The Calm Infrastructure Lab | LUR Growth Issue 003 | October 2026 Practical AI. Stronger systems. Room for the mission. This issue: Before AI sends the report, decide who approves it. Grant reporting is one of the first places AI shows up for small nonprofit teams. A first draft can help. The question that has to come before the tool is simpler: When the report goes to the funder, who approved it? Not who wrote it. Who looked at every number, confirmed it was true, and said yes, this can go out under our name. The short process for this room: 1. Sources — only named places the report may draw from; personal info stays out 2. AI draft — every figure cited; no guessing 3. Staff review — owner + backup; approval in writing 4. Submit — a person submits; the AI tool never does I put this on a Grant Reporting Control Sheet. Both versions are attached: a blank template and a fictional filled-in example. They're also in the Classroom. A longer walkthrough on Substack is coming soon. If you try this on your next report, reply with what was hard to fill in. That is how this tool gets better for this room.

Quick one for this week, friends. I just recorded a video on something I see in almost every organization: the monthly funder report scramble. And the thing I keep coming back to is that it's rarely a technology problem. It's an ownership problem. Here's the idea. Before you add any AI or automation to your funder report, name one person who reads it before it goes out, and name a backup for the month that person is out. One reviewer, one backup. Two names, written down. Why? Because automating a shaky report just makes it wrong faster. A named reviewer keeps speed from turning into risk. The five steps I walk through: - Write down what you promised - Map where each number lives - Gather it once - Draft from approved figures, in your own voice - Human review And I'm honest in the video about when you should not automate at all. Sometimes a spreadsheet and a good calendar reminder is plenty. Your homework takes about ten minutes. Open your next funder report and write down the reviewer and the backup. If you can't fill in both, that's your first project. Full video: https://youtu.be/eDu94GyVaMs More at lurgrowth.com So tell me: which part of your funder report is the biggest scramble for your team right now?

There is a lot happening in AI. In Calm Infrastructure Lab, I’d like us to bring the conversation back to the work our organizations need to do. What is one workflow inside your nonprofit that you would fix before buying another AI tool? It might be preparing a grant report, assembling a board packet, following up with partners, or finding information that seems to live in one person’s head. If you’re comfortable sharing, name the workflow and the part that puts the most pressure on your team. Keep names and sensitive information out of the example. We can use the conversation to distinguish what needs a clearer process, what ordinary automation could handle, and where AI might be useful. One workflow is enough to begin. I’ve attached One Workflow First, a two-page fillable guide you can use with your team. Map one recurring process, choose a small improvement, name the owner and reviewer, and compare the workload before and after. You can type into the PDF or print it. Latoya

Practical AI. Stronger systems. Room for the mission. The Calm Infrastructure Lab | LUR Growth Issue 002 | September 18, 2026 This week: build a board packet someone else can assemble A board packet can look like a document, but it is really a chain of decisions. Someone has to know which reports belong in it, where the approved numbers live, what changed since the last meeting, who can see sensitive information, and who decides that the packet is ready. When those decisions live in one person's head, the packet becomes heavier every month. AI may help gather approved material or create a first draft, but it can't fix unclear ownership or decide what belongs in front of the board. This week, the goal is simple: make one section of your next board packet clear enough for another person to assemble safely. Three updates worth understanding 1. ChatGPT can work with more connected business data: On September 10, OpenAI added a Data plugin to ChatGPT Business. It can help eligible teams analyze connected business data and create reports or dashboards. OpenAI notes that workspace administrators control installation and that connected-account permissions still apply. Read the ChatGPT Business release notes: https://help.openai.com/en/articles/11391654-chatgpt-team-release-notes My practical take: access is not the same as approval. Before connecting a source, name which board questions it supports, which fields are needed, and who is allowed to see the result. A faster report is not useful if it draws from the wrong source or exposes information the reader should not receive. 2. Microsoft is treating AI governance as an ongoing operating practice: Microsoft's September 1 Responsible AI update describes governance that follows the full lifecycle of an AI system. It highlights tool permissions, action monitoring, practical controls, and the people responsible for implementing policy. Read Microsoft's Responsible AI update: https://blogs.microsoft.com/on-the-issues/2026/09/01/responsible-ai-in-2026-how-we-are-adapting-for-whats-ahead/

If AI is on your list but you are not sure where to start, I have put together a practical starter for you in CALM Starters. The free Nonprofit AI Triage Pack includes: • Three copy-and-paste prompts to find a capacity drain, turn notes into an action list and check a draft against its evidence. • A fictional example you can try without sharing organizational records. • A printable prompt pack and a First-Win Worksheet. Set aside 15 to 20 minutes. Pick one prompt, check the result and record your next action. Get the free starter here: https://www.skool.com/the-calm-infrastructure-lab-3421/classroom/2f442ab6?md=0b4a690cdf204e208d5dee6a1c7f872e After trying the starter, tell us: what task did you choose, and what did you need to correct? Keep your examples general and leave private records out of your comment.