The Rise and Fall of Transport Consultants in the Age of AI

A Desk-Based Assessment letter doesn't feel urgent, until you notice it's the DVSA's polite way of asking "can you prove this, in writing, before we come and check ourselves." Most operators can produce the records. Fewer can produce them in a way that actually answers the question asked.

That gap is where consultants have built a business. It's also the gap AI now closes for less money, less time, and, used properly, less risk than most operators assume. This piece sets out why, and exactly where the human still has to stay in charge.

What actually lands in the inbox

Strip away the anxiety and a DVSA compliance action is a defined document exercise. It starts with a 14-day request letter from the Office of the Traffic Commissioner, followed by a Desk-Based Assessment Questionnaire (DBAQ), the DVSA's actual checklist, dressed up as a questionnaire.

If it goes badly, next comes a Maintenance Investigation Visit Report (MIVR) or Traffic Examiner Visit Report (TEVR), and ultimately a Public Inquiry. The evidence required there is largely the same set of records: safety inspection sheets, PMI records, defect rectification evidence, driver licence-checking logs, under considerably more scrutiny, with the operator's and Transport Manager's good repute on the line. Real Public Inquiry decision letters, where operators and their transport managers have lost that repute and had licences revoked, circulate in operator forums for good reason. This is not a category of paperwork worth treating casually, and nothing that follows should be read as suggesting otherwise.

The actual question a DBA response has to answer is narrower than it feels: can you produce systematic, dated evidence against a known checklist, before a deadline, correctly framed? That's document-matching work. It rewards consistency far more than cleverness, which is precisely where this gets interesting.

Operators are already being told to do this

On TransportForum.com, a community of around 47,000 to 48,000 UK operators and transport managers, the guidance thread covering DVSA audit and desk-based assessment documents tells members plainly: use AI tools to draft answers to most of the DVSA's questions, and don't attempt it by hand. That's operators telling other operators. Nobody's selling anything in that thread.

The consultancy market this piece is arguing against isn't abstract, it's active on the same forum, in the same weeks. A thread on choosing an operator licence audit provider names half a dozen firms operators actively recommend to each other, with the advice to check they're on the DVSA's Earned Recognition approved list before paying. That's a live market operating a few clicks from the thread telling members to draft their own DBA answers with AI instead.

Before any tool gets named

Most guidance on this topic jumps straight from "AI can help" to a list of product names. We'd rather you understood where it can go wrong first.

The honest downside. Even a well-built, document-grounded AI tool isn't a neutral instrument. Grounding it in your own documents reduces the risk of it inventing an answer. It doesn't remove that risk. Retrieval can favour whichever passage matches the wording of a question most closely, not necessarily the most relevant one, and a "grounded" tool can still get an edge case wrong. Treat every AI output as a draft with a non-zero error rate, because it is one.

Not all tools are the same tool. A general-purpose chatbot, answering from open internet training data, is the wrong fit here: it can blend outdated regulatory guidance with current guidance and hand you a confident, wrong citation with no way to trace it back. A document-grounded tool, restricted to files you upload and citing the exact passage behind each answer, is the right category for DBA drafting. Specialist compliance software, built for operators specifically, sits in between. The brand matters less than one question: does it retrieve only from your documents, and does it show you exactly where each answer came from?

Verification is not optional, and it is not a formality. Every AI-drafted answer needs a citation to the document and page it came from. No citation, no use. Spot-check a sample of those citations against the source document, every time, not just the first time. Anything referencing a regulatory standard, a rectification timeframe, an inspection frequency, gets checked against the Guide to Maintaining Roadworthiness or the relevant gov.uk guidance directly, not against the AI's summary of it. And a second reviewer, someone other than whoever ran the AI drafting, catches what familiarity with the tool tends to make the first reviewer miss.

Know the technical limits. Tools process a finite amount of text at once. Years of PMI and defect records can exceed what a single query handles cleanly, and documents get chunked in ways that can lose a pattern that only becomes visible reading the full run in sequence. Scanned or handwritten inspection sheets depend on decent text recognition before a tool can read them at all, and a poor scan produces poor retrieval quietly: it won't flag that it misread a date. And uploaded documents typically sit in a vector database inside the tool itself, which is a data asset in its own right, discoverable if a dispute or investigation ever required disclosure of what the tool held and returned.

Two controls do most of the work. Lock the tool to your own documents only for anything DVSA-facing, no blending in general internet knowledge. And require a source citation on every answer before it's used. Beyond that, the human sign-off described further down is the actual risk control. The AI output is an input to a decision. It is not the decision.

GDPR: the bit operators skip, and shouldn't

Driver licence-checking logs, defect records tied to named individuals, and maintenance files are personal data. Uploading them into a third-party AI tool is a data-processing decision with real obligations attached, not an IT detail to wave through.

The Data (Use and Access) Act 2025 amended UK GDPR's Article 22 rules on automated decision-making, giving individuals clearer rights to request human review of, and contest, decisions made solely by automated processing with legal or similarly significant effect. That's the law backing up something this piece has been saying anyway: an AI tool should never be the sole basis for a decision affecting a named driver's record or standing.

The ICO has generative AI and automated decision-making listed among its live priorities through 2026, with a statutory Code of Practice on AI and Automated Decision-Making due under the 2026 Regulations. The rules here are still settling, not finished.

One point worth remembering before anything gets uploaded: tier matters more than brand. Free, consumer-facing AI tools commonly train on user inputs by default; paid or business tiers of the same tool typically come with a contractual no-training commitment instead. Uploading driver records into a free consumer chatbot and into a business-tier document tool from the same company are two different data protection positions, not a rounding error. Check training, retention, and storage location before anything goes in. It's the same due diligence you'd apply to any other data processor, because that's what it is.

What to actually look for in a tool

Retrieval restricted to your own documents. A visible citation on every answer. A written no-training commitment at whatever tier you're on. Retention terms you've actually read, not skimmed. Naming specific products against that list is a moving target, policies change, so we're treating that as its own piece, judged against current published terms rather than reputation.

Where AI genuinely earns its keep

Used properly, document-grounded, citation-backed, an AI tool does structured document review: cross-checking PMI records against required inspection frequency, flagging gaps in defect-rectification paperwork, mapping evidence onto the DBAQ's actual questions, in minutes rather than days.

That's the honest scope of what it replaces: not everything a consultant offers, but the mechanical layer, finding the gaps, structuring the response, drafting the narrative against a checklist a competent transport manager already understands.

So is the consultant actually still needed, or is that just the comfortable answer?

That's the real debate, and it deserves more than "AI helps, but you still need a human" stated as if it settles anything. Both sides of that sentence need testing before they're trusted.

Start with the instinct that of course you still need an expert. Kahneman and Tversky spent a career showing that confident, experienced people reliably overrate how much their judgment is actually adding, a pattern they called the illusion of validity. The transport compliance industry has every reason to feel sure it's irreplaceable: it's built on decades of "you need someone who's seen this before." That confidence isn't proof. Status quo bias looks exactly like professional wisdom from the inside, and an industry doesn't get to mark its own homework on whether it's still necessary.

Now test the other side just as hard. "AI can do 80% of it, so the consultant's redundant" is its own kind of overconfidence, the same illusion of validity wearing a different jacket, this time belonging to whoever's excited about the tool rather than the fee. A fluent, citation-backed AI draft feels authoritative. Feeling authoritative and being sufficient are not the same claim, and treating them as the same is exactly the bias this piece has been warning operators about all along.

So which is it? The honest answer splits the difference in a specific, checkable way rather than a comfortable one. What a document-grounded AI tool plus a competent, engaged transport manager can now do is the part that was always mechanical: matching evidence to a checklist, drafting the narrative, catching the gap a tired reviewer misses. That's most of what a generalist compliance consultant was actually being paid for, and there's no honest reason left to pay someone else to do it. What's left, and what genuinely justifies bringing in an outside specialist, is narrower than the industry likes to claim and real all the same: someone who's sat through enough Public Inquiries across enough Traffic Commissioner areas to know how a specific argument actually lands in that specific room. That's not available in your own documents, however good the AI reading them is, because it was never your documents that contained it. It was every other operator's case the consultant has seen and yours hasn't.

What a consultant would actually still bring, and why that's a smaller list than it used to be

This is the honest inventory, not the sales version.

A consultant's edge was never really "knows the regulations." That part is now a solved problem: it's exactly what a document-grounded AI tool does well, and arguably does more consistently than a generalist consultant juggling twenty clients. The regulations are public. The checklist is public. Knowing them was never the scarce thing, it just felt that way because checking them properly used to take a specialist's time.

What doesn't transfer to a tool trained on one operator's own files is pattern recognition built across many operators: which arguments a particular Commissioner has actually accepted before, what a "Report to OTC" outcome tends to escalate into in practice, how a specific phrase in a DBA response has landed badly in the past. That's real, it's earned through volume no single operator generates on their own, and it's the genuine remaining case for paying an outside specialist rather than doing it all in-house.

Everything else on the traditional consultant's price list, the document assembly, the checklist-matching, the first-draft narrative, is the part this piece has already argued is redundant. Calling the whole profession redundant overstates it. Calling most of what a generalist consultancy bills for redundant does not.

What this means for you: human in the loop, always

"Consultants are redundant" is a claim about the document-assembly layer specifically. It is not licence to let AI output go anywhere near the DVSA unreviewed. Here's where a person still has to be the one deciding:

• Materiality. AI can tell you a defect took 11 days against a 7-day target. It cannot tell you whether this Commissioner, with this operator's history, reads that as minor or systemic.
• What not to say, and how to say it. Disclosure and tone are a judgement call, not a checklist item.
• Representation. If it reaches Public Inquiry, that's a legal and reputational proceeding in front of a Traffic Commissioner. No AI tool stands in for a transport manager or regulatory solicitor in that room.
• Currency. What a DBA actually probes for shifts over time. A tool grounded only in your own files won't catch a change it was never shown.
• Data handling. Covered above. A legal obligation, not a caveat.

Most operators don't lose their licence over one big mistake. They lose it over twelve small ones nobody wrote down. AI is very good at making sure they get written down. It is not the thing deciding what to do about them.

The bottom line

The document-assembly end of transport compliance consultancy is redundant now, for any operator running a document-grounded AI tool with proper verification and a competent transport manager's judgement sitting on top of it. Operators are already being told this by their own forums, not by us.

What isn't redundant is judgement and representation: reading how a Commissioner will actually take a gap, knowing what not to say, standing in the room if it gets that far. That doesn't need a consultant either. It needs the transport manager to be good at their job and to own the final decision, rather than outsourcing it to a consultancy or trusting an AI draft without checking it. AI removes the excuse for paying someone else to do work you were always better placed to check yourself. It doesn't remove the requirement to check it.

If you'd like an independent view of where your own compliance file actually stands, AI-assisted or otherwise, get in touch for a no-obligation conversation.