Quotes out faster, orders in without retyping

Quotes go out sooner and with fewer errors, and orders reach the ERP without anyone typing them in again. In many sales teams the bottleneck is paperwork: finding the current price, the right article number, the last quote for this customer and a realistic delivery date across several systems. A vendor case study in the slides of a 2026 BIHK webinar puts it like this: a quote bottleneck, a quote process that is manual, error-prone and slow, and the warning that while you are still looking for data, your competitor has already sent the quote.

Sales is one of the fastest growing AI areas in Germany. In Bitkom's 2026 survey of companies with 20 or more employees, 14% of the companies that use AI use it in sales, up from 5% in 2025. In a 2025 study by VDMA with Strategy& of 247 industrial companies in Germany, Austria and Switzerland, 53% expected generative AI to have an impact in sales and marketing.

• Quotes out sooner, because prices, articles and past quotes are already filled in when the salesperson opens the draft.
• Fewer errors from retyping, because article numbers, prices and delivery dates come straight from the ERP.
• Orders that arrive as email with a PDF go into the ERP without manual entry, checked by a person.
• A current CRM without extra typing: quotes, orders and follow-ups recorded as they happen.
• Experienced salespeople get time back for customers and complex quotes.

The assistant does the searching, copying and drafting between an enquiry and a sent quote, or between an emailed order and a sales order in the ERP, so nobody has to jump between programs to put one quote together.

Public German examples show the pattern. Cloudsail did not build them. For emailed orders, the Zentis case described on our document processing page shows what automatic order entry can reach. The vendor Nemox reported in 2026 (Logistik Heute) that at the air freight forwarder Cargo Movers its assistant drafts quotes in Outlook, staff check and confirm them, 54% of quotes went to the customer unchanged and 17% with one correction. In vendor material shown in a 2026 BIHK webinar, the steel wholesaler BUHLMANN captures enquiries by phone, email or PDF, the AI suggests materials, and sales staff approve the quote or allow automatic sending by criteria such as quote volume.

• Quote preparation: reads the enquiry from Outlook, a web form or a PDF, matches requested items to your article master, pulls prices, discounts and delivery times from SAP or your ERP, and drafts the quote in your template.
• Similar past quotes: finds what this customer, or similar customers, were offered before, at which price and on which terms, so the draft starts from your own practice.
• Order entry: extracts customer, items, quantities and dates from emailed orders and PDFs, checks them against the ERP and creates a sales order draft for a person to release.
• Deviations flagged: prices that differ from the price list, unknown article numbers, missing delivery addresses or quantities outside the usual range.
• CRM upkeep: records quotes, orders and email contacts in Salesforce, Microsoft Dynamics or HubSpot, and suggests the next follow-up.
• Enquiry triage: sorts incoming enquiries by product line, customer and value, and routes them to the right salesperson.

Prices and delivery promises leave the company in writing, so a person approves every quote and order above the thresholds you set. Those thresholds are the most important design decision, and they are yours.

• Approval thresholds by quote value, discount, margin, customer group or product line, like the weight threshold at Cargo Movers or BUHLMANN's criteria such as quote volume. Above a threshold, a salesperson or sales manager approves. Below it, the assistant may send on its own once tests show it is reliable for that type.
• Special prices, non-standard discounts and payment terms: always a person.
• Quotes that need engineering judgement, such as special machines or custom designs: the assistant gathers past projects and data, and the engineer decides.
• Release of sales orders into the ERP: a person confirms, unless you approve automatic release for a tested order type.
• Credit decisions: checking the creditworthiness of private individuals (credit scoring) is a high-risk use under the AI Act, with extra duties for the company using it, so the assistant does not make or rank such decisions.

The assistant reads prices and master data live from your ERP, learns your practice from past quotes, and runs where your pricing data is allowed to be.

Check what you already own first. SAP S/4HANA Cloud already creates sales order requests from a PDF or image, suggests field values and simulates price or item deviations before conversion, as the SAP partner PIKON described in 2025. Specialised order-entry products exist, such as the one Zentis uses, and CRM vendors add AI features to their own tools. If your orders arrive in one format into a current SAP release, the standard feature may be enough. A custom assistant is justified when you run mixed or older ERPs, when quotes need data from several systems and past projects, or when quotes need engineering judgement that field suggestions do not cover.

• ERP connection: SAP S/4HANA, SAP ECC or another ERP read through its standard interfaces for prices, conditions, stock and customer data. Sales order drafts are written only through an approved, logged step.
• Past quotes and orders: your quote archive from the CRM, SharePoint or the document management system, such as d.velop or ELO, indexed by customer, product and outcome, so the assistant finds comparable cases.
• Email and documents: Outlook or Exchange mailboxes and PDF orders read with extraction tested field by field on your own documents. Orders that already arrive by EDI stay on EDI.
• Pricing data on your own servers or in your own cloud account: price lists, margins and customer terms are among your most sensitive data, so they need not go to an external AI service. An open-weight model, a model whose weights you can run yourself, can run in your data centre.
• Every value traced: each price, article number and date in the draft links to the ERP record or past quote it came from.
• One LLM gateway, a single controlled access point to all models with keys, budgets and logs, so the cost per quote and per order is visible.

Before go-live the assistant runs an acceptance test on a set of your real past enquiries and orders, where the correct quote or sales order is already known. You set the pass threshold per quote or order type before the test, and only the types that pass go live.

In operation the same tests run after every change, and we report quotes and orders handled, the share sent unchanged, corrections and cost per case. We measure hours per quote and per order before and after, on your own cases. Computerwoche named "Der Chatbot schreibt Angebote" among the most common AI failures in the Mittelstand in 2026, which is why nothing reaches a customer untested.

• Field accuracy per document type: customer, article numbers, quantities, prices and dates compared with what your staff entered.
• Quote quality: the share of drafts your salespeople would send unchanged, with one correction, or not at all.
• Price and term checks: no price, discount or delivery date that the ERP does not support.
• Threshold behaviour: every case above a threshold goes to a person, every time.
• No duplicate orders: a re-sent email or a retried step never creates a second sales order.

Quoting and order entry are not high-risk uses under the AI Act. The duties that matter here are data protection for the customer contacts in your CRM and the protection of your pricing data. Credit scoring of private individuals is high-risk and stays out of scope. If the assistant talks to customers directly, for example in a quote request chat on your website, the transparency duty in Article 50 of the AI Act applies from 2 August 2026, and the customer must be told they are dealing with an AI.

We give your data protection officer the technical facts: which personal data of customer contacts the assistant reads from the CRM, where it is processed, what is logged and for how long, and which model provider, if any, sees it. With pricing data on your own servers, no price list or margin leaves the company. You decide, advised by your data protection officer and lawyers. This is not legal advice.

Logs that show which salesperson approved or changed which quote can be capable of monitoring performance, so the works council is involved under § 87 (1) no. 6 BetrVG. We use pseudonymised IDs and agreed retention, and document the design for the works agreement.

Start with the document type that costs the most typing and has the clearest right answer: usually emailed orders in a recurring format, or standard quotes from the price list.

• In a free 30-minute first call we look at your channels, systems and volumes, without access to systems or data.
• A sample of past enquiries, quotes and orders shows which types the assistant can handle and what the correct result was.
• We agree acceptance thresholds and approval rules, then test on the past cases before anything reaches a customer.
• The assistant first works alongside your team, then takes over type by type as tests and live numbers allow.
• Each step has a fixed price, agreed in writing after the first call.