How to build the initial AI algorithm for the business problem
-
They created an AI platform for big brands that are about to put some POS or Display inside of a store; their system can estimate the return over the design. So they can get 10 materials to register and judge them.
-
Type: Small company
Funding: Pre-Seed/Angel - Around 400k
# of Employees: 1-10
Ownership: Private
Annual Revenue (not publicly shareable): Pre-revenue - In the process of closing their first big client in Nederlands UPDATE: They are now post-revenue. Closed the first big client
-
Business Problem(s)
MVP product - They didn’t know how to build the initial algorithm for their business problem.
What would happen if they didn’t solve these problems?
They wouldn’t have a product at all (our algorithm is their product) - This client has gained potentially the most out of our collaboration. Any valuation now is due to that algorithm.
-
An end-to-end Machine Learning solution that takes data from their clients, processes it and predicts what is the potential outcome of their in-store marketing efforts (POS, Displays).
Details:
Their client’s uploads a set of images for their proposed POS/Display designs, Shopnosis uses people to initially label the image and provide as much info about it as possible.
Our algorithm takes the image and labeled information and starts the prediction process.
GCP (Google Cloud Platform) uses the image for similarity and finds similar products that we already have in the database -> this part return 5 most similar products
Statistical models takes the labeled info and run the set of rules to filter out the noise and the best candidate
The combination of both stages are used to determined the final prediction
Increased the accuracy of their system to 70%
UPDATE OCTOBER 2023: In the meantime they hired a full time data analyst to help them with data and stats full time. They proposed to move to the improvement stage of the algorithm, we performed the analysis process and additional tests and proposed to upgrade the current process with ChatGPT based classifier - no implementation needed from Datablooz. The whole infrastructure will be moved to their GCP account.
Results & Key Learnings
They closed the first client of 400k ARR
Got an verbal offer to sell the startup because of the algorithm for $15 million
