This is a standard question I get while we talk about AI projects. Recently I was delivering a seminar and I got the question even from IT company owners.
If you ask such question to any Data Scientist or Machine Learning Engineer they would try to respond the question based on few key areas:-
Data
Algorithms/Models
Accuracy
We can start by asking ourselves some questions like:-
Do you have data? How much data do you have? Do you already collect the data needed?
How is the quality of data? Does it need cleaning efforts?
Do you want to use a precise type of algorithms? How are you going to use the system?
What is the latency criteria and will you be going for real time training or does your data keep changing for your algorithms?
What level of accuracy are you satisfied with? 70–80% accuracy would be good enough?
These questions would help you understand your current state of Digital Systems.
I would elaborate one example, as an answer you come to know that the data which your existing systems collect does not have items which you want to measure or which can be valuable to your AI system-then for you first step would be to start collecting the data and then may be after few months you have sizeable data to start actual AI work.
Following is a typical Data Science project life cycle:-
AI project life cycle (Source:- https://docs.microsoft.com/en-us/azure/machine-learning/team-data-science-process/overview)
Data acquisition and understanding it self is very important step before we get to model training stage.
In a summary, while you have a well defined requirements (this is where most of the time gets consumed in any software development work — sometimes even years …), it can take at least 4–6 months to have a working alpha version.
This is a ballpark estimate for the AI projects and the same thing can not be applied as a rule of thumb. So, as per the use case and exact requirements, project scope, and timeline changes-case on case basis.