklist and the chasing for that depot, with every statutory check left explicitly to a human and logged as such.

Measure two things: how many starters arrive with access and kit ready on day one, and how long it takes to assemble an induction evidence pack for an audit.

If you run a network and want a straight read on whether your records could support this, write to Adopt AI Partners at hello@adoptaipartners.com.

Tell us what your teams keep redoing

One conversation is enough to say whether there is a system worth building, and what would make it fail.

Describe your process