Tachograph Analysis: What AI Can Summarise, What a Manager Decides

AI cannot read a tachograph download directly, because the file is coded for analysis software. It can summarise the report or spreadsheet that software produces: counting infringements, grouping them by driver and rule, and drafting the follow-up list. What to do about each driver, and whether a breach is excusable, stays with the transport manager.

That split is more than a technicality. A summary is cheap to get wrong in a way you cannot see: a chatbot will cheerfully tell you that a driver "appears compliant" from a table it only half-understood. So the workable approach is to let AI sort and describe, to check it against a report you already trust, and to keep every decision about people in a human's hands. The rule figures below were read from GOV.UK's goods vehicle guidance (updated 1 July 2026) in November 2026; always check the current page before relying on them.

Follow me on Instagram@sagnikteaches

What a tachograph download holds, and why chatbots struggle

A digital or smart tachograph stores a driver's activities (driving, other work, availability and rest) on the vehicle unit and on the driver's card. Operators download both and analyse them. GOV.UK says operators must periodically download the vehicle unit every 90 days and driver cards every 28 days, and analyse the information to make sure the rules have been followed.

Connect on LinkedInSagnik Bhattacharya

The raw download is a binary file made to be opened by analysis software, not a document. If you upload one to a chatbot you will get one of three outcomes: a refusal, gibberish, or a confident summary that has little to do with the file. None of those is useful. The practical pipeline is therefore:

Subscribe on YouTube@sagnikteaches

1. Download the card or vehicle unit with your usual tool and store the original file untouched, because it is the evidence.
2. Open it in your analysis software, which decodes it and flags infringements against the rules.
3. Export the result as a CSV, spreadsheet or report, with driver codes in place of names where you can.
4. Give that export to the AI for summary, sorting and drafting.
5. A named manager reviews the summary against the analysis software's own report, then decides.

Step 4 is the only one where AI helps. Steps 2 and 5 are where accuracy and accountability live.

The rules your analysis is checking against

You need a rough picture of the rules to judge whether a summary makes sense. These are the main limits under the assimilated drivers' hours rules, which apply to most goods vehicles over 3.5 tonnes (including a trailer) used wholly in the country or on journeys to the EU, as GOV.UK describes them. Since 1 July 2026 the same guidance also brings light goods vehicles over 2.5 tonnes into those rules when they carry goods on international journeys, so a van operator who crosses a border needs to check again.

Smaller vehicles and exempt operations can fall under different rules. The GB domestic rules in the same guidance, for example, set a daily driving limit of 10 hours and a daily duty limit of 11 hours, with their own exemptions. Which set applies depends on the vehicle, the journey and any exemption, so confirm it before you set up any automated checking. If an AI summary quotes limits, compare them with this page, not with the chatbot's memory.

What AI is good at in this work

Turning a long report into a ranked list

An analysis export for 15 drivers across 28 days can run to hundreds of rows. A chatbot can group infringements by driver, by rule and by week, and tell you in a paragraph where the weight of the problem sits. That is a sorting job, which AI does reliably when the data is in a clean table.

Spotting patterns across weeks

Look at the shape of the data, not the single event. Three reduced daily rests in a week for the same driver, break infringements clustering on one route, or a spike every second Monday are patterns that point to a scheduling cause rather than a driver choice. A tool can surface them quickly if you ask the right question.

Drafting the conversation notes

For each flagged item, AI can draft a neutral factual note for the manager: the date, the rule, the figures, the number of previous occurrences in the period. It should not draft the conclusion. Managers still pick the wording of any warning and the record of what was said.

Writing the weekly summary for people who do not read tables

Directors and customers want one paragraph, not a spreadsheet. A summary like "Four minor infringements across three drivers this fortnight, all break related, all on the Thursday long route" is useful when the numbers underneath it have been checked.

A prompt and an illustrative result

Use a prompt that fixes the data, forbids guesses and requires a trail back to rows.

You are summarising a driver-hours analysis export for a transport manager.
The CSV below has one row per flagged event: driver_code, date, rule, severity (as
labelled by our analysis software), amount_over_or_under, vehicle.

Tasks:
1. Count events by driver_code and by rule. Show two small tables.
2. List any driver_code with 3 or more events, with the dates.
3. Say which weekday and which vehicle have the most events.
4. For each event of severity "very serious" or "serious", write one factual line:
   driver_code, date, rule, amount.
Do not suggest causes, excuses or disciplinary action. Do not use any figure that is
not in the data. If the data is missing something you need, say what is missing.

DATA:
[paste CSV with driver names replaced by codes]

Illustrative figures from a quieter month at the specialist food importer worked through in the next section: 23 events in 28 days. The summary says drivers D2 and D5 account for 14 of them; 17 are break infringements; the busiest weekday is Thursday; one event is labelled serious (a 10.5-hour day without the extension available). What you do next: open the original report to confirm those 23 rows exist and that none was dropped, check the Thursday run on the schedule, and look at what the serious event actually says. The prompt did its job when it saved you from reading 23 rows to find the pattern. It would have failed if it had said "D2 seems tired", which is exactly the kind of conclusion the instruction forbids.

A worked month: one importer, three decisions

Say a specialist food importer has eight vehicles over 3.5 tonnes and nine drivers on a mix of local drops and a weekly run to a distribution centre. The transport manager downloads driver cards every 28 days and vehicle units every 90, runs them through the analysis package and exports a CSV of flagged events. Last month's export has 31 rows.

The AI summary says: nine of the 31 events belong to one driver; 22 are in the "break" category; 19 happened on days that included the distribution centre run. The manager then does three things the tool cannot.

1. Checks the pattern against the schedule. The distribution centre has a loading window that regularly starts 40 minutes late. Drivers wait, then rush the break to make the 4.5-hour mark. The cause is partly the operator's planning, so the right action is to change the route timing and tell the customer, not to warn nine drivers.
2. Looks at the one driver's nine events. Seven are small breaks. Two are daily rests shorter than allowed. The manager reads the original rows, talks to the driver, and records the conversation. A formal warning would be a judgement based on the full picture, and that judgement is the manager's.
3. Looks for what the tool did not flag. A card that was not inserted for two days will not produce infringements, but it leaves a gap. Missing data is a compliance problem of its own, and an AI summary of flagged events is silent about it unless asked.

Time: perhaps 20 minutes to run and check the summary, against an hour of manual sorting. The saving is in the first pass. Everything after it is still manager time, which is the point.

What stays with the transport manager

Every item in the next table is a decision or duty that carries legal, employment or licensing weight. A tool can supply evidence for it; it cannot own it.

Where summaries go wrong with driver-hours data

• Wrong rule set. A chatbot applies the 9-hour limit to a driver on GB domestic rules, or the other way round. Tell: the summary quotes a limit your analysis software did not use.
• Silent row loss. A long CSV is truncated and the summary counts only what it saw. Tell: the counts do not match the row count of the file. Always ask for the number of rows it processed.
• Fixed-week confusion. Weekly limits apply to a fixed week, Monday 00:00 to Sunday 24:00, and a chatbot may use any seven days. Tell: totals that differ from your software's weekly view.
• Merged drivers. Two drivers with similar codes treated as one. Tell: a count that seems too high for one person.
• Invented causes. "Likely due to traffic on that road." The model cannot know. Tell: any cause in the summary that is not in the data.
• Treating no flags as compliance. A report of flagged events says nothing about missing data, unrecorded other work or working-time rules beyond driving. Tell: a clean summary for a driver whose card was not downloaded.

A mistake that slips through when nobody checks the row count

A small contractor supplying a craft brewery's deliveries exports a four-week CSV of 480 rows into a chatbot. The tool handles about 300 and quietly stops. The summary reads well: no very serious events, one driver with three breaks. The manager forwards it to the owner as the month's compliance picture. A fortnight later a routine check of the original report shows two driving-time events from the last week, which were never in the summary.

Nothing was dishonest; the file was longer than the tool handled. The remedy is two lines in the prompt and a habit: "State how many rows you read and the first and last date you saw," then compare with your own count. It takes ten seconds and catches the whole class of truncation errors.

Keep drivers' data out of places it should not go

Driver records tell you who was where and when, and how long they worked. That is personal data. Before any of it goes into a general chatbot, replace names with codes, remove vehicle registrations if you do not need them for the question, and strip anything about a driver's personal circumstances. Use a plan whose data settings stop your content being used to train models, and keep the code list in a separate file that never goes into the tool. If your analysis provider offers AI summaries inside its own product under the contract you already have, that is usually a cleaner route than copying exports out. The tutorial on classifying business data before using AI tools gives a simple way to decide what may go where.

Drivers should also know how their data is used. If you start producing weekly AI-assisted summaries, say so in plain words in your driver handbook or privacy notice, and say that a person reviews any decision. That is good practice and keeps conversations about infringements calmer.

Three fleets, three different answers

A kitchen fitter with two vans

Both vans are under 3.5 tonnes and stay on domestic jobs, so the assimilated tachograph rules generally do not apply, and GB domestic rules may. There may be no tachograph at all. The GOV.UK guidance lists goods vehicles up to 3.5 tonnes used for installation or fitting among those exempt from the 11-hour domestic duty limit but not from the 10-hour driving limit, which is exactly the kind of detail a chatbot gets wrong from memory. The useful AI task here is not tachograph analysis but a simple weekly log of driving time per day, which a chatbot can total against that 10-hour limit. Confirm which rules apply to your vehicles and work with the GOV.UK guidance before you write the log, and check again if a van over 2.5 tonnes ever takes goods abroad.

A craft brewery supplier with a 7.5-tonne lorry

One vehicle, two drivers, deliveries across a few counties. A smart or digital tachograph is likely to be fitted and the 28-day and 90-day downloads apply. For this size, a monthly export and a 15-minute summary is plenty. AI helps with the write-up; a spreadsheet and the analysis software do most of the work.

A specialist food importer with a mixed fleet

The case worked through above. Volume is high enough that grouping by driver, rule and weekday saves real time, and the schedule links to a customer's loading window. This is where AI pays off, as long as the manager owns the interpretation.

Questions to put to your analysis software provider

Before you build any AI step on top of your analysis package, ask its provider five plain questions and keep the answers.

1. Which rule sets does the package apply, and how does it choose between them for each vehicle and day?
2. Does it flag missing data, such as days with no card activity or gaps between downloads, as well as infringements?
3. Can it export every flagged event as a CSV with a stable column layout, so the same prompt works each month?
4. Does it offer its own AI or summary feature, and where is the data processed and stored?
5. How are rule changes applied, and how will you be told when a limit or definition moves?

The answers decide how much work is left for the chatbot. If the package already produces a clear one-page summary per driver, an AI step adds little. If it exports a long table and nothing else, a summary prompt earns its keep. Either way, the package, not the chatbot, is the source of truth for what counts as an infringement.

Tracking whether the summaries are right

For the first three months, keep a simple accuracy log. For each summary, record the row count the AI reported, the true count, and any figure that differed from the analysis software's report. After three months you will know whether the tool is reliable enough to reduce your checking to a sample. If it is wrong more than once in a dozen runs, tighten the prompt, shorten the data, or stop using it for this job.

Run date | Export | Rows in file | Rows AI reported | Counts match? | Notes
05 Oct   | Cards, 28 days | 214 | 214 | Yes | -
02 Nov   | Cards, 28 days | 190 | 190 | Yes | AI named Thursday; matches software

That log is illustrative, but the habit is real. It turns "I think it works" into evidence you can show a colleague, an auditor or yourself in six months.

A weekly routine that fits a busy transport office

1. Monday: download any cards due and store the originals unchanged. Run the analysis software.
2. Monday, 15 minutes: export flagged events, replace names with codes, run the summary prompt, compare the row count.
3. Tuesday: the manager reads the original report for every serious or very serious item and for any driver with three or more events. Records decisions in the driver's file with the reason.
4. Wednesday: change any schedule that is causing repeat events, and tell the customer if their loading times are part of the cause.
5. Monthly: review the accuracy log and decide whether to keep, tighten or drop the AI step.

If you are looking at wider fleet admin, the tutorial on automating services, checks and fuel covers the other paperwork, and AI route planning for small fleets helps with building schedules that avoid the infringements in the first place. For the review habit itself, see setting up human review for AI work.

This tutorial is general information, not legal advice. Drivers' hours, tachograph and working-time rules are set by government and can change; the guidance cited here was read in November 2026. If a decision could affect a driver's job or your operator licence, speak to a qualified adviser or the relevant authority.

Questions transport managers ask about AI and tachograph data

Can ChatGPT open a .ddd tachograph file?

Not usefully. The download is a coded file made for analysis software, not for reading. Run it through your usual analysis package, then give the chatbot the exported report or CSV. Test on a file you already know the answers to before you trust any summary.

Is tachograph data personal data?

Yes, in practice. It records identifiable drivers' movements, working time and rest, so data-protection law applies to how you store it and where you send it. Anonymise names to driver codes before using a general chatbot, or use a tool your analysis provider has built in.

Should AI decide whether a driver gets a warning?

No. Whether an infringement was a one-off, a fault with the vehicle unit, an emergency or a pattern that needs a formal conversation is a management judgement with employment and licensing consequences. AI can line up the evidence; a named person decides and records why.

How often must tachograph data be downloaded?

GOV.UK guidance says operators must periodically download the vehicle unit every 90 days and driver cards every 28 days, and analyse the data to check compliance. Check the current guidance for your type of operation, because the rules depend on vehicle weight and journey type.

Do these rules apply to my van?

It depends on weight and use. The assimilated drivers' hours rules apply to most goods vehicles over 3.5 tonnes, and since 1 July 2026 also to vans over 2.5 tonnes carrying goods on international journeys. Smaller vehicles on domestic work may fall under GB domestic rules or have exemptions. Check the GOV.UK goods vehicles guidance for your exact vehicle and journeys.

Further reads

• AI Software for Trucking Companies: What to Check for a Small Fleet — What to check when buying AI software for a small fleet.
• How to Stop AI Tools Training on Your Business Data — Stop tools training on your drivers' and customers' data.
• AI Security Risks for Small Businesses and How to Close Them — Security risks to close before sharing operational data.
• Can ChatGPT Read PDFs, Spreadsheets and Photos? What Breaks — What breaks when you give a chatbot files.
• AI Shift Scheduling vs a Rota Spreadsheet for Small Teams — Rotas and shift patterns with AI help.
• AI in Fleet Management: Auditing Yard Time and Tolls From Telematics — Use AI on a telematics export to find drivers idling in the yard on the clock and tolls that never reached an invoice.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: GOV.UK, Drivers' hours and tachographs: goods vehicles (assimilated and AETR rules; GB domestic rules; tachograph rules), page updated 1 July 2026 and read November 2026.