This post was originally published by Alexandra Petrus at Medium [AI]
European (pre)-early stage AI startups in the limelight
As AI moves from research labs to commercial products, the dent technology leaves on society accelerates — agreed by mass. In the following three years, the contribution from AI will grow considerably. The European Union predicts that AI solutions will have an economic impact of $12 trillion by 2025. We’re witnessing AI technologies used by Johns and Janes in real-world products (in varying setups targeting B2B, B2B2C or B2C). The need to bridge the gap in the deep tech startups space is more real than a few years ago. People that introduce novel interpretations, to new or old problems, must know the intrinsics that AI can bring to a business. Generally valid, it doesn’t matter what AI technology you use, but if you are efficient and effective at using it, and picking the right problem you solve. What does this mean?
Most of the Programs out there are built with the mission to help companies leverage new tech (AI, ML) and augment existing product lines for efficiency or cost. There are also innovative programs which are out of reach to regular Joes that lack either the traction or the right geo. Or simply programs built with a focus to help AI startups, but require a different technology readiness level than what’s mostly achievable in real world conditions. Nothing wrong. Every effort gets to play a part in building up ecosystems and getting better, more cost effective or solutions out the door, faster. No program can meet all needs or fit all sizes.
But then AI should be for everyone, anywhere, anytime. A credo we share at BeAI.
And here we are, left with a few questions:
- How can we provide easy access to multidisciplinary know-how mandatory for all to gear up an AI-based business?
- How to start an AI-based side project, and iterating towards practical products that solve real world problems.
- How do we encourage discipline and have people build products addressing needs for the change they want to see in the world?
- How do we encourage more diversity into AI products and projects building? To help us design and develop ethical AI.
- How do we encourage integrity as a value in the world and make people aware that their voice and effort is their contribution. There is a serious danger that we are entering a world without values, where AI powers tools that will intervene in our emotional and biological systems, without our consent. It’s for us to avoid such scenarios.
We, as many, want to contribute and see the applied methodology of learning tough lessons the quick and dirty way. #BeAI has been catered to help the ecosystem make good progress on the above five quests, and help early stage deep tech startups give a try.
Real-world AI Programs in perspective
EU funds are one of the promising AI funding sources to rising startups in Europe. Programs like EIC or PathFinder are good ways to further fuel your endeavours. Ping Ecaterina Silova, if you need help applying for such AI funding programs with your deep tech startup.
“Statistically speaking, top applying countries for EU funding are from Western Europe. There aren’t too many Eastern Europeans countries applying — there are a lot of misconceptions around bureaucracy. In fact, the process to apply is straightforward — it takes about 3 months from the application date to get to the evaluation round, and after that it’s really easy to manage.[…] There are no specific pre-requisites for applying for the EIC accelerator other than Technology Readiness Level 6 and demonstrating your startup has secured financing in the past — this shows financial capacity for co-founding.” — Ecaterina Silova #knowledgepill from the #BeAI #preaccelerator Mentor Sessions.
Wondering what is the flip-side? To be able to access the funds for AI-driven startups, one must show a technology readiness level of six. To get a sense of the scale, a large part of the deep technology start-ups, which aim to disrupt a whole range of industries, typically have higher TRL*, but early stage ones span across TRLs three and four (and sometimes five and six) in their venture maturity. Doors open for greater access to capital, but at the cost of early investors (and even later ones) forgoing returns in favour of higher and safer TRL start-ups.
*Technology readiness levels (TRLs) measurement system was invented by NASA in 1989 to assess the maturity level of a particular technology. There are nine technology readiness levels, where TRL 1 is the lowest and TRL 9 the highest. BCG Analysis and Hello Tomorrow paint a good picture in understanding TRLs in context to deep tech startups — see the image below.
This is where an AI Pre-Accelerator comes in to bridge and help scale up. This is why #BeAI is doing something about the gap and started its #BeAI Pre-Accelerator for early stage deep tech startups in Europe and nearby. 70+ applications, 9 countries, all in under 30 days. The needs and opportunities are immense in the space.
The need of many, in a community for all — #BeAI
A wise man (thank you, Guy Kawasaki) once said that if you can get a community around, then start building products for it, as you may have found yourself a good space to be in. The #BeAI team builds programs and meets the needs of its fast-moving and growing community. Focused on education and early business acceleration. It now runs an AI Pre-Accelerator, tomorrow partners in educating 1% of the local space in the basics of AI, and maybe, after that, it will build the first early stage deep tech angel investing platform. Everything is built from the needs of many, in a community for all.
Intrinsics brought by AI in a startup / business — 10 segments view
- From Online to Deep Tech
Will your company/ startup/ business be much more valuable and/or much more effective if it were good at AI? Most likely. There are ten segments where AI brings a particularity when compared to the traditional perspective. It’s mostly the same, but slightly different effort. 🙂 Just think it this way: software or Internet-first companies used to have a few things in common — like A/B testing, short iteration times, and decision making being pushed to engineers & PMs.
Now, deep tech or AI startups demand airtight alignment between different teams, such as strategy, legal, investment, product-market fit, team skills, marcom, packaging, customer readiness, pitch and development. Let’s check each segment for some intrinsic AI novelty in approach.
“AI is very hyped right now. If you don’t know AI, you believe it can do some impossible things. We have to get back to a more realistic set of expectations.” — Christian Merkwirth #knowledgepill from the #BeAI #preaccelerator Mentor Sessions
LEGAL > Having the right amount of data, crunching the important signals and dropping the outliers, having your customer information protected, all of these are under the umbrella of data readiness and it’s something that will determine your propensity to succeed, as a deep tech startup. The mere use of AI in a business has policy, regulatory and ethical implications that depend on the use-case considered from the start or realized late in the journey. Using AI implies using a lot of data and data governance implications are something that a startup will be involved in from day one. Ping Alexandru Stanescu or Tudor Stanciu for a Digital Legal and Governance perspective. They can also cover the type of questions like “Under what conditions and legal frameworks can we use widely available data to train an algorithm? (for instance: journal articles, academic journals, open source or otherwise, images, videos from Youtube), or “registering the name/idea/patent. How necessary is it at this point? When’s the best time to do it?”. Read more on what does gdpr have to say about AI.
“Depending on how advanced a #startup is with its activity, there comes a moment when the team may want to protect their idea and should take into account the registration of authorship rights over their work. There are three categories of such rights — #copyrights related to machine learning, #trademarks related to the commercial use, #trade #secrets related to some operations or sub-ideas better to be kept secret and the patent, which is the ultimate protection on software.”– Alexandru Stanescu #knowledgepill from the #BeAI #preaccelerator Mentor Sessions.
“#IntellectualProperty is your most valuable asset as a #tech #startup — make sure you know what you can (and should) protect, and why.” — Tudor Stanciu #knowledgepill from the #BeAI #preaccelerator Mentor Sessions.
“The #ConvertibleNote has the nature of a #debt towards the first #investor, acquired under more or less complex terms, with a maturity date usually triggered by a second #investment round. The usage of convertible notes began to be popular in #startups’ ecosystem as it delays the moment of evaluation of an idea to a later stage, when it begins to resemble more to a #business #evaluation.”– Tudor V. #knowledgepill from the #BeAI #preaccelerator Mentor Sessions.
“From a legal perspective, I’d recommend limited liability companies — but I would give the legal structure less thought at this point, as it can be easily changed later on. […] The ideal place to found a startup is your own country. At a relatively low cost you can overcome bureaucracy — for instance, tax is low in Romania, if your revenue doesn’t exceed €1 million you only pay 1% in tax.” — Mircea Marculescu #knowledgepill from the #BeAI #preaccelerator Mentor Sessions.
PRODUCT-MARKET FIT > But using AI doesn’t bring challenging new facets only in a Legal & Governance segment. Working on the right business problem to address is an essential aspect for any early stage deep tech startup. The particularity brough by AI is that a developer profile tends to overshoot and a non-developer profile tends to undershoot. ALL startups seek the right product-market fit. Much of the early days struggles are around understanding space and problems. Picking the right problem matched with the right AI applicability to address in a six-to-ten months horizon is a balance that an early stage deep tech startup should strive for. How do you pick the right problem — follow traditional methods of product-market fit (don’t underestimate the MOM test & product-market fit survey by Sean Ellis). How do you identify the AI applicability for it — visit the AI Canvas built by: Ajay Agrawal, Joshua Gans and Avi Goldfarb, Rotman School of Management, University of Toronto. It’s a good proposal for assessing what type of problem to work on from an AI prediction needs angle, as an example. Tudor Goicea works wonders in channeling identifying the right problem.
“The easiest way to identify product-market fit is (by checking your) retention. If after retention you’ve got revenue then you’ve got a good market fit.” — Razvan CRACIUNESCU #knowledgepill from the #BeAI #preaccelerator Mentor Sessions.
“Look for evidence that shows customers are using your product as intended. If they use the product differently, you need to understand why as mental models are hard to change.” — Alina Catalina Banuleasa #knowledgepill from the #BeAI #preaccelerator Mentor Sessions.
TEAM SKILLS > There are 300,000 AI engineers out there, as per Tencent’s estimate in 2017. The demand for AI engineers surpasses by a lot the threshold of people available in tech. That means, from an AI team skills perspective, you will probably have to have a fallback plan IF you don’t have sufficient AI/ML talent to kickstart the efforts. The chances are, that you will unlikely hire someone and, most likely, need to train your internal team in this direction. But, is an AI role the only one you should focus on? By far, no. Most early stage deep tech startups need to ensure they build a monetizable engine from the start, so make sure to bring the right person on board to focus on this and don’t leave monetization out of sight.
“When building a team from scratch, it’s important to look at the people you’re taking on board. The people able to take the company to the next level will need to demonstrate flexibility, a learning mindset and equally important, they will share the values that you have. As a general rule — when in doubt, don’t hire — keep looking instead.” — Elena Năstasă #knowledgepill from the #BeAI #preaccelerator Mentor Sessions.
“Focus on soft skills.You should search for drive, initiative, curiosity and creativity, the ability to grasp new concepts quickly. Look for perseverance and passion.” — Oana Cotofan #knowledgepill from the #BeAI #preaccelerator Mentor Sessions.
STRATEGY > What is the best strategy for applying AI? What would be the leanest way to adopt AI? How broad should the MVP scope be? Or should the focus be on growth, quality or profitability? This sounds like the type of questions many would ask in defining their AI Strategy. But, are these the right questions to ask? Traction is one focus to have in the early days — one cannot get it wrong with traction. Leanest way to adopt it — use packed solutions out there, APIs. The providers diversify in serving multiple needs out of the box. Before you jump in building your proprietary tech from scratch, validate. And when it comes to building a grand scheme AI Strategy, it really goes down to multiple layers. The linked picture envisions it correctly. The grand rule for building your AI Strategy is build up experience first, understand AI and its intrinsic implications on segments of a business/startup and then formulate your strategy. Failing to follow this route will get you an academic approach to a strategy that may not set you for real-world success or dynamic challenges.
“Move fast! Don’t cut concepts! Reduce the scope of features! The MVP has to be complex enough to get initial traction for your concept. Put the rest of the features on the roadmap.” — Virgil Ilian, Ph.D. #knowledgepill from the #BeAI #preaccelerator Mentor Sessions
DEVELOPMENT > Moving onto AI Development. Does it make sense to outsource? Does the annotation effort need to be a team effort? Should you jump in tackling a NLP issue in a local language without much precedent? Does it matter if it’s Azure, GoogleCloud, Amazon, Oracle, IBM or others? Should you make an upfront choice for Pytorch, Tensorflow, Keras? It all depends on your team profile and less so on the many subfeatures, methods, systems or options out there. Having the right contacts in your network to ask around helps. Using AI, contrary to some practices, is not about the type of AI you use but the problem you get to solve and for whom, and understanding that over and over. Build systems that can be a partner in supporting the continuous adaptation and insights loop.
“The business knowledge and understanding of what you want to achieve is very important. The platform you choose for development — that is just a preference. Equally important — incorporate feedback into your product as fast as possible.” — Krzysztof Suwada #knowledgepill from the #BeAI #preaccelerator Mentor Sessions.
“Outsourcing AI only makes sense if you don’t have the technical expertise. If you’re positioning yourself as a tech startup, you’ll find an AI developer to join your team, even if just as an advisor.” — Traian Rebedea #knowledgepill from the #BeAI #preaccelerator Mentor Sessions.
PITCH > How should your AI pitch be built? Should you focus on AI mentions or should tech be agnostic and be focused on the problem you try to solve? Non-tech audiences are part of the acquisition cycles even for the most technological problems solved. Should you have multiple messages for different audiences? Is empathy your go to path? A lot will derive from the other segments and their respective choices.
MARCOM > AI MarCom (Marketing and Communication) is a segment impacting any message a startup delivers. There is a reason growth hacking or content marketing techniques are so ROI friendly and why some startups get to create such meaningful and strategic partnerships that make a difference. Developing internal and external communication for investor relations, government relations, consumer/user/enterprise partner education, talent and team communication is key to how well you sail both hard and good times to maximize your odds. Tricks from frameworks like JBTD, Buyer personas, positioning and customer interviews will help you. Andra Zaharia, Alexandra Pahomi, Ioana Irimia, Georgiana Bodas — they all helped deliver essential aspects into this complex segment that many underestimate, under prioritized or *think* may understand it.
“When it comes to designing your marketing strategy, the first step is to define and understand your market audience. Start by creating a buyer persona — a semi-fictional representation of your ideal customer based on research, surveys, and in-depth user interviews. Buyer personas should be at the core of your strategy as they help you better understand your potential customers. Having a clear understanding of your target audience is critical for any marketing or business decision you’ll have to make along the way — it will influence your content, messaging, product development, paid ads strategy — basically, anything that relates to customer acquisition.” — Georgiana Bodas #knowledgepill from the #BeAI #preaccelerator Mentor Sessions.
“In terms of #content #marketing, be #benefit-driven (What are the benefits that you bring to your clients?) and focus on #specificity (What do you solve and who do you solve it for?). Then, you need to build trust. How? Lose the jargon, particularly if your audience is not technical. When it comes to positioning, don’t try to be many things for many people. That never works! Try to find your people. And answer to these questions >>Where is your place in the market? How are you are different from your competitors? Last but not least, unless you want to get the people excited about the tech and the AI component, talk in a simple way. People don’t buy your product because of the #AI component. They buy it because they need it, regardless if it has AI in it or not.” — Andra Zaharia #knowledgepill from the #BeAI #preaccelerator Mentor Sessions.
INVESTMENT > AI Investment — getting pre-seed investment might not be that tricky. Is it the best time to look for financing? What are its implications? There is growing interest in AI and machine learning solutions. This has created an influx of capital in the form of angel and seed rounds, as well as Series A and B. The larger firms are getting interested in this space too, and there is growing competition. AI hardware is a recent trend, with Elon Musk leading the pack. Musk has invested in almost all of the major hardware AI companies, but there have been other players that have gotten into the AI hardware business as well. Some of these hardware companies have raised massive funds.
“As an investor, you look at the profile of the startup founders asking for financing. 95% of the decision is based on the founders’ skills. They need to be action oriented, learning oriented, solution oriented. These traits indicate a growth mindset.” — Dragos Nicolaescu #knowledgepill from the #BeAI #preaccelerator Mentor Sessions.
“Signs that you’re ready for investment? A solid MVP, a strong team and 1–2 contracts putting your solution to work for another company. That’s when investors will start approaching you, offering their support.” — Alexandru Stanescu #knowledgepill from the #BeAI #preaccelerator Mentor Sessions.
“The best time to look for investment is when you have some clarity on your burn rate. You need to bootstrap as much as you can, and have a clear landing strip to get you to the next financing round. However, don’t rush to accept money just because you need funding. For an AI startup, founders need to work with investors that understand AI.” — Tudor V. #knowledgepill from the #BeAI #preaccelerator Mentor Sessions.
“Investors are always looking at the adhesion of the team, how well it works together. As a founder, proving that your team can iterate fast and is adaptable is something you need to have if you want to attract investment. Equally important, if you really believe in the product you’re working on, your core focus will be making it successful, understanding what your customers want.” — Mircea Bardac #knowledgepill from the #BeAI #preaccelerator Mentor Sessions.
PACKAGING OF IDEA > Jump to execute pilot projects to gain momentum and show traction within six-to-ten months. In parallel, develop your core proposition, validate and discover. Focus in creating network effects and use AI as an accelerator for your idea if fit. If you’re in a niche-specific space, search for those programs where large providers of the niche focus on innovating and reinterpreting solutions for their needs. Like if you target retail chains, join a retail chain Innovation Accelerator to understand more. Such programs may give you a better understanding of how to best package.
“If you pitch a product that’s free, people will not value it enough. A free product doesn’t send the message of responsibility to the clients. You cannot validate a free product, so keep in mind how can you monetize as quick as possible. Having a portfolio (including a presentation, a pitch, an action plan) will help you set up a clear price for what you’re offering. You can later offer discounts as appropriate.” — Raluca Tătărușanu #knowledgepill from the #BeAI #preaccelerator Mentor Sessions.
“Reach out to your users as fast as possible to get essential feedback on your product, start with the end in mind. You could approach the users in different steps. One option is to do smoke and mirrors and then you validate hands-on features.” — Razvan CRACIUNESCU #knowledgepill from the #BeAI #preaccelerator Mentor Sessions.
CUSTOMER VALIDATION > Focus to release early, move fast, capture feedback, improve product, get users, more data and repeat. Ai’s specific is that it’s a loop process, be prepared to create a wheel and not have it as a one off process. Validation never ends. A great way to extend the business’ cycle of validating an idea, product, etc. is to be 100% external, the way Good, Sad, Mad people test their theories. How do you validate customers once you have a few? First you need to understand the problem. What’s the value proposition of the feature? The second question is how do you measure that value proposition and what does it mean in terms of usability? If the value proposition is that it improves the user’s user journey, then do a usability study to learn how. After a usability study you have to check the efficacy of the idea in real world applications. How does the product impact people’s lives and what does it really mean to “improve the user’s user journey?” This is where your customer validation loops will give you the answers you need and if your tests come out clean. This is where customer validation will help you improve and validate your product or feature and get customers. Have you ever validated a product idea by getting users to help in beta testing? You’re able to get feedback quicker, but you might not have the credibility to be in beta. Beta is an intensive experience and not a quick onboarding process. It takes a lot of time and energy. Test markets, user panels, user interviews, listening sessions, focus groups and user validation are all critical parts of product validation. Alpha, beta, release. Iterate. Repeat. Find Product Validation Partners. -Andy “Hamster” Richardson
AI Limitations — prepare to explain yourself
AI-fication rush may be dangerous, and here are some top level takeaways:
- Looking at AI as a secret sauce to pour on top of your existing data processes, without focusing on the cost of opportunity is a clear path to disaster.
- Another weak spot is when ethics and harm potential are left aside or downplayed. Make sure there’s a framework backed by a strategy to keep things on the right path.
- There’s one playground that requires you to be optimist: future scaling. With a reasonable cost in mind, do prepare for success: you will have tens to hundreds of models in production, and growing demands for your machines. Add up continuous retrain and additional pipelines for insights use and a ton of pain if grown in the wild. Therefore, make sure you build your systems and infrastructure to be elastic and soft on scale-up efforts.
- Explainability is hard. Make efforts in this direction. DARPA is a good starting point to understand more.
- Be mindful of biased data. Whether included in your model, or within the historical/training data, bias will skew everything. Opt to have sociologists and psychologists work close with your data scientists, to ensure cleanliness. If your team has an industry expert part of it, you’ve got yourself a good balance.
- Adversarial attacks on AI are dangerous. Never underestimate your footprint and impact potential for what you’re building.
- Skills and culture before tools. DevOps and MLOps are real but not sufficient.
- Accuracy threshold — Your time to value for your customer is your accuracy.:) Don’t get trapped into the intrinsics of AI tech. We are using the tech to augment something and get us better, faster or more cost efficient, somewhere. Stay focused.
To conclude, is it enough to educate a startup to just add technology and aim at technological acceleration? Or look at the tech as a means to make business better? We think customer development techniques are as important as picking the right problem to work on. Technology can only be one of the many mechanisms in reaching the end goal.
Offering 360 support for something that just emerged from the labs and is here for commercial impact is much needed. It’s important to bridge early deep tech startups’ needs to market’s current offerings and expectations. Democratizing access to provide everyone with access to multi-disciplinary resources and knowledge will help more of us gear up an AI-based startup as a side project or main focus. The whole idea behind the effort is to learn to understand expectations and needs from each iteration level, advocate for collaboration, and have more practical products that solve real world problems.
It is time to rethink AI, get informed and participate in the creation of the future. It’s time to #BeAI and demand a future without fear. Join our community on LinkedIn, Meetup, Facebook and be amongst the first to hear about our initiatives and get to contribute. Pre-Accelerator Demo & Pitch Evening on Dec 8th, attend to hear the selected 10 finalist startups as they demonstrate and pitch their solutions!
Thank you for reading this far!
[🔮 A small part of this article has been generated by AI-technology.]
This article was originally published under my LinkedIn profile on November 13, 2020.
Why did an AI community chose to run a #deeptech #preaccelerator? And what #AI means for everything you do in a startup. — There is so much effort and know-how behind #BeAI! The need of many, in a community for all. — There are not enough thank yous for the time and effort of people contributing to paying it forward. #startups #artificialintelligence #innovation #entrepreneurship #strategictransformation
This post was originally published by Alexandra Petrus at Medium [AI]