AI in the furniture industry: how automation helps with made-to-order furniture

Every made-to-measure furniture order is a small project of its own. A conversation with the customer, measuring up, a visualisation, a quotation, approval, the technical design, ordering materials, production, transport and installation – with dozens of emails, photos and files in between.

In many companies, this information circulates in mailboxes, messaging apps and Excel spreadsheets. The salesperson knows one thing, the designer another, and the production manager learns about a change in front colour when the boards have already been cut. So it is the office and the managers who carry the heaviest load, not the machinery.

Calculate it with your own data. Suppose a company handles 40 orders a month, and for each one the office spends 3 hours retyping data and chasing details. That is 120 hours a month, or about three quarters of a full-time job, before you count the cost of mistakes.

The first step is not about AI, but about order. An order should have one place where you can see its stage, the person responsible, the deadline, the documents and all changes. Only then do automation and AI have something to work with.

Such a place is often monday.com, a platform for managing tasks and processes as boards with columns, statuses and notifications. It works well where an order passes through many people and stages. The ERP system stays with what it does well: finance, inventory and invoices.

This is not the only route. Some companies build a similar workflow on SharePoint lists and in Power Apps, and some ERP software has extensive make-to-order production modules. The table below compares these options.

The most expensive mistakes in made-to-order furniture happen when quoting unusual jobs. The salesperson estimates the cost from memory, and when the job is settled, it turns out the margin has vanished on extra hours in the paint shop or more expensive fittings.

Here AI can suggest, not decide. Based on the order history, it finds a few of the most similar past jobs and shows what the materials and labour actually cost. The estimator gets a reference point and sets the price themselves.

There is one condition: the history must contain actual data, not just prices from quotations. If the company does not record labour time and material usage per order, it first has to start doing so. After a few months, there will be enough material for the suggestions to be reliable.

Every day, a furniture company receives emails and contact form messages: specific orders, requests for quotes, questions about delivery dates and enquiries outside its range. AI can sort them, extract contact details and dimensions, assign each enquiry to the right salesperson and prepare a draft reply.

Take a hypothetical company that receives 30 enquiries a day. If a salesperson spends 5 minutes reading, assessing and passing on each one, that is 150 minutes a day on sorting alone. AI can do this part of the work as a first pass, and a person only approves the decision.

If a chat on your website answers customers, from 2 August 2026 it must tell them that the conversation is with AI. With private customers, it is also wise to make sure the chat does not make promises about delivery dates or prices that the company has not confirmed.

Once the quotation is accepted, the data has to pass from design to production without distortion. Ideally, the furniture design software exports the list of parts and materials to a file that the system reads in without retyping. This is more reliable than having AI read dimensions off a drawing.

Where data arrives as scans or PDFs, such as specifications from an architect or supplier quotations, OCR helps. It is technology that turns an image of a document into text and numbers the system can work with. Combined with AI, it recognises what is a dimension, what is a colour and what is a fitting code.

A person always approves the extracted data before it reaches the shop floor. An error of one centimetre in a carcass dimension costs more than a minute spent checking. Only the approved specification triggers the next steps in the list below.

Many furniture company owners do not know exactly how much they earn on a given type of furniture, customer or sales channel. The data exists, but it is scattered across the ERP, spreadsheets and managers’ notes. Power BI, Microsoft’s tool for reports and charts, combines it into a single view that refreshes automatically.

The second useful view is production workload. All accepted orders on a timeline, with their stage, materials and crew, let the production manager plan several weeks ahead. The salesperson can then see what delivery date they can realistically promise a new customer.

A report is only as good as its data. You can calculate the margin on an order reliably only if labour time and material usage are recorded against the order, not estimated after the fact.

Transport damage, a dimension that does not fit, a chip in the lacquer – complaints are an everyday reality in furniture making. With private customers, the company has 14 days to respond. If it does not, the complaint is deemed accepted, so a lost email can cost you a whole new front.

Here, automation keeps track of deadlines and order. Each complaint gets a number, a response deadline and a person responsible, and the system sends a reminder before the 14 days are up. AI can produce an initial description of the complaint from the photos and the email, and prepare a draft reply.

A person decides whether to accept the complaint. Judging whether a scratch happened in transport or during installation takes experience and often an inspection, and a mistake has legal consequences.

The most common mistake is starting with the most interesting use of AI, such as automatic quoting, before the company has its data in order. The steps in the list below make sense in this order, because each one prepares the ground for the next.

Each stage ends with a check of the effect against figures measured before the start: time from enquiry to quotation, number of errors in specifications, time to respond to a complaint. The duration depends on the number of order stages and the systems to be connected.

A workshop that handles a few orders a month usually does not need an elaborate system. The owner knows every order anyway, and a well-kept spreadsheet and a shared calendar are enough.

Nor is it worth starting when every salesperson quotes and takes orders their own way. The system will only lock in that chaos. First you need to agree one way of working, and only then automate it.