# Only 13% of 3,362 automation case studies use any AI at all. Agents show up in 18. — retained record

> Retained reddit record from reddit:AI_Agents.

Canonical: https://highsignal.app/data/records/b8074ce046ba64d2

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:AI_Agents)  published 2026-10-19 02:09 ingested 2026-10-20 02:35

[original source ↗](https://www.reddit.com/r/AI_Agents/comments/1x9kto9/only_13_of_3362_automation_case_studies_use_any/)

## retained text



```
So I've been putting together a database of automation case studies for the past few weeks. Vendor-published stuff, the kind where a named company uses some software on a specific process and reports a result. Ended up with 3,362 of them.

Out of those:

* 436 (13%) claim any AI at all (ML, genAI or agents)
* 84 are generative AI
* 18 have an AI agent, and 7 of those are customer support

The rest is regular software. The most common project is just getting two systems to talk to each other, like the CRM pushing data into accounting. Then ERP rollouts, customer portals, scheduling and dispatch tools. Pretty boring stuff.

AI shows up a lot in customer support (41% of those cases) and in invoice and document processing (about a quarter). Barely at all in payroll, scheduling, onboarding or warehouse work, like 3 to 4%. And the AI cases don't report results any more often. Time saved shows up in 61% of the AI ones vs 58% of the rest, and the non-AI ones mention cost savings more (36% vs 29%).

Reading this sub you'd think agents are running half the businesses out there, and in the published wins it's 18 out of 3,362. I think a lot of the "I built an AI agent that runs my whole business" posts on here would work just as well as a scheduled script and a few if statements, it just doesn't sound as cool.

And when you look at where the AI ones are, it's support tickets, invoices, contracts. Stuff where every one comes in a little different and a rule can't read it. The rest of what's in there is plumbing. Someone used to copy the order from an email into the ERP, then into the invoicing tool, then into a spreadsheet for the boss, and now the systems just pass it along. Even in a lot of the AI cases, the model reads the messy email and then regular integration does everything after that. I think that's why the AI ones don't report better results. The model is a small piece of the project, and most of the hours come back from the boring part around it.

It reminds me of the electricity story, if you've heard it. When factories first got electric motors, productivity barely moved for decades. Paul David wrote a well known paper about it in 1990. Factories pulled out the big steam engine, put one big electric motor in the same spot, and kept the same multi-floor layout with all the belts and shafts. The gains only showed up once people redesigned the whole floor around the work, a small motor at each machine, everything laid out by how the product moved.

I think a lot of companies are in the one-big-motor phase with AI right now. They drop a model into the exact workflow they already had, same handoffs, same copying between systems, same approvals, and then wonder why nothing changes. That's probably a big part of why MIT found around 95% of company genAI pilots with nothing to show on the P&L. The pilot is the motor in the old spot. In these case studies the ones with results mostly did the boring redesign first, and the model, when there was one, came last.

Plumbing doesn't sell though, so that's not what anyone pitches.

Before anyone says it:

* these are vendor success stories, nobody publishes the failures, and the numbers are self-reported
* it's what vendors chose to publish, so it's a skewed sample of what's out there
* I counted something as AI if the source said so, so 13% is probably generous
* most of these aren't dated and the dated ones lean 2015 to 2020, so genAI is definitely undercounted

curious to know your ideas on this
```



document fetched 2026-10-20 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1x9kto9/only_13_of_3362_automation_case_studies_use_any
