If the solution is not technically possible at the moment, how much time and effort is needed for research and development (R&D)? Are these estimates aligned with time and funding limitations, if there are any?
In some cases, a tech team is strong and the idea is very promising, but it might take full five or ten years to develop and be adopted – like quantum computing for solving enterprise-grade problems in the pharmaceutical industry.
You have to be honest – and realistic – about the timing and expenses. You will certainly get this question. Here you need to distinguish between research and common software development costs: the research stage is inventing algorithms to build something that previously hasn’t been possible due to technological limitations, with uncertain results and timelines. The software development stage is building a well-understood solution, which only requires a certain period of time.
Clearly enough, investments at the research stage are much less predictable. However, development can also take much longer than team plans originally, trying to impress investors and overestimating capacity. Make sure you don’t.
In the case of a software product, does the project really need proprietary software and not a white-label solution or SaaS?
Reinventing the wheel might be seductive. However, in some cases, spending resources for the development of a new in-house technical solution can be a waste of time. If you as a startup do not suggest a software innovation, it might be easier and cheaper to purchase a ready technical part and customize it to the particular business needs.
What are the external dependencies (e.g. libraries)? How is external software maintained?
No software is written totally by the company in-house team. Every project in the world uses multiple external databases and code libraries, often open-source, maintained by global communities of developers or by corporations. The resilience of the project depends on the timely update of external software for security and efficiency.
If you’re doing an AI project, what is the source of data? Is it sufficient? Is it available?
The viability of AI projects is extremely dependent on data quality. Algorithms may be inefficient when there is not enough data. Also, inherent biases in the data (e.g. racial) will impact the final algorithm. Furthermore, there may be a chicken-and-egg problem if the customers are a source of data and, at the same time, the main value is delivered using the AI/ML. If the data is not free, its cost should be considered vs potential value compared to using less advanced methods.
If you’re doing an AI project, how is the context-dependence addressed?
Even if there is plenty of data available, it may be gathered in a specific context, often being non-applicable in another. For example, if the network was able to distinguish cats and dogs indoors, it may be unable to do so outdoors.
If you’re doing a blockchain project, why the database should be distributed, in other words, why do you need blockchain?
Many problems that are claimed to be solved with the blockchain can be solved with a simpler cryptographically protected database with a robust permission management system that can also utilize public-key cryptography if needed.
In the case of the original concept of blockchain, the database is distributed among multiple participants with all of them being able to make an input. This is not always needed. For example, an enterprise may need a database to store and process its internal data, in which case it shouldn’t be distributed. Or it may be a database of a governmental body, to which everyone should have access but only the government should be able to validate input data.
If it makes sense for a database to be distributed, does blockchain have to be public?
Blockchains can be generally divided into public and private. Public (permissionless) blockchains are the ones in which anyone can host a node, thus having access to all data recorded and validate database updates. In private (permissioned) blockchains only certain participants can have access to data and validate input.
Public chains significantly reduce the control over the business as the state of the database is now controlled by multiple people scattered across multiple countries. This also means an increased regulatory uncertainty, especially in the case of heavily regulated industries or the ones that are of systemic importance. For these reasons, the case for public chains must be really strong. In many cases, a private blockchain is enough to satisfy business requirements. For example, transaction processing requires only financial institutions participating in the blockchain, sharing medical history data requires only hospitals participation.
If you’re doing a blockchain project, what are the incentives of participants to act for the benefit of the system? What are the ways to break these incentives and how are they addressed?
As blockchain, especially the public one, is maintained by common efforts, and the quality of data, the transaction costs depend on the participants, incentives should be designed in a due way to ensure that the system is sustainable.
An example of where it is problematic is the Tezos blockchain that utilizes the so-called Liquid Proof of Stake (LPoS) consensus algorithm. A consensus algorithm is a way in which validators agree on the new state of the ledger. In LPoS consensus participants can stake a certain amount of a blockchain native token to get a right to either validate transactions themselves or select another trusted person that would do that instead, who would validate a transaction and distribute the reward. Although such algorithms have multiple benefits, the common point of criticism is that incentives for participants to become validators are questionable as they can select someone else, and still receive a significant chunk of reward because of the competition among potential validators, while not spending time and computational resources on network maintenance and governance. This creates a risk of blockchain centralization and various types of attacks.
How is the cybersecurity ensured?
Cybersecurity is a primary feature of any IT infrastructure. Especially for a regulator, who’s main concern is protecting customers.
If you’re doing a hardware business, how is the quality of supplies ensured?
While software businesses are dependent on external libraries, hardware businesses depend on supplies providers for the quality of their solutions.