Scaling A.I. Hardware: A Hedge Play

People are bad at knowing what they want in new product. So how, then, do we scale expensive and long-lead hardware solutions for new products in uncertain markets? At Athelas we’re challenging the need to scale manufacturing through layers of process-limitation, choosing instead to focus on user and function. Our iterative methodology is powered by experimentation, consolidation, off-the-shelf integration, and a bit of Voodoo . “If I had asked people what they wanted, they would have said faster horses.” — Henry Ford. Source: Creative Commons Misconceptions Our viewpoint on product is often narrow and iterative: focusing on issues with something that already exists and improving the product concept within that same context. If a new product is to have staying power and isn’t an iterative improvement that ultimately ends in price and feature competition, it needs to change what we want and how we want it. Since you can ask all the market questions you want and still flop like the recent story, what resources do you commit to a product that is almost certainly wrong? How much weight do you put behind what features? And how do you know when it’s good enough? In other words how do you manage massive uncertainty with conventional project management tools: The following is how is finding product-market fit in the diagnostic blood testing space through maximizing the number of product-iterations through driving hardware iteration time into Agile two-week sprints. Juicero juicer scope, budget, and schedule. Athelas One of these costs $600 to purchase. Source: Bloomberg Iteration Limiting Factor: Mechanical Engineering Since users need to see a product to know if it makes sense, a startup optimizes its chance of survival by controlling burn while enabling as many product-iterations as possible. This is done through minimizing iteration time and maximizing learning from each product experiment, the model of . Lean Startup Build, Measure, Learn Custom mechanical components will generally control the length of an iteration loop for hardware products. The time required to design, mock-up, solicit vendors, agree to terms, order material, manufacture, inspect, integrate, test, refine, and repeat, can take months based on complexity. The worst part is that if you have scoped the application or solution wrong, which you most likely have, the subsequent application pivot erases the effort, leaving you with lessons for next time. The time required to design, mock-up, solicit vendors, agree to terms, order material, manufacture, inspect, integrate, test, refine, and repeat, can take months based on complexity. Source: 52weekturnaround.com Playing Catch-Up The way for mechanical engineering to gain ground on electrical and software is to focus on the : cell phone components, actuation systems, and fasteners, and . This product test can run at odds with the engineering tendency to look ahead to material and process informing design or Design for Manufacturing (DFM): optimizing your architecture based on the technical and economic constraints of these two areas. The effort is merited, but I would argue only after product-market fit is confirmed, unless critical to the core value proposition. However, if you’ve optimized a product that isn’t accepted that forethought becomes wasted time and effort. Realizing the product opportunity first, the later optimization effort will be rewarded in its own right with revenue growth and eventually profit. integration of highly-engineered mass manufactured components tie the system together with processes that enable quick-customization The inclination to complete DFM early is that manufacturing cost, which partially summarizes risk and complexity, kills scalability. Athelas: In-Home Blood Testing An Athelas Challenge: Scaling Custom Housings At , a challenge was tabled for our first 10x manufacturing output increase: Athelas Do we really need to plan, design, and execute a solution around the various limitations of traditional manufacturing: subtractive methods including machining, etc? Could we delay or eliminate all together this body of work and move directly to transformative (molding) processes at scale? The driving force of our build project was to broadly distribute product to collect as much feedback as possible and model trends, quickly. The thought of spending multiple-days or weeks on Design for Manufacturing that would likely be obsolete in a month, although accepted practice, seemed redundant. Having done extensive prototyping by this point, our custom housings conformed with standard additive manufacturing terms: a few hundred dollars per part with a two-day turn. At this price, the device would hardly be a consumer play unless we decided to significantly bleed on each sale until we reached scale. Additive manufacturing is the standard quick-customization process for prototyping and low-volume manufacturing, but it is not great at speed, cost, and often resolution. Segment of Voodoo Manufacturing’s Production Floor (used with permission) Enter . Running a hedge on my subtractive path, I explored companies advancing high-speed and high-volume printing techniques. Founded in the ocean between low-volume custom components and high-volume manufacturing, Voodoo is taking a sly approach to making custom products in a scalable, on-demand fashion. After the collapse of additive manufacturing in 2014, and , former MakerBot Product Managers, realized the issue with additive is that $200,000 industrial printers don’t offer much except better reliability when compared with the hobby-market MakerBot at $2,000. Then came the realization that for one $200k industrial printer, you can establish a factory of 100 MakerBots and turn out 100x the product while hedging on reliability by leveraging your excess capacity. Voodoo Manufacturing Jonathan Schwartz Max Friefeld Source: Voodoo Manufacturing’s Project Skywalker (used with permission) Then came the realization that for one $200k industrial printer, you can establish a factory of 100 MakerBots and turn out 100x the product while hedging on reliability by leveraging your excess capacity. Why you Need a Little Voodoo to Hit Scale Voodoo offered Athelas the ability to forget about process optimization so we could prove out our verticals before locking in our designs. Producing our custom enclosures for and the ability to deliver over was a clear 10x, the bar for a win at Athelas. As we scale with Voodoo, we can learn more about what we’ve overlooked in our market, model a solution, print in-house and test same-day, and fulfill a bulk order in two-days for user deployment. This is the first time I am aware of a startup being able to match the fulfillment capacity of a medium-size OEM, but without needing to finance large capital equipment or pay a large margin to a production partner. The ultimate benefit for Athelas is that by hedging on our internal methods, we never need to hedge or compromise on our users from commitments made to process optimization, tooling, and inventory. 80% less than those produced in an industrial printer 100 units in as little as 48 hours instead of 7 days “Run and Gun” on the Athelas Test Bench. Voodoo offered Athelas the ability to forget about process optimization so we could prove out our verticals before locking in our designs. A Final Thought Custom mechanical components will generally define the speed of product iteration. Additive process offerings have struggled as a scalable solution to mechanical design until Voodoo Manufacturing. Leveraging some “Voodoo” in our process allowed Athelas to achieve an 80% cost reduction on custom housings and find a fulfillment capability of 100s of units in as little as 48 hours. The immediate benefit to Athelas is the elimination of Design for Manufacturing activities until product vertical requirements can become well-defined, an advantage for both Agile integration and customer focus. The core advantage, however, is the ability to quickly and broadly deploy product for data collection without cost being a primary factor. The topic for next time? This is my first post in what I plan to be quarterly insight into how we hack scale-up to drive at 10x growth. I’d love to hear your thoughts and make this a conversation. If you made it this far, thanks! About Athelas Athelas (Y-Combinator ’16) is democratizing blood testing for prediction and peace-of-mind, applied at point-of-care and the home. Packaged as an affordable IoT device, the Athelas platform is enabled by a novel fusion of deep learning and advanced optics. Check us out at and be part of the conversation or continue reading: Athelas.com _At Athelas, we use Convolutional Neural Networks(CNNs) for a lot more than just classification! In this post, we’ll see…_blog.athelas.com A Brief History of CNNs in Image Segmentation: From R-CNN to Mask R-CNN About Voodoo Manufacturing Voodoo Manufacturing (Y-Combinator ’16) is a Brooklyn-based startup that’s on a mission to change the way we manufacture. Learn more at or continue reading: voodoomfg.com _A manufacturing startup’s journey to cut costs and take on injection molding. Co-written with Max Friefeld._medium.com How We’re Building a Robotic 3D Printing Factory