Taking orders on WhatsApp with AI:from message or voice note to ERP
The order comes in as loose text, a photo of a list or a two-minute voice note, often at night, and the next day someone retypes it into the ERP. An AI agent takes it on the spot, checks it against your prices and inventory and leaves it entered. Here is how it works, what you need and how to measure it.
How do you take orders on WhatsApp with AI and get them into the ERP?
To take orders on WhatsApp with AI, you connect an AI agent to the official WhatsApp API, your ERP and your CRM. The agent understands the order even when it arrives messy, as a voice note or as a photo of a handwritten list, checks it against the catalog, inventory and that customer's price list, confirms quantities and delivery date, and enters the order in the ERP without anyone retyping it.
The difference from a menu bot is that customers write the way they write to their usual sales rep, with no numbered options or forms. Whatever the agent cannot settle on its own, like an order over the credit limit or a special product, it hands to a person with the full conversation. It is the core use case in four industries: Consumer Goods, Wholesale Distribution, Construction Materials and Agribusiness.
How orders arrive on WhatsApp, and who enters them
At a distributor, a brand or a building supply yard, orders rarely arrive tidy. They come in as loose text across several messages, as a photo of a handwritten list or as a voice note recorded from the car, and often at night or on the weekend, when nobody is answering.
Then someone reads it, interprets it and types it into the ERP, usually the next day. That manual step is the cost that never shows up in a report, and you pay it in five ways.
- Retyping: every order is written twice, once in the chat and once in the system, with the telesales hours that takes.
- Entry errors: a similar product code, a case instead of a unit, a price from another list. They end in credit notes, re-invoicing and arguments with the customer.
- Orders that wait: what comes in at 10 p.m. gets processed at 10 a.m. the next day and sometimes misses the delivery run.
- Repeat questions: price, stock and order status take up a good part of the sales team's day.
- History outside the CRM: conversations live on each rep's phone and leave with them.
What order taking looks like in each industry
Who orders and what needs checking changes, but the flow is always the same: understand, check, confirm and enter. The table sums up the four industries where taking orders on WhatsApp is the core use case, and each one has its own page with the details.
| Industry | Who orders and how | What the agent checks before entering | What you gain |
|---|---|---|---|
| Consumer Goods | Corner stores, grocers, convenience stores and bars that restock at night, between one visit from the field rep and the next | The channel price list, volume promotions and stock at the warehouse that serves the area | Orders that come in between visits and after hours, without adding field reps |
| Wholesale Distribution | Hardware stores, pharmacies and resellers that send lists by product code, EAN or description | The product among thousands of codes and cross-references, the price list for their tier and their account balance | Telesales freed from repeat questions, and quotes that go out first |
| Construction Materials | Building supply yards, builders and site foremen ordering for the job, often on Saturday for Monday | The price by channel and by project, the account balance, and the delivery date and unloading method | Fewer pricing errors and less re-invoicing with the customer |
| Agribusiness | Growers, cooperatives and farm input dealers ordering supplies in the middle of the season, first thing in the morning | The day's grain price, the payment terms (cash, grain swap or against the next harvest), the farm registration and the agronomic prescription | The whole planting window covered, at any hour |
Summary by industry. Each industry page covers the full picture of that sector.
What you need to automate orders on WhatsApp
An agent that takes orders is only as good as the data it checks. Before talking about technology, make sure these six pieces exist and that someone owns each one.
- A catalog with codes: every product with its ERP code, its unit of sale (unit, case, bundle or pallet) and the names your customers use for it, including brands and cross-references.
- Price lists: which list applies to each customer or channel, with volume discounts and current promotions.
- Business rules: minimum order, delivery zones and days, the cutoff time for the next delivery run, and what to do when an order goes over the credit limit.
- ERP access: reading stock, prices and account balances, and creating the sales order, through an API or a connector.
- The CRM and the number: a number connected to the official WhatsApp Business API, which is how the channel gets integrated with your systems, and Salesforce to log every conversation and order on the customer record.
- Handoff rules: who handles exceptions, during which hours and with what information.
Turnkey: with Sellium, our team handles the process mapping, training the agent on your catalog and the integration with your ERP. On your side, you need access to the data and someone who knows the business rules.
Six mistakes when automating order taking
Most projects that fail do not fail because of the AI: they fail because the agent has nothing to check against, or because nobody defined what happens when something does not add up.
- A menu bot: making the customer pick numbered options and search product by product. The one ordering thirty items goes back to the phone. We explain the difference in chatbot or AI agent.
- A messy catalog: duplicate codes, descriptions nobody uses and ambiguous units. The agent is only as accurate as the catalog it checks.
- Confirming without checking: promising stock or price without asking the ERP. The error shows up when the truck arrives short or the invoice carries a different price.
- Entering without a summary: before entering the order, the agent has to repeat what it understood (products, quantities, price and delivery) and wait for the customer's yes.
- An order left in an email or a spreadsheet: if someone still has to move it into the ERP, the retyping is still there in another form.
- Exceptions with no owner: credit limit exceeded, a special product or a complaint. With nobody in charge, the conversation is left hanging and the customer calls again.
The metrics to measure WhatsApp order taking
Measure before you start. Without a baseline of a few weeks, any improvement is an impression and not a number. These six metrics are enough to know whether the agent is working and where to adjust.
| Metric | What it measures | How to calculate it |
|---|---|---|
| Orders with no intervention | How much of the flow the agent handles on its own | Orders entered with no handoff over total WhatsApp orders |
| Time from message to order | End-to-end speed | Minutes between the customer's first message and the order being created in the ERP |
| After-hours orders | Demand that used to wait or get lost | Orders confirmed outside business hours, per week |
| Entry errors | Order quality | Credit notes and re-invoicing caused by a wrong code, quantity or price |
| Average order value | Whether the agent's suggestions add up | Revenue per order before and after, for the same customers |
| Telesales hours freed | Capacity that goes back to the team | Hours that used to go to typing orders and answering price, stock and status questions |
Agree on the baseline and how each metric is calculated before go-live, using ERP data rather than an estimate.
How an AI agent takes an order, from message to ERP
What used to go through three people and two systems, the agent handles in a single conversation, at any hour.
Identifies the customer
By their WhatsApp number, tax ID or account code, against the CRM and the ERP. It pulls their price list, history and terms without asking for forms.
Understands the order however it comes
Loose text, several messages in a row, a voice note or a photo of a handwritten list: it separates products, quantities and units, and resolves brands, synonyms and everyday names.
Checks catalog, stock and prices
It looks up every item in the catalog, checks stock in the ERP and applies that customer's price list with its promotions, minimums and available credit.
Confirms before entering
It sends back the summary with items, quantities, price, out-of-stock items with their alternative and delivery date. If something does not add up, it asks instead of guessing.
Enters it in the ERP and logs it in the CRM
The sales order goes into the ERP ready to pick and the conversation is logged in Salesforce, on the customer record. When it applies, it sends the payment link.
Reports status and hands off the special cases
It confirms the order number, answers status questions when the customer asks, and hands exceptions to a person with the full context.
Public data on ordering without a rep in the middle
Figures with a published source to size up the case. None of them is a Sellium result: every implementation is measured against the company's own baseline.
7x
More likely to qualify a lead when contacting it within the first hour than one hour later
1,000
Free service messages per WhatsApp number per month: these are the agent's replies when a customer writes in to order
67%
Success rate of AI bought from specialized vendors; internal builds reach one third of that rate
Source: MIT NANDA, The GenAI Divide (2025)
Caveats: the Gartner figure comes from a survey of 632 B2B buyers, and the same survey shows that for decisions that need context they prefer to talk to a rep; that is why the agent hands those cases off. The Harvard Business Review figure measures web leads contacted by phone or email, not WhatsApp orders, and is used as a reference for response speed. Meta updates its rates several times a year: check the current one in its documentation. The MIT NANDA figure measures generative AI projects in companies, not order-taking agents in particular.
Sellium: the agent that takes the order and puts it in your ERP
Sellium takes orders on WhatsApp over the official API, using your real catalog, price lists and inventory, and leaves them entered in your ERP and logged in Salesforce. It is delivered configured: it is not a platform you have to build on.
Sellium
AI sales agent for WhatsApp
Sellium is Vantegrate's AI sales agent for WhatsApp: it understands text, voice notes and photos of lists, quotes with your rules, takes the order, can collect payment with a payment link and logs everything in your CRM, around the clock.
Explore SelliumSecure AI agents
Sellium runs 100% on Salesforce and Oracle Cloud infrastructure, certified SOC 2 Type II and ISO 27001. Vantegrate enables, the customer operates and certifies.
See the security modelISV and Consulting PartnerRuns inside your Salesforce
Sellium reads and writes your objects, respects your profiles and permissions, and works alongside Agentforce and your existing flows.
Vantegrate and SalesforceFrequently asked questions about taking orders on WhatsApp with AI
What sales and IT teams ask before putting an agent in charge of taking orders.
Can it take orders from voice notes or photos?
Can it take orders from voice notes or photos?
Yes. Sellium understands voice notes the same way as text and also reads a photo of a handwritten list: it picks out products, quantities, units and delivery date, and if something is unclear, like a brand it cannot make out, illegible handwriting or an ambiguous quantity, it asks before entering the order.
What happens if a product is out of stock?
What happens if a product is out of stock?
The agent checks stock in the ERP before confirming, so the shortage shows up in the conversation and not when the truck arrives. What it does next depends on the rule you set: offer an equivalent product, drop the item from the order or leave it for the next delivery.
Does it integrate with my ERP?
Does it integrate with my ERP?
Yes. Sellium integrates with SAP, Oracle, Tango, Odoo, NetSuite and other ERPs to read stock, prices and account balances, and to create the sales order. If your ERP does not expose an API, how to connect it is worked out during process mapping, before the quote.
Does it follow each customer's price list?
Does it follow each customer's price list?
Yes. The agent identifies the customer and applies the price list that customer gets, with its volume discounts, promotions and minimum orders, exactly as they are in your ERP. It is the same logic it uses for quotes, which we explain in automated quotes on WhatsApp.
What about orders that come in after hours?
What about orders that come in after hours?
They get taken all the same: the agent works 24 hours a day, every day, and the order is entered so it goes into picking or the delivery run according to your cutoff time. Exceptions that need a person are logged with the full context so the team sees them when it opens.
How much does it cost to automate order taking with Sellium?
How much does it cost to automate order taking with Sellium?
Sellium is priced in two parts: an initial implementation quoted for each company and a monthly subscription with expandable AI credits, with no lock-in. You pay for WhatsApp messages directly to Meta, on your own account and with no markup from us. For the number for your case, request a quote with your order volume and your ERP.
See how Sellium would take your orders
Tell us how your orders come in, which ERP you use and how many arrive on WhatsApp. We will show you, with your catalog and your price lists, how Sellium would take them from the first message to the entered order.