Successful implementation of AI in an organisation goes through three phases: 1) explore & experiment; 2) dev, test, and integrate; and 3) roll out & govern. And although there is no single standard when it comes to a management strategy to navigate this flow in a consistent way, many successful companies tend to follow this pattern.
Explore & Experiment
This initial phase is typically a business-heavy phase, where business owners and their IT counterparts explore a handful of business use-cases and experiment with the sources of data available and the AI technology options, typically during a 6-8 week period.
→ Outcomes: AI models and data pipeline code, along with a clarity of tools required for productionisation.
→ Risks: Substantial number of business use-case, upon exploration & experimentation, will deem unfit to pursue further – i.e., there is a higher risk of use-case abandonment.
→ Teams: Engage with internal or external teams where you as the business manager has most leverage, both on the quality of work as well as cost. Typically, boutiques consultancies & AI Labs can play a very competitive role in this area as they are highly creative and in need to prove their value quickly. Keep your internal Data Science or Data Engineering teams as oversight only, and your BAs & PMO to manage timelines & budgets.
Dev Test & Integrate
As exploration exercises find viable use-cases, in this IT-heavy phase teams of Engineers work to incorporate live input & output data pipelines in the previously-experimental projects, and provision the end to end solution on the cloud or on-prem environments.
→ Outcomes: A fully connected pipeline, from input data to creation of insights, to assimilation of insights into the target systems
→ Risks: Highest risk in this phase is around engineering solutions that are brittle or inflexible, as there’ll be a lot of pushing and pulling (changes) required to get the solution working so that it can create business value.
→ Teams: Often having external Cloud consultancy and software product PS teams to work with predominantly your internal Data engineers, IT, infrastructure, & security groups will create best results.
Rollout & Governance
For most of us in the industry, AI is still such a blackbox of a technology so it takes much longer to feel comfortable with a solution that is AI-infused. Good solution-rollout and service governance approaches make sure that the stakeholders of AI & the insights generated can stand behind it in testing times. Additionally, with AI, governance is not just about the program governance, but also to make sure that the drifts in the AI models can be monitored and captured, such that they continue to deliver the business results desired
→ Output: On-going business stakeholder confidence in the value that the Insights continue to deliver
→ Risks: Often most neglected aspect of an AI program is the rollout & governance, due to general lack of experience of implementing large scale AI programs.
→ Teams: Ideal for BAs & PMOs to manage this phase, with support from a Managed Services vendor for repetitive & BAU tasks. Resident Data Scientists & Engineers need to review the model performance and accumulation of missing data in the data pipelines, if any.
At Syntasa, we have seen tremendous potential for a platform that is able to connect the entire value-chain of AI in Digital: from creating data ingestion pipeline to applying machine learning algorithms, and to finally disseminating the insights back to action and activate on some of the most popular marketing clouds like Adobe. Syntasa is also a Cool Vendor, according to Gartnerʼs Personalisation study. Our platform simplifies the technology challenges by creating an entire “app”-like framework for creating & managing hairy data pipelines, applying super complex AI models to the data, and then sending the actionable insights back to web and business applications with an inbuilt micro framework of DevOps and cost reporting. Loved by Data Scientists & Data Engineers (who can use their existing tools with Syntasaʼs platforms), by IT and Infra teams (who can simply plug in Syntasaʼs containerised solution within their existing cloud or on-prem data lakes environments), and by business users (who can configure and see their complex business strategies laid out in easy-to-navigate graphs).
If any of this resonates with you, email me (Apoorv Kashyap) and share your thoughts or experience with your own AI projects.