AI quotation automation: from WhatsApp enquiry to quote in minutes

The short answer

An AI quotation agent reads an enquiry that arrives on WhatsApp, email or a web form, works out what the customer wants, looks up prices and stock from your own pricing sheet, and prepares a draft estimate in your accounting tool. A person checks it and sends it. The aim is not to remove your sales team from the loop. It is to remove the 20 minutes of searching, copying and calculating before they can make a decision.

We cover the build in more detail on the AI quotation agent page. This article explains why quoting is slow, how the pipeline works and how to start without risking a wrong price reaching a customer.

Why quoting is slow

For distributors, traders and other B2B sellers in India, a quote is rarely a single price lookup. A typical request touches several sources at once:

• Pricing sheets. Rates sit in Excel files, often with different columns per customer group or city, and several versions in circulation.
• Customer-specific terms. Rates, minimum quantities and discounts differ by customer group, and the rules live in people's heads.
• Stock and availability checks. Someone has to confirm that the quantity requested is in stock or can be supplied.
• Taxes and charges. GST and any freight or packing charges are often worked out by hand.
• Approvals. Large discounts, new customers or unusual items need a manager, who may be on the road.
• Messy input. Customers write "10 nos 25mm PVC elbow, need Thursday" in a voice note, a photo of a handwritten list or a forwarded PDF.

Each step is small. Together they mean a customer waits hours, sometimes a day, and a faster competitor wins the order. Speed matters most in the first hour after an enquiry, which is when buyers are still comparing.

How an AI quotation agent works, step by step

1. Intake. The agent watches the channels you choose: a WhatsApp Business number, a shared mailbox or a web form. Voice notes are transcribed, and photos or PDFs are read with document AI.
2. Extraction. A language model pulls out the customer, items, quantities, required delivery timing and any special conditions. Ambiguous lines are flagged, not guessed. This is the kind of work generative AI does well.
3. Product matching. Free-text item names are matched to your catalogue codes using embeddings and your rules, as described in RFQ and product matching. Low-confidence matches are listed for a person to confirm.
4. Pricing from your sheet. The agent reads rates, minimum quantities and discount rules from your pricing sheet. It does not invent or estimate prices. The sheet stays the single source of truth.
5. Availability check. It checks stock levels and reports shortfalls, with substitutes you have approved.
6. Draft estimate. A draft quotation is created in Tally, Zoho Books or your ERP or CRM, with GST applied through your own tax settings, and a plain-language summary goes to the salesperson.
7. Human approval. A person reviews, edits and approves. Nothing reaches the customer before that, unless you later choose to auto-send low-risk, repeat orders.
8. Follow-up. The quote and its outcome are logged against the customer in your CRM, so the team can chase open quotes and see which ones convert.

What goes in and what comes out

Input	What the AI does	Output
WhatsApp text or voice note	Transcribes and extracts items and quantities	Structured request
Photo or PDF of a list	Reads it with document AI	Line items with confidence scores
Free-text item names	Matches to catalogue codes	Matched SKUs, doubtful ones flagged
Pricing sheet	Applies rates, discounts and tax	Priced lines and totals
Stock data	Checks availability	Available, short or substitute
Accounting tool or CRM	Creates a draft estimate	Estimate ready for approval

A worked example (hypothetical)

This is an illustration with assumed numbers, not a client result. Replace them with your own.

Suppose a distributor sends 600 quotes a month. A salesperson spends about 25 minutes per quote today, gathering the request, finding prices, checking stock, working out totals and typing the estimate. With a draft prepared by the agent, assume review and correction takes 6 minutes. Take the loaded cost of a staff minute as Rs 5.

Formula: monthly saving = quotes x (manual minutes - assisted minutes) x cost per minute.

Item	Calculation	Result
Manual effort today	600 x 25 min x Rs 5	Rs 75,000
Assisted effort	600 x 6 min x Rs 5	Rs 18,000
Gross monthly saving	600 x 19 min x Rs 5	Rs 57,000

Subtract the running cost of the system before you decide. The staff time is not the whole case either. If faster replies win even a few extra orders a month, the value of those orders may outweigh the time saved, but you can only know that by tracking win rate before and after. At 60 quotes a month the same arithmetic gives Rs 5,700, and a custom build is hard to justify. Volume and quote complexity decide the case.

What to automate first

Do not start with your hardest customer. Start with the quotes that are frequent, rule-based and low risk:

• Repeat customers ordering from a stable catalogue.
• Standard catalogue items with clear pricing rules.
• Requests that arrive in one dominant channel, usually WhatsApp.

Leave custom engineering quotes, tenders and one-off items for a person until the system has proved itself. If you want a ready map of where to begin, the AI readiness checklist helps you rate each workflow.

Risks and guardrails

• Wrong prices. Prices come only from the sheet through fixed lookup logic, never from the model's memory. Anything not found in the sheet is escalated, not estimated.
• Invented items. Models can produce plausible but wrong answers, known as hallucination. Matching against your catalogue and showing confidence scores keeps this contained.
• Discount leakage. Set discount limits per role. Anything above the limit goes to an approver.
• Stale data. If the pricing sheet or stock feed is out of date, so is the quote. Agree who owns each source and how often it changes.
• Customer data. WhatsApp conversations contain personal data. Confirm your obligations under the DPDP Act with your adviser, and decide what is stored and for how long.
• Audit trail. Keep the original message, the extracted fields, the draft and the approver for every quote, so disputes can be settled quickly.

How to get started

1. Collect 30 to 50 real enquiries from the last few months, including the messy ones, with the quotes you sent.
2. Clean up one pricing sheet and write down your pricing and discount rules in plain words.
3. Note how long a quote takes today and how often customers chase you.
4. Share these with us through the contact page.

We then prototype on your own enquiries and report how many lines the agent got right, and how many went to review, before you commit to a full build. Related reading: the business process automation service for connecting the quote to your accounting and stock systems, the WhatsApp AI assistant for the customer-facing side, and build vs buy AI for SMEs if you are weighing a ready-made tool.