44. Maya Gupta and the history of Artifact Puzzles

Artifact Part 1 of 3: An essay about Maya Gupta, the Artifact Puzzle company's founder, the history of the company, its Ecru line, and the Hoefnagel Puzzle Club (about 4400 words; 4 photos)

Maya Gupta image source

This is the first of a 3 part series of postings about the Artifact Puzzle company. This one is an essay about the history of the company, focusing on its founder Maya Gupta. The second (to be published in about 5 days) will be an in-depth review of the company’s puzzles, and the third, 10 days from now, will have walkthroughs of my assembly of the puzzles that informed that review (under a spoiler alert) as well as research about their images.

My research for today’s essay included finding articles about Artifact and its founder online and a fair amount of back-and-forth correspondence between me and Maya Gupta. She expressed herself so well that instead of rewording her perspectives and information I will let you read many of her statements in her own words (sometimes lightly edited for context). Except where noted, the quotations come from her correspondence with me.

Artifact Puzzles was launched 2009, making it one of the oldest laser-cutting wood jigsaw companies. It was founded as an intended side-gig hobby by Maya Gupta, who at the time was an Associate Professor in the Electrical Engineering program at the University of Washington (UW) in Seattle.

But I can’t begin the story there. I need to go back to when Maya was a young girl. She saw the movie Star Wars and became fascinated with the light sabers. That plus something that her mother said led to a change in what young Maya wanted to be when she grew up.

Me: Do you enjoy being a professional craftsperson who makes “artsy” puzzles?

Maya: Ha! I love experimenting, and I love ideas (artistic and otherwise), but personally I’m actually terrible at the handiwork needed for arts-and-crafts, and I’m thankful we’ve been able to grow a company of puzzle-makers who are much better than me at the actual craftsmanship. As my mother told me when I announced as a kid that I wanted to be a carpenter when I grew up, “Maya, no, there are other people who are going to be much better carpenters than you. Find something else to do.”

Of course, she had other childhood interests as well, such as assembling jigsaw puzzles, playing poker, and music. I don’t know what other career paths she considered as a child but by the time she was a teenager she was settling into Electrical Engineering.

Maya at age 15. (image provided by Maya; cropped)

Me: In one article you say that your choice of specializing in Electrical Engineering was influenced by an early interest in lasers. I’d like to know at what age you got interested in lasers, and what was it about lasers that attracted you to EE.

Maya: Me and I think every other kid got our early exposure to lasers from Star Wars! But more seriously when I was around 16 I started studying electromagnetism and it just seemed super magical. I loved that the way you learned about it was through the math. (I'm a big math geek.)

Maya began her college studies at Rice University, where she earned both a Bachelor of Arts in Economics and a Bachelor of Science in Electrical Engineering.

It turned out I was better at the big ideas of information than light [itself] and I ended up specializing on the signal processing and AI side of electrical engineering. I managed to marry the two areas during a summer at Rice working on extracting information from Terahertz echoes; terahertz waves are longer than infrared but shorter than microwaves.

She then went on to the Stanford University School of Engineering where she earned her PhD in 2003. But by then her specialty had moved away from light and electromagnetism and focused on the machine learning and artificial intelligence (AI) side of things. How machines learn, and then apply what they learn as artificial intelligence, is now in the news every day, but back then it was an obscure specialty of theory and research. Her PhD dissertation developed a practical application of machine learning theory called linear interpolation using the principle of maximum entropy.

After receiving her PhD Maya was hired as a professor in Electrical Engineering at the University of Washington where she devoted all of her time to the academic treadmill. There she expanded on her research and grew an AI research team that focused on how to make good predictions from noisy signals (like interpreting the location and type of a boat from underwater engine noise from a few miles away).

In the academic world she developed an impressive list of peer-reviewed academic publications. In 2007 Maya received a Presidential Early Career Award for Scientists and Engineers, an Office of Naval Research Young Investigator Program award, and a departmental Outstanding Teaching Award.

The career-focused Maya had left her childhood interests behind but they hadn’t been completely forgotten. As one can imagine, although her academic career was proceeding well her work-live balance had suffered. By 2008 as she approached earning tenure at UW she began to think it was time to take up a hobby again. In 2011 she told a UW student reporter:

I needed a hobby that would be compelling, engaging enough that I couldn’t just ignore it, and playing a musical instrument is too easy not to do. (source)

Then, through serendipity, it came to her:

I hadn’t done a jigsaw puzzle in years, but I was at the airport heading to Hong Kong and I picked up a puzzle as a gift for a friend’s kids. But then the 14-hour flight was so long and boring I ended up doing the puzzle myself! It was so fun I looked into doing more jigsaw puzzles, but couldn’t find any images I liked, so I decided to start my own jigsaw puzzle company.

Around that time, laser cutters were pretty new, and starting a puzzle company seemed like a great excuse to buy myself a $10,000 laser. (source)

Actually, it seems to me that her childhood interest that might have been most influential in this decision might have been poker. What could be a more immersive gamble than starting your own business? And Maya was becoming an expert at understanding gambling. In her academic literature she describes AI as teaching computers how to gamble about finding the right answers from their training information sources. She told attendees at a machine learning seminar at Georgia Tech:

I loved poker as a kid, because it was all about making good guesses and probability, and that’s really what machine learning is about too: teaching computers to gamble well. (source)

I didn’t want to pester Maya with questions about all of the work that had to be done to start up the company. I give some insight into that stage of launching a new puzzle-making business in my essays about StumpCraft, Rocky Mountain Creators, and Puzzly UK. But there would have been a lot of decisions that had to be made and implemented, money to be spent, problems to solve, and trial-and-error lessons to be learned before she posted the Artifact website in 2009 and began selling puzzles.

That 2011 article in the UW News traced the founding, early growth and raison d'être for the company. As the student reporter describes it:

Her good friend Maria Berg soon joined Gupta in the fledgling business. “We got things rolling and it sort of took on a life of its own.”

They wanted to push the boundaries of the traditional jigsaw puzzle, Gupta said, “so that every puzzle would be kind of different, and the cuts would be different.” She said it’s easy to cut a puzzle where all the pieces look “sort of noisy and jaggedy.” They try to make puzzles that are elegant, even beautiful, in design, and challenging to the solver. “You also want something that sort of surprises the do-er,” Gupta said, “so that they have this sort of Gestalt moment where they’re like, ‘Oh, I get it. I get what’s going on!”

The reporter asked Maya what I think was a very thoughtful question: “Is there perhaps an inner connection between your professional work and the design elegance you work to achieve in puzzles?” Maya’s response was:

I don’t know that there’s a direct connection, but a lot of beauty and, let’s say aesthetics, crop up in signal processing. … It’s the same sort of design problem. An engineering problem that feels beautiful, that really feels elegant, is simple. Often the best-engineered systems and the best natural systems have a lot of symmetry.

Thus, Artifact was launched as a small business from her home in Seattle – a hobby side-gig just as she had planned. In fact, she told the student reporter that it was intended as an experiment in the economics of micro-manufacturing. Perhaps that was a return to an earlier interest from when she had earned her BA in Economics in parallel with her first Engineering degree. When asked about the future prospects for the business Maya said:

I think that we’ll grow – we’ll double a couple more times. But I think it will always remain a fairly small company. This isn’t something that’s going to turn into a multimillion dollar enterprise – it’s just too small a market.

Making and selling wooden jigsaw puzzles is a labour-intensive hands-on process. Aside from the computer-controlled laser-cutting machine and the automated process for taking online orders and generating the shipping documentation, making and selling laser-cut wooden jigsaw puzzles today is pretty much like it was for hand-cut puzzles over 100 years ago.

Here is how Maya’s long-time Head of Operations Jorge Rodriguez describes how Artifact operates now, but it would have been the same for Maya and Maria back in those early years, just on a much smaller scale:

Everything we do besides the laser cutting is done by human hands. We prep your puzzle for cutting, we clean your puzzle, separate each piece, pack it up and place all labels and stickers, manually pack your puzzle and prep to ship. We want the time and love we put into each puzzle to come through from the moment you get your puzzle in the mail and see the care we took into making sure it arrives safely.

Initially Maya and Maria ran the company out of Maya’s home but as the business grew they had to move it to Maria’s more spacious one. All of this launching and running of a new business was while she was still doing her day-job as a university professor leading a team doing widely cited research in the field of AI.

In August 2012 that led to a change in Maya’s career path: She was recruited to participate in Google’s machine learning research and development in the Silicon Valley. (Remember, this was at a time when for the rest of us artificial intelligence was still just the stuff of science fiction.) Initially, this was to be a one-year sabbatical gig, but Google asked her to stay to initiate and lead their “Glassbox” machine learning development team focused on more controllable and reliable artificial intelligence.

That required Maya to move herself and her rapidly-growing Artifact company to the Silicon Valley. Maya invited Maria to accompany her but she chose to stay in Seattle doing her own “day job” as an author, and for a while, continuing to design puzzles from there. Maya would need to train up a new fabrication team in California.

The move had the serendipitous effect of facilitating further expansion of the company. The volume of business now required more cutting machines, more staff, and instead of being a home-based business it would operate out of rented industrial space in Menlo Park.

I’d say the key thing is execution. Anyone can go down to their local tech-shop or buy a cheap Chinese laser and laser-cut a decent 50 piece puzzle. It’s a lot harder to cleanly cut a 300 piece puzzle. But then laser-cutting a batch of 100 300-piece puzzles at a price that keeps you in business really takes a lot of fine-tuning and optimizing the details for every step of the process.

As the company continued to grow into a scale that can reasonably be described as a small factory and warehouse rather than a workshop, it moved again to a larger space in Fremont, across the San Francisco Bay.

Jumping ahead to 2019, after several years of continued growth and R&D, Artifact introduced the Ecru line of puzzles.

A few years ago we figured out our “Ecru” process for using laser-cutting puzzles on matte images (for less glare) without causing a lot of laser-scorching and smoking, but it costs us a lot more to make those puzzles. We agonized over whether we should just switch to the Ecru process for all our puzzles and charge more … but some quick customer tests showed plenty of them preferred the glossier Artifact prints, and no one likes higher prices!

So we decided to launch Ecru as a separate line of puzzles, sort-of like Toyota has Lexus :). Over time, we’ve improved our basic manufacturing process a little so that the quality difference between the Artifact and Ecru line isn’t as big, and we’ve been able to bring down the price on the Ecru line a little, so the key differentiator for Ecru now is the matte image.

As a consequence of the matte images, our Ecru puzzles also have to use a smaller color range so the matte image doesn’t get burnt, so there’s a lot of artworks we can’t make as Ecru which we release as Artifact Puzzles. And then because Ecru puzzles have to be more expensive, we tend to do more fine art for Ecru that appeals to our older puzzlers who are willing to spend a few extra bucks for matte images.

Me: Why was the new line named Ecru?

Maya: Ecru is a very natural color like unbleached linen or eggshells or light sand, and we wanted to capture that the Ecru puzzles are a bit rawer and more natural looking than the polished super-bright look of our standard Artifact Puzzles.

An ecru colour palette (image source)

That introduction of a new line and the more spacious new accommodation in Fremont meant that Artifact was well positioned to respond to both the challenges and opportunities that arose in next year.

In early 2020 the Covid pandemic began!

The “social distancing” restrictions that were initially introduced internationally created a huge demand for non-screen home entertainment. That sparked a new international jigsaw puzzling craze similar to the ones that had occurred in 1918 and in 1933 (which I have previously written about here, and here.)

As with the other jigsaw puzzle companies, in March 2020 orders began to pour in. All wooden jigsaw companies have an annual boom-bust cycle, with the vast majority of their sales occurring in the shopping period around Christmas. But this unexpected sudden growth both bigger and longer lasting than the predictable pre-Christmas rush.

For Maya personally this was a mixed blessing. Her company was booming but now it demanded her personal attention full time. Do you remember how the phase “supply chain problems” entered our vocabulary in early 2020? Sourcing adequate supplies of plywood, photographic paper, and the company’s distinctive wood and magnetic closure boxes, as well as obtaining more cutting machines, ventilators and other equipment, became time-consuming challenges. Before long, the pandemic brought another change to Artifact’s business environment: Competition emerged as many wooden puzzle start-up companies, domestic and world-wide and some of them well-funded, emerge to fill the booming demand for puzzles.

Artifact could no longer be a side-gig. She had to choose between leading her company through a time of great opportunity and challenge, or continuing to participate in Google’s cutting-edge artificial intelligence R&D. She chose Artifact and resigned from her job as Principle Scientist for Google’s Glassbox team.

Maya didn’t have to leave AI completely, but now, instead of Artifact, it became her side-gig. In 2021 she was welcomed back as an Affiliate Associate Professor in the University of Washington’s Electrical & Computer Engineering Department. She has continued to do “a lot” of AI research activities including remaining an action editor (the scholar who is responsible for managing the peer-review process for specific manuscripts and making the final publication decision) for the Journal of Machine Learning Research.

Believe it or not, during the years of pandemic upheaval Maya somehow managed to find time to start another related puzzle business. She recognized that although one of wooden puzzles’ main features is their durability, many and perhaps most people don’t want to assemble puzzles again or save them as family heirlooms. They want to test themselves with a puzzle’s challenge, succeed, and then move on to other ones.

For such people the cost of buying and the bother of selling make wood puzzles not just more expensive but also rather inconvenient compared to the much less expensive cardboard ones. So she founded the The Hoefnagel Puzzle Club (which unfortunately I cannot participate in because it is only available only for Americans due to high international shipping costs.)

Hoefnagel is a revival and technological updating of the wooden puzzle library format that first began in 1908 and thrived in the 1930s. Basically, it is a private subscription library in which people check out puzzles instead of books.

The Hoefnagel Puzzle Club logo; presumably a joke about the pace at which some of us (like me!) assemble puzzles, and a reference to the fact they the Club puzzles are usually delivered by “snail mail”.

During their membership members can receive as many wood puzzles as they can keep up with assembling. These are not just Artifact and Ecru puzzles. The club has (literally!) thousands of wooden puzzles in its library made by over 50 high-quality makers from around the world. They even have hand-cut puzzles, including ones made by some of the most expensive hand-cutters like Stave and Elms.

Hoefnagel members pay a fixed monthly, half-year, or yearly fee regardless of how many puzzles they receive, and also a flat fee (currently $10.95) to mail the pack to the next member in its queue. In their FAQ Hoefnagel explains: “We charge the shipping costs separately from the membership fee because some members go through a puzzle every few days, while others take a couple of months with each pack, so this is a fairer way to handle shipping costs.” Once someone is a member they can get club credits by selling Hoefnagel their used puzzles if they meet or exceed Hoefnagel’s quality and condition standards.

Maya used her computer science, math and organizational skills to develop a way to use the Post Office’s payment and tracking systems to aid in automating her own computer system to keep the puzzles in expeditious circulation and to optimize “the happiness of our members and keep shipping times and costs down.” It is based on “wishlists” that members create identifying the puzzles they would like to assemble. They do this by browsing through Hoefnagel’s extensive library of over 1000 puzzle-packs (usually comprised of two regular-size puzzles or one large-size one.) New puzzle-packs are created every month so it pays to keep one’s wishlist up-to-date.

Naturally, since Maya is an AI expert, as members add to their wishlists and set their priorities while browsing the system learns their preferences and can direct them to other available packs they might want to consider adding to their wishlist.

After a member is done with a pack (they don’t need to assemble both puzzles unless they want to) they repackage the puzzles in their mailing box and notify the company. Then Hoefnagel sends the member a prepaid mailing label for the outgoing shipment. As soon as the post office receives the package its tracking system notifies Hoefnagel. Then Maya’s purpose-built algorithm and its AI choses and sends the best replacement pack from the member’s wishlist.

Hoefnagel has enough members and puzzles in circulation that no one should ever wait more than a few days before the next high quality puzzles arrive (unless they have a small wishlist and need to wait for one of their choices to become available). Also, since it is to Hoefnagel’s advantage to have puzzles in circulation rather than returned to its own warehouse, sometimes the company sends members with long wishlists a free 2^nd pack. This eliminates the potential for having a gap between puzzle-packs. As will be mentioned in the upcoming Part 2 of this series (the Review), the system seems to be very popular and is running smoothly.

Also, the Covid turmoil brought Maya another opportunity (and challenge.) She had grown up in Seattle and longed to return to the Pacific Northwest. Since she had resigned from Google’s R&D she now had that opportunity. Maya had moved her company before, and she could do it again. In 2021, she established another Artifact/Ecru workshop in Port Townsend, Washington but after a brief time she moved the whole company to that town.

We completely closed the Fremont, CA puzzle factory at the end of 2022. We tried to run both locations for a while, but that was a painful lesson in the extra costs of logistics – did you know you can Uber a semi-truck? They do take a few days to show-up, rather than a few minutes.

Port Townsend is not an industrial suburb of Seattle or Tacoma. It is a beautiful town with Victorian architecture on the sparsely-settled Olympic Peninsula, a ferry ride and two hour drive away from the hustle and bustle of city life. Its population is only about 10,000 but it has a vibrant arts and culture scene and ready access to outdoor recreation opportunities.

Since PT is actually slightly closer to my home in Victoria than it is to Seattle I knew as an occasional tourist about that community’s charms, so it never occurred to me to ask Maya what drew her to Port Townsend. But local Port Townsend reporter did ask that question:

When asked why she decided to bring Artifact Puzzles to Port Townsend, Gupta shared her love of the Pacific Northwest and stated “weather was an important factor.” In addition, Gupta was attracted to PT because of its reputation as a “woodworking town” and said that while she loved Silicon Valley’s spirit of innovation, the company “feels like a better fit for Port Townsend with its culture of making beautiful things.” (source)

That reference to PT being a woodworking town would probably be based on it being the venue for the annual PT Woodworkers Show as well as the home of the Port Townsend School of Woodworking and the nearby Northwest School of Wooden Boatbuilding. (In a world where wooden boats have largely disappeared, more than 20% of the boats in Port Townsend marinas are still made of wood.) The reference to weather might be because, like Victoria, the town is in the rain-shadow of the Olympic Mountain Range and so it doesn’t get the precipitation for which Seattle is famous.

I imagine that the employees that moved to Port Townsend with the company are glad to have left the traffic and high housing prices of Fremont behind them for a beautiful and quiet town that is still only a ferry ride and two-hour commute from the urban amenities of Seattle.

That local reporter had better luck than I did at getting a sense of the current size of Artifact as a business. For this April 2024 article Maya told him they currently sell over 10,000 puzzles each year. Like all of the laser-cut puzzle-makers that kind of information is usually treated as a trade secret. Even jigsaw puzzling’s foremost jigsaw puzzle historian Dr. Anne Williams, Professor Emerita of Economics at Bates College and the author of The Jigsaw Puzzle: Piecing Together a History, can only guess at which is the largest American wood puzzle-maker:

Sales figures for the toy industry in general are hard to come by, and I have never seen any on wood jigsaw puzzles. My guess is that Stave (due to high prices) and Liberty (due to high unit volume) have the highest annual revenue in the U.S. I don't think I can venture a guess about the rest of the world. (source; recent personal communication)

I knew better than to ask for sales figures but here is all that I could get from Maya:

Me: It is hard to get a sense of the size of companies from their online presentation, but I would like to tell people about the scale of Artifact/Ecru. How many employees do you now have? How many laser cutting machines do you use?

Maya: Ha! That’s a hard question to answer as it changed a lot during the boom-then-bust years of Covid, and there are a lot of folks who we consider part of our team that choose to work part-time, or just seasonally, or here-and-there on puzzle designs. Let’s just say we are a tiny operation powered by dozens of great people.

“Social distancing” and demand for wooden jigsaw puzzles has subsided somewhat in the past three years, however without any data available to make such a claim I get the impression that demand for wooden jigsaw puzzles is settling in at a much higher level of demand than before the pandemic. That would mirror what happened after the previous 1908 and early 1930s jigsaw puzzle crazes.

Maya’s “experiment in the economics of micro-manufacturing” has emerged from the pandemic years as a thriving business somewhat larger than she had envisioned, and is probably second only to Liberty Puzzles as America’s largest wooden jigsaw puzzle company.

You know how sometimes you eat a big meal and you feel all full, but then dessert comes out, and you find you do still have room for dessert? That’s how I am with projects, and Artifact Puzzles has been my “dessert project” since I started it in 2009, back when I still ran a big research group at the University of Washington.

Then I moved to Google Research in 2012, and Artifact was still a fun on-the-side hobby business. But then in 2020 Covid caused a massive boom-and-bust cycle in the puzzle business as people first quarantined and puzzled, and then abandoned their living rooms. To cope with all that, we went from coming out with 5-10 new puzzles a year at Artifact to more like 50 puzzles a year, making Artifact a bit more of a full-time job for me ;) . . .

I’m now working on hiring an Art Director and empowering our senior people more, so I can take a step back and turn Artifact into more of a hobby than a workday.

Coming up:

#45 Artifact Part 2: An in-depth analysis and review of Artifact and Ecru Puzzles

• Introduction and evaluation standards

• Selection of images

• Selection of sizes and cutting designs

• Cutting designs

• Wood

• Printing

• Cutting execution

• Packaging, fulfillment and shipping

• Miscellaneous and conclusions