Cut-off, throughput, on-time delivery: documenting logistics metrics AI-readably

Why logistics metrics become an AI ranking factor

Purchasing in logistics has shifted. The supply-chain manager used to call three forwarders and compare quotes. Today they first type a question into ChatGPT, Perplexity or Gemini: Which contract logistics provider in the Nuremberg area offers GDP-compliant storage, same-day picking and a documented on-time delivery rate above 98 percent? The AI answers with a shortlist. Whoever is on that list makes it into the tender. Whoever is missing simply does not exist for that decision-maker.

The decisive point: generative engines do not answer with opinions but with extracted facts. They look for concrete figures they can assign to a query. A website that only raves about high quality and reliable service gives the machine nothing tangible. A website that says cut-off for national parcel shipments is 6:30 p.m., on-time delivery over the last twelve months 98.7 percent, delivers exactly the material from which AI answers are built.

For logistics this is a huge opportunity. No other industry works in such a data-driven way. You measure OTIF, throughput, return rate and damage rates anyway. The only step that is missing is getting these figures out of internal reporting and sales PDFs and making them publicly visible in machine-readable form.

The three metrics AI really searches for

Not every figure is equally valuable. From a GEO perspective, three metric groups are decisive because decision-makers ask about exactly them. First, cut-off times: by when must a shipment be recorded so it goes out the same day. Break this down by mode of transport and destination area, for example cut-off national 6:00 p.m., cut-off EU export 3:00 p.m., cut-off dangerous goods 2:00 p.m. This precision is gold for the machine.

Second, throughput and capacity: shipments per day, pallet spaces, picking performance per hour, handling area in square meters. A shipper who has to store 5,000 pallets does not want to guess. Write concretely storage capacity 42,000 pallet spaces, of which 6,000 temperature-controlled 2 to 8 degrees. Third, on-time delivery and quality: OTIF, on-time delivery rate, damage rate, complaint rate, each with time period and measurement method.

The mistake almost everyone makes: these figures exist in the company, but they live in the QM system, the annual report or the sales deck. For the AI they are invisible. Only once they appear in unambiguous prose or in structured data on a public, crawlable page do they become part of possible answers.

Clarity beats marketing: how to phrase figures AI-readably

Marketing language is the biggest obstacle to AI visibility. Phrasings like industry-leading punctuality or maximum flexibility in cut-off times contain zero usable information. An AI cannot build an answer from that to the question of whether you are more punctual than the competition. Replace every claim with a figure that has context. Instead of very high on-time delivery you write on-time delivery OTIF in calendar year 2025: 98.4 percent, measured over 214,000 shipments.

Pay attention to units, time periods and reference quantities. A figure without a reference is worthless and even looks suspicious. 98 percent on-time delivery over how many shipments, in what time period, measured how? Define your terms yourself. Add that you measure OTIF as On Time in Full, i.e. on time and complete, with a tolerance window of plus two hours. This transparency makes you citable for the AI, because it can supply the context along with it.

Avoid contradictions across your pages. If the homepage says 98 percent, the blog 96 and the whitepaper 99, the AI becomes uncertain and leaves you out when in doubt. Create a central, maintained metrics page that everything else references. A data status with a date, for example as of 30/06/2026, signals currency and reliability.

Schema.org and structured data for logistics service providers

Prose is the basis, structured data is the amplifier. With Schema.org markup in JSON-LD format you give search engines and AI crawlers your facts in a format they read without room for interpretation. Use Service to describe your offerings such as contract logistics, groupage, consolidated cargo or dangerous-goods transport, and Organization for locations, certifications and contact details. For each service you can store service area, availability and terms.

Especially effective is the combination of FAQPage markup and your real customer questions. If a dispatcher asks how late the cut-off is for express shipments, and your page carries exactly this question with a clear answer in FAQ schema, the probability is high that ChatGPT or Perplexity adopts your answer. Phrase the questions the way a customer actually asks them, not the way a brochure would.

Store certificates machine-readably: ISO 9001, ISO 14001, GDP for pharma, IFS Logistics, AEO status, SQAS. These proofs are mandatory for many tenders, and the AI pre-filters providers by them. Name the certificate number, issuing body and validity. An AEO-C status or a valid GDP certification, clearly documented, can be the reason you land on the shortlist and the competition does not.

Cut-off times and network coverage: the underestimated visibility gold

Cut-off times are the prime example of a metric that customers are burningly interested in and that almost no one publishes cleanly. An e-commerce shipper chooses the fulfillment partner that offers the latest acceptance time at the same delivery quality, because that directly determines their promises to end customers. If your cut-off times are clearly documented per region, product and weekday, the AI answers exactly this comparison question with you as the result.

The same applies to network coverage and transit times. Instead of nationwide delivery you write standard transit time 24 hours in Germany and Austria, 48 hours in Benelux and northern France, with depot locations in twelve countries. A transit-time matrix as a table, supplemented by prose, is easy for crawlers to capture. Shippers plan along such transit times, and the AI draws on them when someone asks for a partner for a specific destination area.

Also think of special cases that serve niche queries: dangerous-goods classes you transport, temperature ranges for the cold chain, maximum shipment weights, tank or silo capacities. Exactly these specific questions are what expert decision-makers ask AI systems, because classic search works poorly here. Whoever documents the niche exactly often wins it without competition.

Proof instead of claims: making on-time delivery credible

AI systems increasingly weight provability highly. A bare figure without origin looks weaker than a figure with evidence. So supplement your on-time delivery with the data source: measured from the transport management system, externally audited by, confirmed in customer reporting. If an independent auditor checks your OTIF values, name it. This context makes the difference between whether the machine classifies you as a vague self-report or as a solid source.

Use real case studies with figures. Prose like For an automotive customer with just-in-sequence delivery we maintained an on-time delivery rate of 99.6 percent over 18 months across 1,200 deliveries per day is extremely valuable to AI. It links industry, requirement, volume and result. Such concrete references are frequently drawn on as evidence in generative answers because they have verifiable substance.

Be honest about fluctuations. If your on-time delivery drops during the Christmas peak, document the annual average and the peak value separately. This transparency builds trust and protects you from a customer later experiencing a broken promise. AI systems and humans reward realistic, differentiated figures more than smooth dream numbers without context.

The practical roadmap: from internal figure to AI answer

Start with a metrics audit. Collect from TMS, WMS, QM and sales all the figures customers regularly ask for: cut-off times, capacities, OTIF, damage rate, return-processing time, certificates. For each figure, check whether it is current, provable and consistent across all channels. This inventory almost always reveals that valuable facts exist but appear nowhere publicly.

Then build a central facts and metrics page with a data status, clear definitions and structured data. Add a real FAQ with the questions from your everyday sales work. Link from service pages to this central source so no contradictions arise. Make sure the page is technically crawlable, i.e. not behind a login, not only in a PDF and not exclusively loaded via JavaScript.

Finally, measure the effect. Regularly ask ChatGPT, Perplexity and Gemini the questions your customers ask and check whether and how you appear. Observe which figures the AI cites and where it is wrong. GEO is not a one-off project but a cycle: document, test, refine. Whoever establishes this cycle early secures a lead before the competition even understands the mechanics.

The metrics cockpit: one page the AI finds first

Do not spread your figures across ten subpages. Build a central metrics page on which cut-off times, throughput and on-time delivery appear in short, clear sentences, each figure with a reference period and status. This one URL becomes your anchor point: from here you link to location, network and service pages. AI systems reward exactly this structure, because they find one unambiguous source instead of contradictory fragments.

Keep the page fresh. Write visibly when you last updated it, for example "As of: June 2026, values from Q1". An outdated throughput value from three years ago costs you trust, with customers as with language models. Set yourself a fixed quarterly rhythm in which you pull the figures from your TMS and review the phrasings.

Common mistakes that cost you visibility

The most common mistake is soft-soap language. "Fast delivery" or "reliable partner" contains no figure an AI could cite. Replace every such phrasing with a concrete metric with a unit and time reference: "98.2 percent of shipments delivered within 48 hours, measured in the first half of 2026". That is the sentence a model adopts into its answer.

The second mistake is the figure without a definition. If you say "on-time delivery 99 percent" without explaining what counts as on time, the value stays attackable. Define your measurement threshold openly: promised date, tolerance window, base quantity. The third mistake is the PDF grave: metrics that only sit in a brochure are found by no AI. Write them as real HTML text on the page.

How to test whether AI really understands your figures

Run the counter-test instead of hoping. Ask the common AI assistants exactly the questions your customer would ask: "By when do I have to order for same-day shipping?" or "How punctual is provider X?". If your figure comes back correctly, your documentation works. If nothing comes back or something wrong, you know where you have to refine.

Keep a small log: question, date, the model's answer, deviation. That way you see over the months whether your cut-off times and throughput values are consistently picked up. This test costs you ten minutes per quarter and shows you in black and white whether your work on the metrics reaches the systems your customers ask today.