Before the quote: how AI can help organize a customer’s request

The work starts before calculating a price

In October 2024, DHL Supply Chain described generative AI applications for organizing data received from prospective customers and supporting the initial analysis of quotation requirements. Engineers and sales teams use that preparation to develop solutions and proposals. DHL says it built the applications with support from BCG X.

For a smaller business, the relevant connection is a familiar routine: gathering enough information to respond well. Before pricing, someone has to establish what was requested, which version applies, what is missing and which conditions need a conversation. That stage deserves attention in its own right.

A request can be scattered across several places

Consider an inquiry that starts by email, gains a spreadsheet and is corrected in another message. The salesperson may remember the change while the person preparing the quote uses the earlier attachment. The problem is not necessarily missing information. It may be difficulty identifying the current version and making changes visible to everyone.

Mapping that path helps identify where to intervene. Which documents arrive? Who needs them? How is a revision communicated? Where are open questions recorded? Before automating preparation, the team needs to agree what makes a request clear enough to proceed and who confirms unresolved points.

A preparation test, with review built in

An initial application could extract items, quantities, deadlines and conditions from the materials received. The output would be a summary with references to the original passages, missing fields and conflicts between versions. The team would check it before using it. A quantity that cannot be found should remain an open question rather than being filled in with an assumption.

The workflow should be shaped around the company’s documents, systems and criteria. Preparing a request does not mean approving credit, promising availability, granting a discount or sending a proposal. Each of those actions has rules and owners that need to be defined separately.

What needs to be organized first

Start with a category of requests your team knows well. Gather varied examples, including hard-to-read attachments, revisions and incomplete inquiries. Define the essential fields and the sources accepted for confirming them. Identify restricted information and who may access it; a test does not need to expose documents from across the company.

Choose who will review the output and how corrections will be recorded. If a sales system is involved, establish when information may be written to it and how the team will recognize a failure. Until those conditions are resolved, producing a summary for review may be a more useful initial scope.

How to tell whether preparation improved

Observe the current process before testing: time spent gathering information, repeated questions to customers and proposals reworked because a requirement was misunderstood. Track those same points in the new workflow. Separate reading time, review time and time waiting for an answer. Making one stage faster does not guarantee that the whole quoting process improved.

Record omissions and incorrect interpretations, including those involving difficult documents. Ask the salesperson whether the summary helps them speak with the customer. Ask the proposal’s author whether the references make the request easy to check. The quality of review matters as much as the speed of the first draft.

An application with a clear beginning and scope

DHL’s example shows generative AI supporting the preparation of commercial work. Adapting the idea requires choosing a stage, assessing the available data and identifying who remains responsible for the delivery. The next step does not have to be automating an entire quote. It may be finding a better way to get clear information to the person preparing it.