It may be daunting to think about how you could implement artificial intelligence (AI) in your organization. We’ve got the answers to your questions about getting started with artificial intelligence. 

While AI for industry can have a clear and explicit return on investment (ROI) through greater profits, it is sometimes harder to measure the return on a nonprofit or government agency. You probably want to measure your ROI in terms of one or more of the following:

• Impact: Is the AI system helping you create a larger impact & better accomplish your mission?
• Savings: Is it saving you money by helping staff accomplish tasks faster or with a better result? 
• Improved customer service: Do the results make it easier for your key audiences to get the information or services they need?
• Back office: Is your back office more efficient because of the AI system (e.g., speeding up invoice processing, getting payments out faster, improving inventory control)?

Starting your first AI project

Don’t start with a critical system. 

Failure should be an option. If the AI fails, it shouldn’t disrupt anything. 

Have plentiful data. 

Understand the data you have. Try to pick data that is:

• Clean, without any gaps or outliers.
• Structured, maintaining the same structure across the data sources. You don’t want, for example, to input books and spreadsheets because the AI may not know how to interpret them differently. 
• Well-labeled so that the AI knows what success looks like. If you are inputting thousands of old grant applications and you want the AI to predict which new grant applications will be successful, then you have to tell the AI the key labels, such as the grant name, their final score for each category, whether they got funding, whether they were successful in implementing the grant. Without this data, the AI won’t know how to interpret future applications.

Know your goals!  

What do you want to accomplish? It is hard to know if you’ve reached your goals if you haven’t defined them.

Two options for getting started with AI 

1. Build with an internal team (self-hosted)

There are many tools available to organizations that want to build their AI capacity in-house. When getting started with artificial intelligence, many organizations will opt to use a suite of AI tools that you can easily get online instead of hosting their own. Amazon, Microsoft, and IBM provide toolsets to help you get started, even though they are considered to be quite technical in nature. Because of this, we recommend staffing your team with people who have programming and technical backgrounds.

2. Hire a team of experts

When you hire an external AI firm, look for one that has worked with similar types of organizations to your own. This will help you better assess if the ROI will be worth it for you since firms with similar clients have likely worked on similar products to what you are looking to build for your team. 

Regardless of the type of solution you go after, you’ll want a diverse team working on your project internally. AI crosses boundaries in some unique ways, so only bring on people who are interested in this work. This may include someone from HR who wants to use AI to speed up resume reviews, a programmer who wants to jump into AI, the CTO who wants to experiment with AI, or a writer who can help troubleshoot some of the language matching implications. If you bring in only technologists, you’ll get a solution, but may not understand the implications of it.

More robust, more accurate, and less biased

You should set reasonable expectations when getting started with an artificial intelligence project by not expecting an AI solution to work right out of the box. It should be able to have a quantifiable degree of success, but you’ll want to keep training it over time. This is what is called “machine learning” (check out our last post for more on this). One of the biggest problems with AI is that it uses existing data to create models and predictions for the future. 

If you are feeding biased data into the system, then it will produce biased results at the other end. For example, if you’re using AI to look at past grant applications to recommend who to fund in the future, you might be missing that the last Foundation President had different priorities, so you’ll build those old priorities into the AI system. Consider what biases you might be building into a new AI system without knowing it. 

For more ideas on how to use AI, you can request a recording of our most recent webinar “How to Get Started with Artificial Intelligence.”