In this series we explore the most important parts of moving AI/ML projects from lab scale to production.
If you haven't started exploring or experimenting already, start today! This is a series on the journey to scaling AI/ML and every journey starts with the first step. There are a few ways we have seen (and been involved in) this happening:
1. Pay an external vendor to create a POC
a) Work with an expert in the field and not a traditional software vendor who you will be paying to figure it all out as they go.
b) Work with a vendor that has AI/ML code or software as a deliverable, not just insights in a report or presentation.
c) Involve the business outcome owner in setting the criteria for success so that any deliverables don't just sit in a repo somewhere and get forgotten.
2. Run a POC internally
a) This is a great way to get enthusiastic team mates to understand the reality of kickstarting AI/ML from scratch, with knowledge being retained.
b) We recommend you use a cross functional team rather than just doing it as an IT project. Focus on informing business outcomes and not just tech choices.
c) Communicate the initiative broadly to get stakeholders on the journey to understanding what might be possible and how it fits with strategic outcomes.
d) Success needs to be learning and not a software deliverable (we know that it's really hard!)
3. Run a POC collaboratively
a) Blend internal and external domain experts, data owners, business owners and technical resource.
b) Make it time boxed if there is a dedicated resource allocation or be realistic about the time availability of part-time resource.
c) Discuss ownership of any code developed or IP that results from the process. If you hold on too tight you'll have to work with a pure services company that is motivated by how many hours you can be billed for.
Whichever way you decide to get going, ensure its not just a 'tick box' exercise where the outcomes are not linked to an agreed pathway to next steps. Experiments should inform the journey. Start today!
Next, Step 5: Cloud ML