It Told Her When It Was Guessing

I sit on the board of a small foundation in New York City, and its executive director is my friend. I'm not a neutral party to this story, so I'll start there.

The Caigh It Forward Foundation (CIF) was created to honor the life and legacy of Caileigh Duggan, "a vibrant young girl whose compassion, courage, and spirit continue to inspire." It runs two volleyball tournaments (Caileigh's sport) in Breezy Point and across NYC, funds scholarships that recognize super kind kids, and brings school talks and kindness days to teenagers around the city. All of it asks the same thing: be the person who reminds someone else of their worth. The family tells Caileigh's story at caighitforward.org.

Maryellen (O'Donohue) Duggan runs it. I love this organization and the people behind it.

The record is a stack of flyers

Here's something I didn't understand about grassroots organizations until I helped create and support one.

If you asked Caigh It Forward for its institutional documents, you'd get a stack of event flyers. Speeches from tournaments and events. Scholarship announcements. Social media posts. That's the record. That's what a foundation looks like when it's built by volunteers around a mission instead of by a paid staff.

There's no annual report (yet). There's no strategic plan in a binder. And nearly every tool built for this sector quietly assumes there is. The onboarding starts at "upload your annual report," and an organization like CIF could get lost. Not because it doesn't know who it is. Because it was never written down in the format the software expected.

I wasn't looking for compliments

When BeMo's beta opened, I asked Maryellen to test it.

Asking a friend to try your product is usually a good way to collect compliments. That's not what I was looking for. I got the most careful, most useful critical feedback of any tester in the beta.

BeMo starts with a Knowledge Base, the place an organization's own material goes in so the software has something real to work from. She flagged that our directions for filling it made the assumption I just described:

"I know your direction in the KB video talks about loading mission statements, annual reports, etc. But, wondered if smaller orgs would benefit from a longer checklist that included things (if they should) like: social media posts, event flyers, local speeches. Which is what we have most of."

She told us when something we'd built asked too much of the person using it:

"I think you may be asking users to make a big leap to working this way."

And when files failed to load, she didn't push through and politely tell us the rest was great. She stopped testing and told us to fix it. Then she filed the ticket.

That's what a real tester sounds like. None of it was complaining. It was a roadmap in plain language, from someone who wanted the thing to work.

There's no checkbox for kindness

In the middle of all that friction, Maryellen asked BeMo, in plain chat, to find funders for a foundation that works to increase kindness, or funders focused on girls' volleyball. Try that search in a database sometime, or with an LLM.

It surfaced two grant programs she wouldn't have known to search for: The Call for Kindness Grant and the NAGWS Legacy Award. Read those names again next to her mission. Then it went further and suggested corporate sponsors from inside her own sport, Mizuno and Molten.

And then it did the thing this whole story is about. When she asked for specifics on those two brands' giving programs, BeMo told her it was guessing.

"Loved that the search brought up corporate sponsors like Mizuno and Molten. However, when I asked about specific information about these two brands' grants, BeMo came back with that this was speculation on their part not data from BeMo database."

Speculation, not data. The software said so, unprompted, at the exact moment it would have been easier to sound confident.

And notice where it happened. An organization whose record is flyers and speeches gives a tool less to work from, so it's guessing more often. Which is exactly why it has to say so.

She didn't read that as a limitation. She read it the way you'd read it in a colleague: as a reason to trust everything else it had told her.

The colleague test

Think about the people you actually trust at work.

It's not the ones who always have an answer. It's the ones who tell you when they're sure, tell you when they're not, and know the difference. Confident wrongness is the fastest way to lose someone's trust, in a person or in a tool. And most software, especially most AI, is built to sound certain at all costs, because certainty demos well.

We're making the other bet. When BeMo shows an organization what it knows, it also says what it doesn't. Maryellen found that out by accident, mid-chat, on an unrelated question, and her note said more than any copy we've written: "I love that BeMo can tell you the 'gaps' that exist in your Knowledge Base."

She read that as a starting point, not a shortfall. Knowing what's missing is what tells her team where to go looking.

She also told us the gaps were too hidden and we should surface them sooner. Both things were true. That's the point.

And none of that feedback sat in a folder. Hers went to the top of our next founders meeting, alongside everything else our beta testers have told us, and it's been shaping what we build ever since.

Ask it something it can't know

BeMo didn't earn Maryellen's trust by being impressive. It earned it the slow way. Useful twice. Honest once, at a moment that mattered. Wrong in ways it admitted.

A few weeks ago, I wrote that software earns trust the same way people do. I believed it then as an argument. I believe it now as a description, because I watched it happen with someone who had every reason to go easy on us and chose not to.

If your organization's whole story lives in flyers and speeches, most of this sector's tools weren't built with you in mind. That should change, and not only by us. And if your organization is bigger than that, with a paid staff, an annual report, and a real grant calendar, the bar is exactly the same. We built BeMo for small and mid-sized nonprofits across that whole range, and I'd hold it to what follows too.

So here's the test I'd give any AI tool before trusting it with your organization, whatever its size: ask it something it can't know.

What it does next tells you everything. The ones worth keeping will tell you when they're guessing.