Questions to Ask a 3PL About Its AI and Automation: A Checklist

Before signing, ask a 3PL what its automation does to order accuracy, speed and stock visibility, which jobs use AI and which use plain rules, where your data goes, how exceptions are handled, what technology fees you will pay, and how you can trial and exit. Ask for numbers and a sample report for each answer.

Warehouses describe almost any software as AI now. A barcode scanner with a rule that prevents a wrong pick is useful and may be all you need; a "smart" label on a manual process is not. The point of these questions is to turn the label into a job, a measure and a fee. Use the list as a script on the call and as a scoring sheet afterwards.

Follow me on Instagram@sagnikteaches

How to run the call so the answers are comparable

Send the questions in writing before the call and ask for written answers for the numbered ones. People say more in a sales conversation than they commit to in an email, and a gap between the two is useful to know about. Run the same questions with every 3PL on your shortlist so the scoring sheet at the end compares like with like.

Connect on LinkedInSagnik Bhattacharya

Describe your own business in four lines first: order volume per month, average lines per order, how many products you carry, and any special handling such as fragile items, batches with expiry dates, or kitting. A 3PL's answer to "how accurate are you" means nothing until it knows what you ship. The examples below follow an illustrative brand that sells dog treats and accessories, ships 1,800 orders a month with an average of 2.3 lines each, and carries 140 products, some with best-before dates.

Subscribe on YouTube@sagnikteaches

Area 1: Systems and integrations

1. Which warehouse management system do you run, and is it yours or licensed? A good answer names the system, how long they have used it and who fixes it when it breaks. A licensed system is not a problem; one nobody on staff understands is.
2. How does it connect to my store? Ask for the list of platforms with a tested connection, and whether it is a direct integration, a third-party connector or a manual file upload. Ask how often orders and stock levels sync, in minutes, and what happens when a sync fails.
3. Who owns the connection when something changes? If your store platform updates, whose job is it to make the connector work again, and within what time?
4. Can I see live inventory by SKU and by location, and export it? Ask for a demo login or a recorded screen share, not a slide.

Red flag: a connector described as "being built" for a platform you use. You would be funding their development. Stock sync problems are the commonest source of oversells, and the tutorial on syncing stock across Shopify, Amazon and eBay explains what a sync actually has to do.

Area 2: Accuracy and speed, in numbers

5. What is your order accuracy, and how do you measure it? The definition matters. Ask whether it counts only orders with the wrong item, or also missing items, damaged items and wrong quantity, and over which period.
6. What is your order cut-off and your dispatch time? For example, orders in by 2pm leave the same day. Ask what share of orders meet that, and what happens in the week before a holiday.
7. How is inventory counted, and how often? Ask for the cycle-count schedule and the inventory accuracy figure by location, and how a variance is resolved and credited.
8. What checks stop a wrong pick before it leaves the building? Scan-to-verify, weight checks, camera checks. Ask which of these is automated and which relies on a person.

Do not treat any figure as a benchmark from me. I am deliberately not quoting a "good" accuracy percentage, because definitions differ and I have no source to give. Ask the 3PL for its figure, its definition and a report that shows it, then agree the number you will measure in the trial. For the diagnosis side, tracing pick errors back to their cause shows what a root-cause review should look like.

An example of a strong answer, illustrative: "Accuracy is measured as orders shipped with the correct items and quantities divided by orders shipped, from our own audit and from customer-reported errors, reported monthly. Here is last month's report with the two numbers shown separately." A weak answer is a single percentage with no definition.

Area 3: What is AI, what is a rule, what is a person

9. Which jobs in my account use AI, and what exactly does each do? Common answers: demand forecasting, slotting (deciding where stock is stored), pick-path planning, carrier selection, address correction, damage detection from photos, and chat or email handling for your customer questions. Make them name the job.
10. For each, what is the fallback when the AI is wrong? Who sees the exception, and how fast?
11. Is the AI your own, or a vendor's feature, and who is the vendor? Fine either way, but you need to know whose terms apply to your data.
12. What proof do you have it works for brands like mine? A reference customer you may call is the best answer. A case study with no numbers is the weakest.

Sort the answers into three bins as you go: AI (a model learning from data or reading images or text), rules (fixed logic, which is often fine), and manual (a person doing it). A warehouse with strong rules and scanning may beat one with a vague AI story. If you want a sense of what a small 3PL could realistically be using and what it brings, the tools a small warehouse might run is a useful reference.

A filled-in example for the dog treats brand shows the sorting:

Area 4: Your data and who can see it

13. What of my data is used by any AI feature? Order data, customer names and addresses, product images, supplier documents?
14. Is it used to train a model, yours or a vendor's? Ask for a yes or a no, then the contract clause.
15. Which third parties process it? Ask for the list of sub-processors.
16. Where is it stored, and for how long after I leave? And how do I get a full export, in what format, within how many days of ending the contract?
17. Who at your company can see my customers' details? Role-based access and a log of who looked are good signs.

If you have customers in the EU, data-protection law applies to the personal data your 3PL handles for you, and you remain responsible for it. A contract that states what the 3PL does with that data, and a data-processing agreement, belong in the paperwork, and a qualified adviser should read them. A 3PL who cannot answer question 14 quickly has probably not thought about it.

Area 5: Exceptions, returns and communication

18. Walk me through a mis-shipped order from the first signal to the credit. Listen for who is told, how long each step takes and how it is written down.
19. How are damaged, short and late deliveries reported to me? A daily exception list is better than a monthly summary.
20. How are returns handled? Is the item inspected by a person or a photo check, how quickly does stock go back on sale, and how is the decision to restock, refurbish or write off made? Ask for the average days from the return arriving to the stock being sellable.
21. Who is my named contact, and what is the response time? Ask for it in writing.
22. What do you do on a day when volume is double? For instance, after a promotion you did not warn them about.

Test the exception answer with a scenario. Say a customer receives the wrong flavour of treats on a Friday afternoon. Strong answer: "The customer-service ticket and the pick record are matched the same day, we confirm what happened by Monday, and if the error is ours the cost of the item, postage and the re-send are credited on the next invoice." Weak answer: "We'll look into it."

Area 6: Reporting and the fees that come with technology

23. Which reports do I get, how often and in what format? Ask for a sample, and whether I can pull raw data without asking you. For what good reporting looks like from the 3PL's side, see weekly client KPI reports from WMS exports.
24. What technology, integration, setup or minimum fees apply? Ask for each fee by name, the unit it is charged on and when it starts.
25. Do any AI features cost extra, now or after a trial period? Some features start free and move to a paid tier.
26. How do I check my invoice against your activity? Ask whether invoices can be matched to a line-by-line activity report, because that is how you catch double charges. If the answer is a PDF total with no activity lines, you will be checking invoices on trust.
27. What are the contract term, notice period and exit costs?

A quote that looks cheap per order often carries fees in lines you did not compare. The tutorial on comparing every fee on a 3PL quote sets out a way to put quotes into one table.

Red flags worth walking away from

• "AI" offered as an answer to every question, with no job, report or number attached.
• No sample report. If a 3PL cannot show what you will see, you will probably not see it.
• A refusal to say whether your data trains a model.
• A required long contract before any trial, or exit fees that are larger than the trial would cost.
• Stock sync described as "usually fine" without a failure process.
• Different answers from the salesperson and from operations. Ask to speak to the warehouse manager, and compare.
• No reference customer of a similar size and product type.

One red flag is a reason to ask more. Three is a reason to move on.

Score the answers on one sheet

Give each area a score from 0 to 3 per 3PL: 0 for no answer, 1 for a vague answer, 2 for a clear answer without evidence, 3 for a clear answer with a report, reference or clause to back it up. Weight the areas to your business. A brand with best-before dates should double-weight area 2's inventory questions; a brand with heavy returns should double-weight area 5.

Weighted scores (score x weight)
Area                         Weight  3PL A  3PL B
1 Systems and integrations      2      6      4
2 Accuracy and speed            3      6      9
3 What is AI, rule, person      1      2      3
4 Data                          2      4      6
5 Exceptions and returns        2      4      6
6 Reporting and fees            2      4      6
Total                                   26     34

Illustrative: 3PL B wins on accuracy and data answers even though A's systems answers were stronger. The sheet does not decide for you, but it keeps a smooth sales call from outweighing a weak answer in an area that matters to you.

Use an assistant to compare the written answers

Once two or three 3PLs have replied in writing, paste their answers, with names removed, into ChatGPT, Claude or Gemini. Do not paste your customers' data.

Below are written answers from three warehouse providers to the same list of 27 questions. Label them A, B and C.
For each question, give a one-line summary of each provider's answer and flag where an answer is: missing, vague (no number, report or named system), or inconsistent with another answer from the same provider.
Then list, for each provider, the three questions I should ask again on a second call, and why.
Do not rank the providers or recommend one.

A useful reply, illustrative, will point out that provider A gave an accuracy figure in question 5 without a definition, and that provider C said no data is used for training in question 14 but listed a third-party AI vendor in question 16. What I would fix: assistants tend to smooth over contradictions and be generous with vague answers, so read the three flagged items against the originals. It is a way to cut reading time, not a substitute for the follow-up call.

The email to send before the call

A short, specific email gets better written answers than a questionnaire attached to a vague hello. Here is a version for the dog treats brand, with the business facts filled in. Replace them with yours.

Subject: Fulfilment enquiry: 1,800 orders a month, 140 SKUs

Hello [first name],

We sell dog treats and accessories online and are choosing a fulfilment partner. We ship about 1,800 orders a month, an average of 2.3 lines each, from 140 products. About 40 of those have best-before dates.

Before a call, could you answer the attached 27 questions in writing, and send a sample of the weekly report and invoice a client of our size receives? Where you use AI or automation, please name the job it does and what happens when it is wrong.

We are also looking for a 30 to 60 day trial on part of our range. Could you tell us how that would be structured?

Thank you,
[your name]

Two parts of that email do most of the work. "Name the job it does" pushes a salesperson from slogans to specifics. And asking for a sample report and invoice shows you the real product before you have signed anything.

Reading a vague answer: three before-and-after examples

A vague answer is not a lie, but it is an opening. Here is what to ask next, using real-sounding replies.

"Our AI optimises picking." Ask: which part, and what changed in pick time or errors for a client after it went in? A better answer: "Pick paths are sequenced by the system so a picker walks one loop; average lines picked per hour went up after the change, and here is the report." If they cannot show the report, the claim is unproven for you.

"We have 99 per cent accuracy." Ask: measured how, over which months, with which errors counted? The better answer separates picking errors found by audit from errors reported by customers. A single clean number with no definition tells you nothing about whether it would hold for your fragile or dated products.

"Your data is safe with us." Ask: does any AI feature read it, does any vendor process it, and what is in the contract? A better answer is a named list of processors, a statement on model training and a data-processing agreement ready to sign.

Questions for the warehouse tour, if you can visit

If you can walk the floor, or ask for a live video tour, look for evidence of the answers you were given. Are the locations labelled and are items scanned at receiving? Is there a quarantine area for damaged or unidentified stock? Do returns have a marked area and a process? Are the bins for your products tidy, with the oldest dated stock at the front? How many staff have a screen or scanner in hand, and how many are working from paper? None of this is proof, but a gap between what they said and what you see is worth a polite question.

The trial period: how to test the claims

A trial is where the answers meet your products. Agree this in writing before it starts:

1. Scope. One product range or one sales channel, say 20 to 30 per cent of your volume, so a failure costs little. For the dog treats brand, that might be the accessories range and none of the dated products.
2. Length. Thirty to sixty days, and it must include a stock receipt, a returns batch and one busy period. If it does not, extend it.
3. Measures. Order accuracy on the definition from question 5, dispatch time against the stated cut-off, inventory variance on a cycle count you witness, days from return arrival to restock, and the time to resolve each exception.
4. A weekly review. Fifteen minutes, with the report in front of both sides.
5. Exit terms. If the measures miss by an agreed margin, you can end the trial, retrieve stock and data within an agreed number of days and pay no exit fee.

During the trial, plant a few deliberate tests. Send an inbound shipment with one wrong quantity on the delivery note and see whether it is caught at receiving. Place a small test order with an unusual address format. Submit a return with a missing item. How the 3PL handles three small oddities tells you more about its systems than any pitch.

When the trial ends, look at the numbers for what they say, not for what you hoped. If accuracy missed but exceptions were fixed quickly and credits came on time, a second month may be reasonable. If the data was hard to get, the technology claims have already failed. For how a 3PL sets up a new client properly on its own side, onboarding checklists with SOPs and service levels is a good yardstick for what your onboarding should have included.

Questions brands ask before choosing a 3PL

Should a 3PL be using AI at all?

Not necessarily. Many reliable warehouses run on barcode scanning and good rules. What matters is accuracy, speed, visibility and fair fees. Ask what the technology changes for you in numbers, and treat the label 'AI-powered' as a prompt to ask for the specific job it does.

What is the most important question to ask?

Ask them to walk you through what happens to one order that goes wrong, from the first signal to the fix and the credit. The answer shows their exception process, how fast their data updates and whether a person owns the problem.

How long should a trial with a new 3PL last?

Thirty to sixty days, covering at least one stock receipt, one returns batch and one busy week. Start with a limited range of products. Agree the measures and the exit terms in writing first, so the trial can end cleanly if the numbers miss.

Can I ask a 3PL not to use my data for AI?

Yes, ask in writing. Request a clear statement of what data goes into any AI feature, whether a third party processes it, whether it is used to train a model and how to opt out. Put the answer into the contract, not just an email.

Further reads

• AI Robots for Warehouses: Do They Pay Off for a Small Operator? — Judge claims about robots and automation at a small operator.
• Zapier vs Shopify Flow: Which Automation Tool Suits a Small Store? — Decide what to automate on your own side of the handover.
• AI Accounting for Ecommerce: Reconciling Payouts and Fees — Plan the finance side once orders flow through a 3PL.
• 3PL Billing: AI Reconciles Storage and Pick Fees Against WMS Data — Check each invoice line against warehouse activity data.
• 3PL Pricing Calculator: AI Quotes Storage, Pick and Pack Rates — A 3PL operator builds its own rate card and quote calculator: cost-based storage, pick and pack rates, a worked quote and a margin check.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: no vendor pricing is quoted; the questions are built from common warehouse operations and integration practice, with every figure in the examples illustrative.