The NPO AI Capability Gap Is a Time Allocation Problem

TL;DR

Pick one NPO workflow, run it 10 times with guardrails, and measure results. Most leaders are dabblers; become an integrator by redesigning one workflow that matters.

Key points

• Time × Curiosity = Capability (reps beat vibes)
• Most NPOs are Dabblers at the Individual rung - move up one rung on one workflow
• Use the 4-point guardrail framework: data boundary, quality standard, human ownership, workflow rule

Most NPOs don’t have a tools problem. They have a reps problem.

Most NPO leaders aren’t avoiding AI because they lack access to tools. They’re avoiding it because they haven’t carved out time to build reps.

This post is a treasurer’s guide to moving from dabbler → integrator with guardrails: pick one workflow, run it 10 times, and measure what changed.

You’ll leave with:

• A simple maturity model (spectator → dabbler → integrator → AI-native)
• A repeatable “10 reps” method to redesign one workflow that matters
• A 4-point guardrail framework (data boundary, quality standard, human ownership, workflow rule)

Why now?

AI is not going away. It’s also highly polarizing. Some people see it changing the future of knowledge work. Others see it producing endless slop [1]. Both views have merit.

There is a pervasive anxiety many feel towards AI. In my experience the loudest critics have spent the least time understanding what it’s actually useful for. If you’re leading an NPO, you need to know what’s happening, not just what the critics say. Some people inside your organization will have a visceral reaction to AI. This is OK - give them standards, structures and show them you are intelligently implementing AI into workflows and you will do a lot to ease concerns.

The formula is Time x Curiosity

Curiosity is your willingness to try without waiting for someone to tell you what to do.

Time is self explanatory, you need to carve out time for yourself to experiment.

I think time is probably the biggest blocker for most people I chat with. I get it, you are busy. Some people are genuinely so busy they can’t find 20mins a day to experiment. Most of us aren’t though. This is a convenient excuse or fallback. Block your calendar, spend some time in your evenings on personal workflows. I promise you will reap some rewards if you carve a little time.

Curiosity is not to be discounted. As an NPO leader you should look for curiosity as a trait of future leaders and staff as you interview. It’s going to be an increasingly important trait for success in this new strange world of knowledge work we are heading into.

I like the gym analogy for building capacity. Get your reps in! Pick a frontier model [2]. OpenAI, Anthropic, Google each have a great frontier model that is accessible at the basic subscription price (usually 30-50 CAD per month per user). Don’t split hairs here, try out something and stick with it for some time. If you switch your gym program every week you are going to flounder and not see progress.

Think of a workflow (a muscle group or capability to train) and try approaching problems with AI through conversation and tooling. I ran ‘summarize this grant guideline into a checklist’ 10 times over two weeks. First three attempts were generic. By attempt seven, I’d learned how to constrain tone and format. Now it’s a 15-minute workflow that used to take 45.

Maturity curves

There are two ways to think about AI maturity in your organization: where you are strategically (org capability) and where you are tactically (day-to-day workflows). Most NPOs are further behind than they think on both.

A) Org capability ladder (strategic maturity)

This is about how AI fits into your organization’s operating model:

1. Individual productivity Faster drafting, summarizing, ideation. Low governance, people using tools quietly.

2. Team coordination Shared prompts, templates, review standards, common tooling across the team.

3. Org operating model Policy, data boundaries, ROI logic, training programs, intake/routing systems.

4. Sector collaboration Shared playbooks, safe datasets, shared procurement leverage across organizations.

Most NPOs are at the Individual rung- staff using AI quietly, no shared templates, no governance. That’s fine for now. But if you want compound wins, you need to move to Team: shared prompts, review standards, and common expectations for what “good” looks like.

B) Workflow maturity ladder (day-to-day use, tactical)

This is about how you personally use AI day-to-day:

1. Spectator Rare use; mostly hears about risks. Maybe a detractor. Didn’t carve out the time or curiosity.

2. Dabbler Ad-hoc prompts; novelty phase; inconsistent outcomes. No real pay-off other than flashes in the pan.

3. Integrator A few redesigned workflows; templates; measured time saved. Now we are seeing some benefits.

4. AI-native AI is default for drafts/triage/research; humans own decisions. Possible multiples to productivity.

Most NPO leaders are Dabblers - they’ve tried ChatGPT a few times, got inconsistent results, and stopped. That’s not evaluation; that’s insufficient reps. The move from Dabbler to Integrator is picking one workflow and redesigning it. That’s where the capability gap closes.

Tying the two together

Here’s what matters: you can be AI-native in one workflow while being a spectator elsewhere. An org can have world-class grant summarization while never touching donor communications. Maturity isn’t binary - it’s workflow-specific.

It’s also important to take a step back and realize the goal isn’t to be AI-native everywhere. The goal is to move up one rung on one workflow that actually matters to your work. It’s frankly impossible to be AI-native everywhere in a large organization with the amount of process friction you need to overcome.

My “integrator” moments (applied AI)

My meatiest project is a timeline constructor. I built a tool (or co-built with AI) to capture and structure evidence for insurance matters. This job is an absolute nightmare and the kind of thing I lose sleep over, especially if there are facts or numbers hallucinated. I needed a deterministic (read: repeatable, same thing every time with the same evidence input) tool to build a timeline and collect and organize evidence - photos, emails, discussions, calls. The tool takes scattered evidence and outputs a chronological timeline I can verify and share. What used to take 8+ hours of manual sorting took about 90mins to scope and 30 seconds to run each time I want to add new evidence.

Another way I was able to use AI as leverage in a crisis was to meet CRA reporting requirements by issuing credit notes to hundreds of members shortly after a sudden equipment failure blocked the Club’s normal operational start. I could have bought software, but that would mean spending money I’d later refund to members. I needed a lifeline so I decided to build the tool to notify all members from a trusted dataset I had access to of their refund amount with a serialized number and time and place to pickup. This would have taken dozens of hours manually to mail each member, it took less than 4 to scope and build. Note this is an abstract use of AI, I didn’t feed sensitive info into a chat, I used a coding assistant tool to help spec and build a program and test that program with dummy data. This is the integrator tier. Once you get comfortable knowing the strengths and weaknesses of your toolkit, you reach for it when it can really make an impact.

AI in my experience as a treasurer has been unfortunately useless at numerical/quantitative tasks. Bank recs? Tried two frontier models and both caused more harm than good. This was sad to me because this is the type of work I wanted to offload to AI where I could. This is why you should be skeptical of anyone promising AI will handle everything. It won’t - at least not yet. It is not there yet for “agentic” work right from your browser.

Funny enough this website was my pet project at the start of this year, entirely built through frontier coding assistant tools like Codex and Claude Code. For me the proof of these tools being valuable is front and center when I use it to build something that was previously beyond my capability. Think abstractly, you can build software much easier than before if you are able to get comfortable using AI beyond the chat interface. These tools I would classify as advanced or more technical because they interact with code or a terminal (this scares people if they are not familiar). But it’s a simple function of curiosity and time to explore what is possible. You may surprise yourself like I did.

Guardrails and operating principles

The not-so-fun part of AI is the policy implications. Going back to the strategic maturity ladder, there’s more friction around policy at each rung. An individual volunteer doesn’t need strict policy to help with communications. A complex email routing tool built in-house for donor feedback will need serious de-risking for security, false positives, and general snafus.

The good news: AI policy is just another form of org culture. Having something consistent to point to helps your staff feel like you’re handling AI intelligently, not letting it run wild.

If you aren’t comfortable drafting policy yet, stay in the Dabbler tier before making a leap. Learn the strengths and weaknesses of your chosen toolkit before you draft a policy that’s outdated before you can put it into practice—or so onerous it scares people from using AI at all.

When you’re ready, here’s a simple 4-point framework:

1. Data boundary: what never goes in I don’t paste donor names, financials, or internal board discussions. Public grant guidelines? Fine. Internal strategy docs? No.

2. Quality standard: what “good” looks like For variance commentary, “good” means accurate numbers, board-appropriate tone, no jargon. I verify every claim before approving.

3. Human ownership: who approves, who’s accountable AI drafts, I decide. If the output goes to the board or a donor, I own it—not the tool.

4. Workflow rule: if it adds steps, redesign it If using AI creates more work than it saves, I stop. The goal is to remove steps, not add review loops.

I want more

I’m presenting this at Baker Tilly Nova Scotia’s NPO Speaker Series on February 5th, 2026. If you can make it, I’d love to chat.

Free resources: I’ve put together a starter pack with guardrails, workflow templates, and a 30-day reps tracker: ai-starter-pack

If you want help implementing this in your organization, reach out after you’ve run the 30-day experiment.

Glossary / definitions

[1] Slop - Merriam-Webster’s word of the year for 2025. Their definition: digital content of low quality that is produced usually in [large] quantity by means of artificial intelligence.

[2] frontier model - As of January 2026 these would be GPT-5.2 Thinking/Pro (OpenAI), Claude 4.5 Opus (Anthropic), Gemini 3 Pro (Google/Alphabet). A great way to track model performance is https://lmarena.ai/leaderboard

Each of these models is near parity and extremely powerful compared to what was available 12 months ago.