We have covered a lot of ground in the last few weeks. If you’ve missed any of the installments, I encourage you to take a look.
- Week 1: The Human-Centered AI Framework: Five Principles for Fundraising with AI at Your Side
- Week 2: Your Monday Morning AI Review, Where the Human Leads
- Week 3: Is Your Donor Database Missing the Most Important Information?
- Week 4: The Human-AI Partnership That Transforms Annual Giving
- Week 5: The AI Briefing That Told You Not to Make the Ask
- Week 6: The Thirty-Second Rework That Makes an AI Draft Human
- Week 7: When the AI Connects the Dots on a Forgotten Donor
- Week 8: Five Questions Your Board Will Ask About AI
- Week 9: The Honored-or-Handled Test for AI Messages
To recap, we have walked through the Human-Centered AI Framework. We know the four pipeline stages, the Rule of Seven, the six-step solicitation process, the honored-or-handled test, and the donor-as-hero principle. We understand the Monday morning review and what a good briefing looks like.
But our relationship-building system doesn’t know any of it yet.
Now Available
In Holding Fire: A Skeptic’s Framework for the Fundraiser with AI at Your Side, Stephen Christopher Nill, JD, lays out the Human-Centered AI Framework—his five-principles approach to keeping humans at the center of every donor interaction, even as AI quietly does the work no one fundraiser could ever do alone.
This book was written by a skeptic, for skeptics.
Before our first real Monday morning can work, the AI inside the RBS has to be familiar with the framework. Otherwise, the answer to the opening question of Chapter 1 of my book, Review my full donor portfolio. Flag anything that needs my attention this week and draft responses where appropriate, comes from whatever the platform’s default logic provides, which usually means generic recommendations that read nothing like a human-centered briefing.
The work of installing the framework in the AI is simpler than it sounds, and once it’s done, the AI operates by the framework’s rules. Pretty cool, huh?
The Specification
The Human-Centered AI Framework Operating Specification is the rulebook for a framework-aware AI. The Specification covers the pipeline stages, the Rule of Seven counting rules, the inference methods for thin records, the output formats for a Monday morning briefing, and the ethical hierarchy with its honored-or-handled test. Every rule is written for the AI to apply directly to a donor record.
You’ll find the Specification at charitychannel.com/framework. When your RBS’s AI reads it, the framework moves from the book (and these articles) into the platform that holds our donors.
Four Tiers of Platform
How we show the Specification to the AI depends on what the platform supports. Most platforms fall into one of four tiers. Each tier has a different method to get the AI to read the Specification:
Tier 1, the fully configurable platform. Some platforms give us a place to save standing instructions the AI carries into every conversation. The field has different names by vendor (custom instructions, configuration directive, operating context), but the idea is the same. We paste the Specification once, save the field, and the AI reads it before every briefing, every solicitation readiness review, and every donor draft. This is the strongest setup available, and the one where the framework most quickly becomes how the AI works.
Tier 2, the upload-and-reference platform. Some platforms don’t have a persistent-instruction field but do let us upload a document the AI can read. We upload the Specification as a PDF or a text file, and at the start of each session we ask the AI to operate by what it finds there. Some platforms hold the uploaded content steady across long sessions, and some let it drift. A short reminder at the top of major queries keeps the AI grounded if the output starts sounding generic.
Tier 3, the chat-and-paste platform. Some platforms have neither field, and their AIs start each conversation fresh. Most general-purpose AI assistants work this way, too. For these, we paste a shorter version of the Specification, called the preamble, at the start of each substantive session. The preamble carries the load-bearing rules in a length most platforms accept in a single message. The ready-made preamble is on the framework page; we copy it once, save it where it’s reachable, and paste it in at the top of each Monday morning.
Tier 4, the closed-methodology platform. Some platforms apply their own engagement methodology, prospect scoring, or recommendation logic, and the methodology is built in rather than configurable. We can paste the Specification at the top of every query and the AI will adjust at the margins, but the underlying logic stays the vendor’s. The pattern that works in Tier 4 is to use the platform for what it does well, like record-keeping and contact logging, and run framework-aligned AI work in a separate environment, with appropriate handling of the donor data.
The full breakdown is on the framework page, including the diagnostic questions that place any platform on the map, the preamble, and the vendor-facing notes.
When the RBS Doesn’t Have AI Yet
Not every commercial RBS has AI capabilities yet, though the major platforms are moving fast. If the platform doesn’t have AI today, the framework still has a place in the work, and the months until AI arrives are not lost time.
A general-purpose AI assistant works alongside the RBS in the meantime. The workflow is more manual. We copy a donor record from the RBS, paste it into the assistant along with the preamble, ask the same question we’d ask a built-in AI, and bring the relevant pieces back to the RBS. The session is less seamless than a built-in AI offers, but it produces the same kind of human-centered briefings and drafts.
This is also a good time to talk with the vendor. We say that we’re working under the Human-Centered AI Framework and that we’ll evaluate the AI features by how well they let us operate the framework. The vendor-facing notes at the framework page are written for exactly such a conversation; forwarding them along is how a vendor gets the message. (Vendors, are you listening?)
The First Monday, After the Setup
The Specification is guiding the AI through one of the methods we just looked at: The persistent field is set, or the document is uploaded, or the preamble is pasted into the chat box. The donor records are in their current state, complete in some places and thin in others. It’s Monday morning, the coffee is ready, and we sit down to do the work.
The briefing won’t look exactly like the polished one from Chapter 1 of my book. Real donors and real records produce a different experience from a teaching demonstration. The first briefing will reveal both the situations the AI caught well and the gaps in the record where the AI couldn’t say much. Some donors will appear with confident recommendations grounded in our notes. Others will appear with the AI flagging that the record is thin and asking us to verify what’s been logged. The Specification tells the AI to behave that way when data is sparse, and the honesty serves us better than confident-sounding guesses would.
Most first drafts produce two reactions at the same time. The draft is impressive, and the draft is not quite right. Both reactions are accurate. The AI provides structure and tone drawn from the record, and our edits provide the specifics the AI couldn’t see, including the detail about last quarter that resonates with this donor, the warmer opening line that sounds like us, and the reference to a conversation three months ago that hasn’t been logged yet.
When we make those edits, we use the next 30 seconds to capture the new detail in the donor record before moving on. The sentence that sounded too generic to send was generic because the record didn’t have the personal detail we just added. Next week the AI’s draft for that donor will be better. The Virtuous Data Cycle is working in our real RBS now, not in a teaching example.
By the time the coffee is finished, every donor who needed attention this week has gotten it, the gaps in the record are fewer than they were, and the AI is better positioned for next Monday.
The framework was built so that one Monday morning a week could carry the Scale Shift, and the Specification is what lets the RBS carry it with us.
In my book, Holding Fire, Chapter 10 walks through this whole installation in detail, including the diagnostic questions for placing any platform on the tier map, the preamble for Tier 3 work, and the conversations to have with vendors whose AI isn’t there yet.
Next week, I’ll take our discussion in a new direction. Instead of walking through one chapter per week, I’ll explore specific fundraising challenges where ideas from multiple chapters converge, starting with what a Monday morning with AI actually looks like.
Stephen Christopher Nill, JD, founded CharityChannel in 1992, building it into one of the largest online communities of nonprofit-sector practitioners in the world. He has edited, designed, and published dozens of books for the nonprofit sector through CharityChannel Press. Over four decades, Stephen has built, administered, and advised fundraising programs in the United States and internationally that have raised billions of dollars. He has served as senior vice president of development at a hospital system, chief advancement officer at a West Coast university, director of development at a parochial school he helped to found, and founder and executive director of a charity that feeds the homeless in Southern California.
The Grant Professionals Association presented him with its first President’s Award. The Grant Professionals Certification Institute later honored him with the Pauline G. Annarino Award for foundational contributions to its existence. Stephen has been a frequent commentator on charitable giving for CNN, Fox News, MSNBC (now MS NOW), and other news shows.
He has provided training to attorneys on advanced charitable gift planning techniques on behalf of the State Bar of California’s continuing legal education program. As a lawyer and consultant, he advised hundreds of nonprofit organizations and educational institutions.
Stephen’s engagement with artificial intelligence reaches back to the 1970s, decades before the current wave. Holding Fire: A Skeptic’s Framework for the Fundraiser with AI at Your Side draws on that long history and on his career inside development offices.
He writes weekly to development professionals at charitychannel.com. These days he’s busy writing his first novel and composing symphonic music in his music studio.
