Shifting AI projects from Islands to Cornerstones

We recently started working with a new customer that had no shortage of AI ideas. In our first meeting, we were presented with a big print-out of all the use cases their team had come up with, neatly divided into quadrants for impact versus readiness. A steering committee that represents the different parts of the business was set up to drive things forward. But despite the strong start and clear budget, there hasn't been a standout AI case yet that really kickstarted momentum internally.

The team itself already highlighted one problem here; for none of the steering committee members, data & AI was their core business, and being a business with high-intensity customer contact, there is always something urgent that needs attention every day. But is that any different for other companies? Data & AI initiatives will always exist next to the daily operations.

How can you get from all these initial ideas into that first use case that is touching the daily workflow and moves the needle internally? Here are some questions we think are good conversation starters to get the discussion started.

Are we creating Islands or Cornerstones?

In the beginning, it’s all about showing tangible results, and preferably as quickly as possible. So the natural flex is to go for low-hanging fruit. An AI note-taking app that listens to meetings and replaces manual creation of minutes. A chatbot where you can upload documents and discuss the data behind them. An app that can process photos and find common patterns. 

They all are possible nice ways to have a first entrance of AI in the daily workflow. But these examples are Islands, not Cornerstones. They focus on a small area and optimize a piece of a process. However, they are not connected to the overall business knowledge or history. Not working across different domains. And it's pretty hard to translate this into a big business case and justify large next investments.

‍What are our Cornerstone use-cases?

Cornerstone AI use-cases enable you to do something small that can easily be extended. Going from one step in the process to another. Implementing something that works in one department to the next. Bringing in more data to grow the impact. Extending the focus from internally to suppliers or buyers.

‍What does it take to create a Cornerstone?

1) It always starts with your internal data

This is where all the unique knowledge about your business lies and where the basis is created to apply things across different functions because you can combine finance, sales, customer, and other data. The biggest potential is to create something that makes a difference in cost savings or revenue potential because it uses all the historical and daily knowledge in the company.

2) Every AI tool that you implement is both a data source and a destination

Many AI tools will request access to various systems, including your CRM for customer data, files, calendar, or other team-used tools But this has a few issues:

• Data fragmentation: Data is scattered across multiple tools that don't communicate, creating islands that only solve one thing. And it will be hard to control which exact data goes where.
• Knowledge silos: AI-generated data remains confined to a specific tool and is not integrated with all the other knowledge inside your business, it stays in a silo.

Utilizing AI tools, you can concurrently leverage these instruments as both the source and destination for your data. You can extract, store, and merge the generated data with your internal data. Send a specific set of internal data to the tool for processing, but in a controlled manner that allows you to contribute domain knowledge and maximize its value.

3) The infrastructure that you use enables starting small and building out

One of the reasons why many projects in IT generally fail, is they are very big and complex to implement. The same holds for data projects. To create a cornerstone AI use-case, you need data infrastructure that does the whole flow from data collection, cleaning, getting ready for AI, and storage. 

To not end up in another year-long project, it is important to set this up in a lightweight way on a limited set of your data. Initially, you can roll out this first case fast and at a sensible budget, quickly showing value & ROI to all stakeholders. By selecting the right scalable tooling, you will still be able to extend this easily to the next projects once the value of the first project is crystal clear.

‍How do you create that first Cornerstone AI use-case?

1) Create a design for the infrastructure

Most conversations start from the user's needs. “If we implement this cool AI tool, we can do XYZ”. But from our view, it needs to be top-down and bottom-up, taking into consideration the perspective of your internal data, which will lead to more balanced and achievable projects. 

This means, first getting a very good understanding of all the different tools that are used and the data that is stored here. Then, map out a clear picture of the data you have that AI can work with. Based on the overview, a design can be created that is focused on getting as much domain and internal knowledge as possible to AI models.

As a final step, the design should have a very Lego mindset. Especially since it is still early days and new techniques and tools are launched every week. By using small pieces that can be swapped to alternatives when you go from a lean first use-case to more scale & higher quality.

2) Select a use-case with a clear €€ value

The earlier example of an automatic meeting note-taking AI is useful, but pretty hard to translate into tangible savings or upside. Yes, team members will be happy that they don’t have to manually write minutes. But what is the real financial impact of that? Moreover, for these generic use-cases, better off-the-shelf solutions are likely to emerge in the near future. 

Compare this to using your internal data and history to enable:

• Sales teams to write proposals in 50% less time so they can close more deals.
• Purchasing teams to reduce inventory by 30%.
• Customer support to handle 70% of tickets automatically with a smaller size.

These are just a few examples, but they all have a tangible financial impact. That impact enables more use-cases to follow at a larger scale, with a proven case and internal enthusiasm as drivers. 

3) Start with one team, and just a few data sources

The examples mentioned share two additional similarities: they focus on a single team and only require a subset of the complete company data for operation. This means that you can start with a relatively small investment in technology and resources.

Take things step by step and branch out to multiple teams and more data sources., because you have not built an Island. You’ve built a Cornerstone.