Everyone needs to try out AI in their business. It has the potential to be ‘everywhere’. Yet, not everyone needs to go full-hog and build an entire factory of AI solutions. Only a few companies get to a point where they start seeing many problems to be solved with AI (or many opportunities missed) and yet struggle to make use of AI in a consistent, efficient & predictable way. These companies need to take an organization-wide view to discover, prioritize & solve problems using AI. Here, we are focused on the question – “When & how do you go full hog into building an AI Organization?”
So, do you need an AI Organization?
You need an AI Organization only when all the below conditions are met:
- You laid out a strategy and agreed at the Top Level (read CEO/Board) that AI is a key pillar of your company strategy
- You tried to solve a couple of business problems. Failed or succeeded, but learned ways to double-down on solving more problems
- The business is looking for ‘daily problems’ to be solved with data.
Not one-off issues, not research & analysis. Day-to-day decisions directly linked to business performance. - Your analytics team is stretched (time/skill/attitude) to address these needs
Ok. You need one. So, what do we need to start?
Now that we have settled that, let’s start with the ingredients to build a new AI Organization. A strong AI Organization needs to have 5 ingredients : Target Operating Model, AI Lab, Factory, Visual Command Center and Data Pipeline
AI Lab
Factory
Command Center
Target Operating Model
Data Pipeline
The quality of any output depends on its ingredients. So, let us make sure we have good quality ingredients. Some tips on each ingredient below:
- Due to the experimental nature of many AI initiatives, from data exploration to feature engineering and modelling parameters, they need to be processed/piloted through a ‘lab’ structure that is separate from other operational systems
- The environment, team as well as toolsets for the Lab need to be flexible to handle uncertainty and ever-changing demands of the analytical approach.
- This ingredient is core to making the AI organization “fail fast and iterate”.
- The AI Lab needs to be practitioner-led, cross-functional and should be able to find creative solutions to meet the business objectives
- Analytical solutions iterated and perfected in the lab move to the factory which provides an environment for running analytics jobs 24/7 with certainty.
- To reliably execute critical solutions at scale, the Factory environment needs to be robust, maintainable and secure
- Technologies underlying the factory need to help execute efficiently through continuous deployment processes, integrated with core systems & actively manage risks
- The algorithms produced in the Lab are most likely very hungry for feedback. And the factory needs to be designed to continually feed these learnings back into production and/or lab algos.
- Getting users to access the decisions enabled by AI and the underlying data anywhere and anytime is the most critical aspect of adoption
- Command Centre needs to pull together all the AI assets together visually, enable distributed access to underlying data and put the decision-controllers on the key executives’ mobiles
- This is the most visible ingredient of the recipe.
- Fairly divergent views exist on what’s the right command center approach, but if your company allows decentralizing decisions, the Iron Man style Heads-Up Display for all decision-makers is probably the best way forward.
- This is the easiest and most standard ingredient to start with. But the most difficult one to master.
- It refers to the standards you follow to identify, prioritize, & produce the opportunities offered by AI.
- From Mckinsey to NASA, there are many organizations that publish a template to start off on this one. You can adopt any of them as baseline
- Key here is to identify what works for your organization & keep modifying this framework.
- This is an oft-ignored & misinterpreted element of the 5 ingredients.
- This ingredient, usually centered around a technology platform, is a streamlined way to acquire, process & make data available to analytical purposes.
- Increasingly, there are industry standard solutions available to make data pipeline smooth & high-quality.
- Data acquisition, a core part of the data pipe, is a never-ending job that needs to be tackled from multiple perspectives. To ensure the pipe supplies fresh data, you’ll need to keep reinventing and rediscovering ways to capture data every now and then.
You have the need and the ingredients. Where do you start?
While anyone in the organization can be AI Champion, choose someone who can be a ‘product owner’ in the traditional sense. Someone who provides the voice of the customer, defines the success criteria, reviews & prioritises the objectives and becomes the first point of contact to the external teams
This person will need to put together an initial squad team to be effective. While a full-scale AI Organization may need different teams for each of the ingredients, having a single cross-functional squad in the initial days provides multitude of advantages. Make this squad a two-pizza size team with the essential skillsets of Data Scientist, Data Engineer, Designer, ML Engineer & a DevOps Engineer.
Ideally, this Champion Team comes from the people who have tried to solve the initial problem set.
The Champion team’s first goal is to get to the first prototype of the organization as quickly as possible. You don’t want the organisational patience to run out before an initial win is achieved.
Look for a business team that is ready and patient enough to work with an early start.
To check if the business unit has a good fit, ask the following questions:
- Do they have defined problem(s) that create reasonable business impact? Solving these problems doesn’t have to be the Top objective of the last quarter but needs to be in the Top 3 for the near future (2 – 3 planning cycles).
- If there is a solution, are they able to take it to execution? Or are there major operational bottlenecks? For eg., 1-to-1 personalization models are great, but the campaign management system is not enabled to run 1-to-1 campaigns.
- Do they have enough data, in any shape/format, to solve the business problem? We don’t need ALL data, but just enough to start. If there is no data / too limited data, we have a dead-start problem on hand. Effective data collection can be a major distraction for the Champion Team that is better avoided in early stages.
Design a prototype consisting of the AI Lab, Factory & Command Centre for solving the selected set of problems
Resist the temptation to build all processes & frameworks needed for a full-scale organization. There will be a lot of incentives from third parties, especially vendors, to adopt a full-scale org model. Don’t! Avoid the risk of over-engineering a unit too much for today.
You can use the following, or a simpler one, as a checklist of essential components to pull together the outputs for the prototype of the AI Organization
Capture the results generated but go with a clear view that these will always be less significant in the initial weeks.
More importantly, capture which ingredients, assets & code are working, and which are not. These differ with every organization and objective. Yet, this is where the most gains & scale will come from.
Taking a fine-brush strokes at this stage is critical. If required, work with Lean Sigma or other continuous improvement frameworks to learn faster.
This may sound the simplest step and a cliché, but this the most effective in moving towards the long-term AI Organization model. This requires a separate focus of its own, with the approach continually evolving with each iteration. So, we will cover this in the next post.
The above recipe is by no means extensive or the only way to take the early steps towards an AI Organization. However, it is based on some of the practical tips I’ve observed as a practitioner that work for organisations time & again. So, hopefully, you’ll find these useful and effective for your business as well.
Are there other tips that you’ve found useful in your journey? Or do you see some big unanswered questions? Please do share your views & questions through comments below.
All the best in your journey to AI Success! If you like this, do share and comment.