For the past 11 years, my company has developed workforce training and education programs for a wide range of humanitarian nonprofits. And during that time we’ve watched our clients’ staff do amazing things despite workforce shortages and resource constraints – from operating health clinics in underserved neighborhoods to maintaining systems that provide advance warnings for hurricanes and drought.

However, the shockwaves created by recent cuts to U.S. foreign aid, plus a falloff in philanthropic giving domestically, are forcing many nonprofits to make painful cuts, laying off experienced staff and scaling back programs in the communities they’d served for years. 

The bitter irony of this situation is that humanitarian organizations are experiencing these pressures at a time when private corporations are generating unprecedented wealth, in part because of the boom around AI.  Which begs the question- could the AI boom that’s currently enriching tech companies also benefit humanitarian orgs?

Closing the Gap

For the past year and a half, my own company has been experimenting with using AI to deliver coaching and support to frontline workers.  While our focus for the AI offerings was originally on  the private sector (given nonprofits tend to be late adopters), I asked a colleague who worked at USAID whether our AI agents could help address the humanitarian workforce gap.  

He chuckled ruefully and said: “Look – everyone needs what you’ve got but nobody wants what you’ve got.  Humanitarian professionals want to hire other humanitarian professionals who can become their friends and colleagues.  They don’t want to work with a robot, no matter how good the robot is.”

But while my colleague’s observation proved true early on, in the months that followed the USAID closure we found that many nonprofits – including cash-squeezed domestic nonprofits – were becoming more open to using AI for training and technical assistance: if only because they had no other option. 

Case in point: we recently developed an AI agent for a nonprofit that provides capital for banks in developing economies to loan to clients for energy efficiency and renewable energy upgrades.  The AI agent is capable of providing technical guidance and sales coaching to help bank staff identify promising investments for specific clients and make the business case for “green loans” to fund those investments.   

Previously, this type of support was provided entirely by human technical experts.  However, based on what we were told, the organization only had about 20-25 qualified experts to support 35,000 to 50,000 frontline bank staff.  

Breaking down those numbers:

• If each of them worked a solid 50 weeks a year (no extended summer holidays, no sick days), and… 
• If each of them did eight 45 minute consultations with frontline staff per day (unlikely given their other responsibilities)…
• That would equal 50,000 Frontline workers receiving 45 minutes of support each per year, and of course there would be bottlenecks and delays if calls for technical assistance came in waves rather than a steady drip of eight per day.   

By contrast, the AI coach we developed can give 4 to 5 hours of support per user – a capacity increase of at least 6.7x – and it can do it 24/7, on demand, in local language.

Hours and Dollars

While nonprofit organizations aren’t out to make money, they still need to demonstrate measurable impact from the money they spend on programs (“mission ROI”.) 

So… what is the ROI of using AI to address capacity gaps?  

Another client of ours runs a program to help young people with developmental disabilities find and maintain employment.  One of the school districts they work with has about 2,000 students who qualify for the program.  And – this being the United States – the district was underfunded and completely overwhelmed. 

When we initially proposed developing an AI coach to support these students, at a price of $50 each, the client had a bit of sticker shock.  “That’s $100,000 for 2,000 students – we could hire two full time human social workers for that!”

However, some quick napkin math told another story:

• The average wage of a social worker in the US is $32.50 per hour
• Depending on whether the social worker is union or non-union, their benefits would add another 20% to 40%
• The school year is 36 weeks

If we divided $100,000 by the annual pay for a school social worker ($32.50 x 1.3 for benefits x 8 hours per day x 5 days per week x 36 weeks per year = $60,840) that would actually pay for 1.65 social workers.

And how many hours would those workers be able to devote to students?  1.65 x 8 hours per day x 5 days per week x 36 weeks per year = 2,376.  

However, those hours wouldn’t all be student facing time.  Conservatively, a typical school social worker would spend at least:

• 30% of their time on reporting and documentation
• 10% of their time in meetings
• 5% in consultations with parties other than the student
• 5% of their time on paid leave 

That leaves 1,188 hours of actual interaction with students.  

The AI system? 5 hours x 2,000 licenses = 8,000 hours of student interaction for $100,000.  Once again, 6.7x the capacity for the same investment.