Share-out and close
Show what you built. Name what stays manual. Leave with a way to keep it working.
Share-out
Three or four teamsTwo minutes each
- What did you automate?One sentence: your smallest useful unit.
- What did it hit?The duplicate, the blank field, two answers in one box.
- What stays manual?The call you would not hand to a rule.
Share-out timer
2:00Ready: 2 minutes
Speaker notes
Pacing for the 25 minutes. 4:05–4:19 share-outs. 4:19–4:23 Keep it alive and Make it findable. 4:23–4:25 What stays manual. 4:25–4:28 Before you automate the answer. 4:28–4:30 Want more, then hand to Charles at 4:30 sharp.
Pick teams from different shapes if you can: one intake → triage, one pile → catalog, one many sheets → one table.
Press Start as each team begins. Amber at 30 seconds. At 0:00, thank them, Reset, next team.
If a team hit a planted problem, ask: rule, or manual review?
Running long? Cut sections 2 and 3 to their headline line and go straight to section 5. Do not eat into Charles's time.
Keep it alive
How would you know it broke?
From my course: “External systems change all the time.” An update ships, a feature gets renamed, a connection changes. Your automation stops.
Every client project has the same five parts
- Trigger
- Extraction
- Decision (rule or manual)
- Write-back
- Log
The last one tells you it broke.
Silent failures
No error. It just stops.
No new rows, and nobody notices. Build one check that notices silence.
Zell’s next step has the same idea: a weekly “no movement in 14 days” alert.
A simple log
Every row gets a status and a timestamp.
“done” when it worked. “needs review” with a reason when it didn’t.
If the newest timestamp is old, something stopped.
Who maintains it?
Name the person.
7 of 11projects raise cost and upkeep.
Marsal: the stewardship manager maintains it. Michigan Medicine: a monthly check of the interest list.
IDEA’s Adjust step asks:
- What could break if something changes?
- Who else is affected by this process?
- What ongoing maintenance is required?
CAI team: make this your evaluation question. How would we know it broke, or paid off?
Speaker notes
About two minutes. Land one line: the log is the part that tells you it broke.
Optional 30 seconds: my Rube Goldberg story from the Module 1 video. Six months later something changes, the whole thing breaks, and I have no idea why.
Point back to the hands-on recipe: the backlog function writes “done” plus a timestamp. That is a log.
Make it findable
Capturing information is automation.
You are automating your future self’s ability to find things.
1 · Easy enough that you’ll actually do it
Don’t: build templates, tagging systems and color codes.
The System Graveyard: the productivity apps you used enthusiastically for a week, then never opened again.
Do: a system you’ll use consistently, even if it’s imperfect. Quick capture, inside tools you already use.
2 · Add context to your notes
Avoid abbreviations. Write “conference”, not “conf.”
Write for someone else. In six months, that person is you.
Keyword vs. semantic search
Keyword search
Matches the exact words or phrases.
Semantic search
Matches meaning. A search for “Financial Forecast” also finds “Budget Planning” and “Resource Allocation”.
So you don’t have to guess your future keywords. Capture enough context that the meaning comes through.
Key takeaways
- Design for how you’ll actually search. Balance that against what’s easy to maintain.
- Don’t make an all-or-nothing switch. Start with one project.
- Use it for a week or two. Then honestly assess: am I actually using this?
Good enough and consistent beats perfect but complicated.
Which projects is this?
- Ginsberg: the catalog is what makes Maizey useful.
- FJA: a prompt library is a catalog, with names, tags and owners.
- Oakland County: retention tags on new records.
Speaker notes
About two minutes. This is Module 3, the Capturing and Accessing Information video.
Example if you want one: a calendar event called “Echelon dinner” with no other context. What on earth is this?
Ginsberg, FJA and Oakland County teams: this section is your project.
What stays manual
Straight from your clients’ projects.
- Accept, waitlist or denyZell
- Legal, regulatory and IP callsMTA
- What gets deletedGinsberg
- Expert review of the shortlistOVPR
- The coach’s read, and anything public-facingWomen’s Basketball
- Every FOIA responseOakland County
Intake → triagePile → catalogMany sheets → one table
When a row doesn’t fit the rules: mark it “needs review” with a reason. Don’t guess.
Some manual steps are the point. They keep you informed, build connection, or catch errors.
Speaker notes
Quick. Ask the room: which of these is yours? One hand per team.
The OVPR line is worth saying out loud: reviewer trust is the product.
The last line is Value of Manual Processes from Module 1: the report you assemble by hand may be how you stay informed.
Before you automate the answer
Consider that the current system may be fine, with no real problem to solve.
Sometimes the task shouldn’t exist at all. Maybe you don’t need an automated report. Maybe you need a five-minute conversation with your team.
Automation decisions require judgment, not just math.
Next · 4:30
Charles Kemp
Discovering the Problem Behind the Problem
Speaker notes
If you want the “Solving the Wrong Problem” slides on screen, this is the moment.
Hand-off line: before you automate the answer, is it the right problem? Then Charles.
Want more?
The course
Systems Thinking for AI & Automation
Part of the Applied AI: Data Analysis, Workflows, and Decisions series.
Today’s ideas come from it: the ROI question, IDEA, tool choice, and capturing information.
My ROI calculator
Is it worth the time to build?
Check before you start.
tinalasisi.github.io/roi-calculator-github-pages/
Try it on your own week: list five repetitive tasks. Run IDEA on each. Check the ROI. Pick one.
Speaker notes
Say the series name in full. Leave the calculator URL on screen while Charles sets up.
The last line is the Module 1 self-audit activity.
Share-out
Type as you go. It saves in this browser. The My notes button (bottom right) downloads everything from every page as one file.
Site: tinalasisi.github.io/automating-without-overwhelming-yourself · Source on GitHub