Pairrot

Suppose you need to share a dataset for analysis, but you want to protect the privacy of your customers and also avoid any potential legal consequences. Pairrot allows you to create a dataset that’s fully synthetic, i.e. does not contain data of real people or entities, but has the same statistical properties as your real dataset. Because it has the same properties, it is as good as the original for analysis or the development of machine learning models. Also, because it does not contain real data, it is privacy-preserving and GDPR compliant.

The product was conceived within the European Data Incubation program in half 2019. By February 2020 an MVP was developed using funding from this program. The software was pivoted on multiple occasions, most notably during the COLLISION conference 2020. We are in the process of foundraising. Currently the system is being tested in two companies: VRT, the Belgian broadcasting company, and an hospital in Italy under NDA.

We can use as a proxy for our market the anonymization (data masking) market
TAM: “The Data Masking Market was valued at USD 483.90 million in 2019 and is expected to reach USD 1044.93 million by 2025, at a CAGR of 13.69% over the forecast period 2020 - 2025”

One of the biggest issues companies face nowadays is how to make use of their customer’s data. While it is widely accepted that data is the new oil, its liquidity as an asset is subpar. Data is not exchanged easily at all, and for good reasons. Firstly, there is a need for respecting customer’s privacy and avoiding the negative impact of any data leakage. Secondly, privacy regulations such as GDPR and CCPA introduced very prominent legal barriers to data usage.
As a consequence, working with customer’s data comes with large legal and operational risks. These risks effectively hinder the efficient exploitation of data. An often-cited solution to get around privacy concerns is data anonymization, which is often described as the process of removing all personally identifiable information. However, in practice, classic anonymization is very weak as it can either result in easy re-identification of customers or in a large utility loss.

When you are doing data analysis, especially on big data, you are uninterested in each specific record. You might as well remove some records and your analysis would follow without any significant disturbance. This is particularly obvious when you consider that machine learning models are not trained with all the available data, in order to reserve some for testing the performance of the models. What really matters is the statistical patterns present in your data.
Therefore what if, instead of the original data, you would be able to use data that stem from the same underlying statistical distribution of the original dataset, while none of the original records are included? At all levels of sophistication, the results drawn from analyzing such data would coincide perfectly with results obtained when studying the original dataset. The crucial difference is that such synthetic data would also be perfectly privacy-compliant, as its records do not correspond to any real-world entity.
This what PAIRROT does, it generates syntetic data that are as good as the original for analysis but are perfectly privacy compliant!

Several indirect competitors exist that offer mere data anonymization services (Aircloak, PRIVITAR). Such anonymization was recently proven to be sub-par for GDPR or CCPA compliance or satisfactory unidentifiable data. Direct competitors are www.statice.ai, www.mostly.ai (5M funding)i, hazy.com (6M funding)

PAIRROT differentiates itself from these competitors in a variety of ways.
Firstly, the user experience is central in PAIRROT's service. This is why the platform is offered with an accessible graphical user interface. Additionally, PAIRROT's availability as SAAS and as a Python module enables users to adopt the service in a manner that best matches their IT infrastructure and needs. While Mostly AI offers an accessible graphical user interface, it lacks other options to optimize the users' experiences. The other direct competitors lack a graphical user interface, making their system available only to a team of experienced IT specialists.
Secondly, through Aindo’s strong academic ties, Aindo is in direct contact with world-renowned AI researchers. This creates a unique opportunity for Aindo to incorporate new scientific advances at an early stage. PAIRROT, as well as the direct competitors, currently are effective for tabular data. This means that all data must be of a particular format (numerical or categorical variables). PAIRROT and Hazy are currently already able to take slightly more advanced data structures into account (for instance: date-time information). However, through its network of experts in the fields of AI and computer science, Aindo will be able to incorporate more advanced data types such as time-series. This will open up an entirely new realm of possibilities, for instance in the field of quantitative finance.
Thirdly, Aindo’s experience with AI applications across various industries allows for a more domain-specific client outreach strategy, in which customers are made aware of PAIRROT’s relevance for their highly specific needs.

We are targeting mainly a B2B approach. We plan on having a SAAS B2B solution and a deployed software. SAAS and the licensed version will be essentially the same software, but the deployed version will be installed on the customer’s premises.
In the case of the SAAS solution, we envision a pay-per-use (Fee-for-service) revenue model. For the deployed version it will be a licensed model.
Today Aindo makes money through consultancy, while PAIRROT is in a pre-revenue stage. However we want to switch from a consultancy-first revenue model to a product-first one.

Development and marketing efforts

We are looking for investment of at least 500K

As the technology is unseen before and completely disruptive, a notable risk of this technology is that users will not recognize this technique is useful to them. Also, when aware of the potential, they might still use it for the wrong use case. It is therefore important to devise a marketing campaign that raises awareness in potential customers and is also explanatory.

PAIRROT won the European Data Incubator (EDI) competition in 2019/2020

PAIRROT won the European Data Incubator (EDI) competition in 2019/2020