Categories
Big Data and AI cost model Logistics

How can AI be used to cost furniture?

Machine learning and AI are being applied everywhere. Why not use it to work out how much it costs to make bespoke furniture?

Please subscribe to the Cost of Everything Podcast on Apple podcast or Spotify. You can also find me on Twitter @costofpodcast and on LinkedIn.

How and why would you want to use Artificial Intelligence to estimate the cost of furniture? In this post we discuss some interesting research about the application of machine learning to early stage estimating of bespoke furniture.

On the go? Why not listen to podcast episode where I discuss the article (click this link for the episode). You can listen on most platforms!

The Value of a Library The Cost of Everything

In honour of libraries week, which is taking place between the 4th and the 10th of October here in the UK this episode is all about the cost and value of a library. Facts and Figures from the Reading Agency: Link The Download link to study by Kantor: Link Previous episode with Doug Hicks: The model is more important than data – The Cost of Everything | Podcast on SpotifyPrevious episode covering QALY: Fight Club and the CBA of Covid-19 – The Cost of Everything | Podcast on Spotify You can contact me or send me recommendations on what you would like to see on future episodes, via the LinkedIn page here, or by email: podcast@thecostofeverything.net.Checkout the website for the latest blog: thecostofeverything.net/Follow the podcast on Twitter @costofpodcast
  1. The Value of a Library
  2. Fight Club and the CBA of Covid-19
  3. New New York Rail Tunnel, Nigerian Project Risk and Upcoming Cost Events
  4. Learn how to cost your product
  5. The Cost of the Olympics

Why do you need to estimate furniture costs?

Furniture is one of those areas where everyone wants something fashionable, but unique pieces are often desired more and have more value. Mass customization is a trend that is becoming more accessible with automation and better technology.

The researchers are trying to use machine learning on historical cost data to identify the likely cost of custom furniture at an early or preliminary design stage. So the designs are different, but essentially the components, the likely production processes and labor are about the same. But, how do you get an accurate estimate at the early stage, remembering that generally the information available at an early stage in the production lifecycle is limited? And why do you need to be accurate?

If you are a customer buying a made to order piece of furniture that you’ve ordered online, you want to know how much it will cost, at the click of a button. You do not want someone from the furniture manufacturer coming back to you later saying that the costs have gone up. That is not good for business, so the manufacturer also wants to have an accurate picture of the cost to produce the furniture. It helps them to plan the whole enterprise.

What is Cost Estimating?

The article states that estimation is about predicting cost in terms of direct costs such as labor, materials and sales as well as indirect costs like overheads. The basis for the rest of the paper is the premise that the historical data can be used to predict at an early stage the cost of future products using machine learning techniques.

The paper argues that using traditional estimating, takeoffs from the drawings, understanding the process times for each steps and the associated labour, with an overhead takes too long. It also requires a lot of historical data to be available. And you need competent people to be able to draw out these costs accurately and quickly.

Understanding cost drivers and the associated features of past products that generate dependent variables on cost are needed. Although, quick and dirty (back of a cigarette packet) type estimates can be generated if the person or people pulling the numbers together at the early stage are really experienced, and know where to look for historical data that reflects the custom product which has been requested.

If you have the data, you can do parametric estimating, where statistical relationships are developed between the historical cost data and product features. Examples include the use of regression analysis or artificial neural networks. There are lots of sources online and through organisations like the International Cost Estimating and Analysis Association which provide detailed description of techniques for analysing historical data. A summary of these techniques will be on the website the costofeverything.net soon.

What is Machine Learning?

I have produced several episodes on the podcast which discuss big data and machine learning. With David Porter (Link), Sanathan Rajagopal (Link), and Ricardo Valerdi (Link). .

In terms of machine learning, the researchers, rather than attempting to use image recognition algorithms to train the model to identify features of custom furniture, instead use the numeric data. The data they use includes the process operations type and time, labor usage data, production time, and batch size. They then associated this data with previous items of furniture produced in terms of dimensions and material.

The learning algorithm bases the cost estimates on the material data (as a key dependent variable). In the research they used a variety of statistical based methods (such as linear regression and random forest) on the data for around a 1000 products from the previous 5 years.

By using these parameters, essentially the algorithm can “piggy back” onto the understanding that already exists of the likely manufacturing processes to be used for the product. The data they have extracted from the relationship between these processes and historical cost are then applied to the new product.

Is this Machine Learning?

If it was a machine learning algorithm I would expect it to be able to pick up when, for example a new manfucaturing route has been developed and what the new cost drivers might be. This might not be linked at all to the dependent variable they selected as the basis for future cost estimates.

I agree that drawing statistically valid relationships using data from the previous 5 years should get you to some idea of how much a new order will cost at an early stage, this does not seem new. That isn’t really the challenge of early estimating. If you already know what the product will look like, because those requirements will no longer change (the customer has placed an order, the manufacturer knows the processes to produce it), the estimate can be derived fairly easily.

There is definitely a time saving from having a system that enables data to be retrievd, calculated and presented quickly. Which I think is what this paper offers. I didn’t see anything in the paper about assessing the cost of rework, or whether the company that produces customised furniture have a framework of options they give to the customer (for example, the way IKEA have a set number of products but have a lot more options for combinations of those components – mass customisation, but based on a framework of a few parts they produce).

My main question about applying this approach to the case is, with the time effort and resource to develop and implement the system for a small or medium sized manufacturer of furntiure – is it worth it compared with their already established methods?

My opinion is that the methods applied in this paper are fairly established, maybe not widely, but they exist in industry. It is a unique application of the method, definitely, but it is not a novel technique.

References

O. Kurasova, V. Marcinkevičius, V. Medvedev, and B. Mikulskienė (2021). Early Cost Estimation in Customized Furniture Manufacturing Using Machine Learning, International Journal of Machine Learning and Computing, Vol. 11, No. 1

Follow the podcast on Twitter @costofpodcast

Join the LinkedIn page: Link

Share this post:

Leave a Reply