A GM walks into the shop at 7 a.m. The first 30 minutes used to be: open CCC, check yesterday's deliveries, pull the WIP list, scan CSI, glance at the schedule, eyeball the parts queue. Five or six different screens. Half-information assembled into a mental picture.
Most GMs don't do all of that. They do two of the six things and guess at the rest.
An AI-generated ops brief flips the script. Pull the overnight data, have an LLM write a structured narrative, deliver it to the GM's inbox before they sit down. Five-minute read. Complete picture.
Here's what a good brief looks like, what it actually replaces, and how to build one.
What a Good Ops Brief Looks Like
An opening summary (2–3 sentences). Then a structured breakdown. Example format:
Morning Brief — Phoenix Shop — Tuesday, April 20
Yesterday was strong on deliveries (7 vs 5 target) but WIP grew to 18 days aged. Three supplement approvals are blocking production on #RO-12481 and #RO-12488—both need carrier follow-up today.
Deliveries: 7 yesterday. 28 MTD vs 25 target. On pace.
Production: 12 cars in WIP. 4 aged over 14 days: #RO-12431, #RO-12445, #RO-12457, #RO-12471. Two of these are waiting on Allstate supplement approval (submitted last Wednesday).
Today's schedule: 6 intakes, 3 deliveries scheduled. Tech capacity is tight—Jose is out, Martinez and Huang are doubled up.
What needs your attention: Supplement follow-ups on 12481 and 12488. Customer callback on 12457 (asked for ETA update). Bay 4 is empty and we have work assigned to it—Nguyen is on paint rework.
This is information a GM could assemble in 30 minutes. The brief assembles it in seconds and delivers it already prioritized.
What It Replaces
Nothing formal—because most GMs don't do a formal morning review. That's the point.
The brief replaces the fuzzy mental picture that GMs assemble from whichever dashboards they happened to open. It's a forcing function: the same information, in the same structure, every morning, whether the GM is a spreadsheet obsessive or a floor-walker who hates logging into anything.
At an MSO, the compounding benefit is consistency across shops. Every GM sees the same structure, which means leadership can ask the same questions across shops and get comparable answers.
What's Actually Happening Behind the Scenes
Three layers:
1. Data Layer
Your warehouse has yesterday's data. Cycle times, WIP ages, delivery counts, supplement statuses, customer messages, CSI responses, schedule conflicts. Most of this is already there if you've done the warehouse work.
If your data isn't in a warehouse yet, this is the first blocker. You cannot build a brief on top of CCC-exported Excel files.
2. Query Layer
A set of SQL queries produces structured JSON describing the state of the shop: {deliveries: 7, deliveries_target: 5, wip_count: 12, wip_aged_over_14: [...], scheduled_intakes_today: 6, ...}.
This is mechanical. Write it once per KPI. Run nightly.
3. Narrative Layer
An LLM takes the JSON + a prompt that describes the brief format + a few examples of what a good brief looks like. Output is prose.
This is the AI part, and it's actually the easier layer. Prompt engineering gets you to a credible brief in a week. The hard part is the first two layers.
What Separates a Good Brief From a Bad One
Specificity
A bad brief: "Production is trending down this week."
A good brief: "Production is down 12% week-over-week. Driver is the Maple St shop (deliveries down from 11 to 6), where two techs were out Tuesday and Wednesday. Phoenix and Dallas are on target."
The difference is whether the LLM has access to the underlying numbers. Give it the JSON, tell it to cite specific ROs and shops, and it does.
Prioritization
A bad brief lists everything. A good brief puts "what needs your attention today" at the top.
This is prompt work: tell the LLM "start with the 2–3 things the GM must act on today, then provide the context."
Actionability
"Allstate supplement approval is blocking two ROs" is useful. "Supplement-approval-cycle-time is 3.2 days" is not. GMs act on specifics, not averages.
Brevity
A good brief is 150–300 words. Longer and the GM skims it. The discipline is the prompt: cap length explicitly and tell the LLM what to cut first.
Honesty About Unknowns
If a data feed didn't update overnight, the brief should say so. "CSI feed from Carrier X is 36 hours stale—treat with caution." This is a simple pre-check on data freshness before the brief runs.
Deployment Patterns
Delivery Channel
Email at 6 a.m. works for most MSOs. Some deliver into Microsoft Teams, Slack, or a custom dashboard the GM opens first thing. The channel matters less than the time: the brief must land before the GM's day starts.
Personalization
One brief per shop, addressed to that shop's GM. Regional managers get an aggregated brief covering their cluster. Corporate gets a roll-up across all shops.
Building this once is marginal additional cost compared to building one. Do it right from the start.
Feedback Loop
Add a "this brief was helpful / not helpful" link. Track it. Adjust the prompt over time based on what GMs actually read.
Where This Breaks
Data Quality
If your warehouse has bad numbers, the brief has bad numbers, and it looks worse than no brief because it's presented authoritatively. Do not build a brief on top of a flaky pipeline.
LLM Hallucination
Occasionally an LLM will cite an RO number that doesn't exist or misread a number. Validate the output: every fact in the brief should be traceable to the input JSON. Production briefs should include this validation pre-flight.
GM Skepticism
If GMs don't trust the first few briefs, they stop opening them. Build trust by letting the GM click into any number to see the source data.
The ROI
A 5-minute read replacing 30 minutes of manual dashboard-tour is 25 minutes per GM per day. At a 10-shop MSO, that's 4 hours of GM attention per day, redirected from information-assembly to action.
Compounded across a year, that's roughly 1,000 hours of leadership attention. That's the ROI. Not the LLM cost, which is negligible.