Upload
victor-osyka
View
546
Download
0
Embed Size (px)
Citation preview
How to Build Deep Learning Companyv.1
Speech @ Yandex DataFest
September [email protected] http://Russia.ai and www.almazcapital.com https://www.facebook.com/victor.osika/
Who is Victor
2
==
Been on both sides of the table: startup founder, venture investor at US/Russia fund
www.almazcapital.com. On boards of Carprice, StarWind, Nival, 2Can-iBox, Yaklass, RoboCV etc.
4 years in VC, 1 yr. startup co-founder, 3 yrs. in consulting, engineering + LBS MBA edu.
Neural networks…
3
==
• …an abstract categorizer of signals from noise
• …evolutionary-iterative code, invented by the Universe
• …replicated by our technology civilization (enough: computing, data, and architectures)
The tech is in hype now = opportunity
What is [deep learning] company?
4
==
Company is:
• Make team full-time, reliable, diverse in competencies
• Find right tech+product focus (see further problem with curiosity)
• Give convincing promise to investors: “we need $300k for a year to focus and deliver X”
• Find revenues - find use cases. And compete with dozen(s) of teams globally who are doing the same
• Scale the company, evolving “boring biz competencies”
• Sell company. (Or IPO it, 3% of VC funded IT companies)
Key problem with great scientists/engineers
5
==
Curiosity and freedom as a core value:
• “Disturbing me in my cozy introspective research”
• “Don’t touch me, boring biz guys.”
• “Customers are lamers”
• Market feedback is often perceived as an annoying factor, limiting curiosity
What is NOT business
6
==
• NOT doing some bootstrap/outsourcing then disappear
• NOT only make some fun with teammates
• NOT only publish some sci papers
Business is - to capitalize on deep learning.
Founder’s core question to himself
7
==
“Go and die for it”
Or
No go, be employee
From tech to biz – by oneself or together?
8
==
NOT business yet• Edu• Field of science• Technology in this field• Employment in this field
Business• Product, not contract work• Business model for this product• Road to market, competing with others• Company to scale it up• Team for this company
Huge shift in thinking required
OR
Adding business co-
founder?
Co-founder with biz passion – who he is?
9
==
Product obsessed
and/or
Business/management obsessed
and/or
Having business experiences already
--> In other words as cool as you but in entrepreneurship
Case: Nicira and its 3 founders from academiahttps://en.wikipedia.org/wiki/Nicira
Product vs. Custom development
10
==
Custom development company – serving other’s unfocused
requests
If you do this, please always keep intellectual property yours
Product-focused company, usually financed by investors
along lifecycle
From milestone to milestone. Give a promise then deliver
Who you want to be?
What kind of DL company you want to do?
11
==
Scientific company - ML company with
radical tech improvements
• Cross-disciplinary team
• Aims to develop new tech
• E.g. DeepMind, Vicarious etc.
Research lab, not company
• Develops new knowledge
• And outsources it• E.g. Open.ai, Caffe
library etc.
ML company with incremental tech
improvements
• Inspired by others’ papers
• In house ML optimization by CS people
• Very clear product focus
Product company, productizing some ML tech stack but
doing very fast and efficient business
● e.g. Targeting some market or some vertical (see the next slide)
● e.g.: Prisma, People.ai, MSQRD etc.
Who you want to be?
Map by Shivon Zilis
12
==
Partnership in a right way
13
==
How big customers work with startups
• Tenders, trials, very long sales cycles, offers “customize us something”
• Big co’s try everything in IT but rarely buy = don’t overvalue those false-positive signals
• Offer asymmetric/perceived as asymmetric partnerships
How big companies analyze startups before acquisition
• They look at everything from tech and biz point of view then try to replicate in-house
• Can start hiring your people during negotiations
• No way to “demonstrations in China” and alike, consult Valley lawyers plz!!
Fundraising of your first and second money
14
==
First• FFF – friends, family, fools• Angels
• Experienced• Non-professional
• Grant money by development institutions• Some want to hire you “and give your team
10% stock option”
Second• Accelerators• Seed funds• Venture funds
• Local• US/global
Legal setup
15
==
• How• US ($1.5-3k) or offshore ($4-7k) like BVI. No Cyprus pleeeeease
• Why• Fundable in future• Right agreements with any investors• IP accumulation• Start vesting early
Why it is important to ask advisors
16
==
Better to ask advisors than to be sad afterwards.
● E.g. How to think about product vision● E.g. Where to get first data (“no data - no learning”)● E.g. How to get enough computing● E.g. How to make right team DNA and right governance● E.g. How to survive first long b2b sales cycles● E.g. How to keep IP inside● E.g. Lawyers are lazy guys, don’t think in economic terms and
don’t go extra miles● E.g. Non professional angels can make your company
unfundable in future by taking half of the company etc.● E.g. Big corps pursue too narrow interests● Etc. etc. etc. etc. etc. etc. etc. etc. etc. etc. etc. etc. etc.
So what?
17
Almaz Capital is looking for promising deep learning / machine learning companies and teams who only thinking
about DL startup
Feel free to reach me at any stage with any questions. My pleasure to assist you in your endeavors from VC
perspective.
Looking for your questions: [email protected]
www.almazcapital.comwww.russia.ai
facebook.com/victor.osika
19
APPENDIX
Productizing
20
==
Product Management - create an efficient and manageable feature set from endless feature list• Develop and refine concept
• Define overall feature set
• Identify risks and estimate development costs per feature
• Liaison with tech team to clear feasibility
• Determine the exact feature set for a successful MVP
• Prioritize features and build product roadmap
• Discover and prioritize company success metrics