What we observed from 86 real automation projects

Abstract network of glowing copper and peach data pipelines on a dark background

Most automation content comes from vendors or analyst surveys, not real requests. So over several weeks we read 86 real automation and AI-agent jobs and tasks posted by businesses — what they asked for, the tools they named, the budget, the language they used. This is a read of that sample, not the whole market.

IWhat's actually being asked for

Strip away the buzzwords and most requests cluster into a handful of recurring jobs. Lead management and follow-up — get a lead into a system, qualify it, keep it moving without a human dropping it — is the single biggest bucket. Close behind is a broad, often under-specified ask: "build me an AI agent" or "automate our workflow" with the actual scope left for a discovery call. CRM integration and sync and AI voice agents round out the clear leaders; after that the tail gets long fast.

Lead management & follow-up

17%

AI agent / broad workflow builds

14%

CRM integration & sync

AI voice agents

Everything else (17 smaller categories)

56%

What 86 real automation jobs and tasks were actually asking for, grouped by primary use case. The tail is long: 17 more specific categories — invoice processing, content generation, booking systems, and more — each made up 5% or less on their own.

Lead management and CRM sync combined make up roughly 1 in 4 requests — automation demand is still mostly about getting data to move correctly between systems, not exotic AI use cases.

IIThe tools that keep coming up

Whatever the use case, the vocabulary is consistent. Businesses describe what they want in terms of the same handful of platforms:

n8n

61%

Make.com

58%

Zapier

47%

Google Sheets / Workspace

30%

Claude / Anthropic

24%

OpenAI / ChatGPT

24%

Airtable

20%

GoHighLevel

19%

Share of jobs/tasks that named each tool or platform. Most requests named more than one, so this doesn't sum to 100% — it's a read on which vocabulary businesses already reach for when they describe what they want built.

n8n, Make, and Zapier are still the default reference points even in requests that are fundamentally about AI agents — the no-code layer hasn't gone away, it's absorbed the AI layer. Claude and OpenAI appear in roughly 1 of every 4 requests, almost always as the reasoning step inside a workflow, not as a replacement for one.

IIIAI is table stakes, not a differentiator

44%

required an AI/LLM layer

not just app-to-app plumbing

56%

framed as ongoing work

a retained function, not a one-off build

21%

were fix-it jobs

stabilizing something already built, not greenfield

64%

came from 5 countries

US, Germany, UK, Australia, Canada

Four more numbers from the same 86-request sample.

Nearly half of the requests we reviewed needed an actual AI/LLM layer — extracting data from messy text, drafting a reply, qualifying a lead in natural language — not just moving a record from A to B. That's no longer the exotic ask it was a couple of years ago; it's baseline scope for a meaningful share of automation work.

The other two numbers say something about how the market is maturing. Over half of requests were framed as ongoing work, not a one-off build — automation is increasingly treated as a retained function, like bookkeeping or IT support, rather than a project with an end date. And roughly 1 in 5 requests weren't asking for something new at all — they wanted an existing automation that had broken or drifted fixed and stabilized. Enough automation has been built badly enough, by enough people, that maintaining it is now its own category of demand.

IVWho's asking

Real estate

E-commerce / retail

Tech / SaaS

Marketing agencies

Other identified industries

44%

Not stated / unclear

31%

Client industry, where it could be identified from the listing itself. Real estate is the single largest identifiable vertical, but industry was genuinely unclear in roughly 3 of every 10 requests — automation buyers tend to describe the workflow, not their business category.

Real estate is the single largest identifiable vertical — lead routing, AI voice agents, and CRM work for agents and brokerages show up again and again. E-commerce/retail and tech/SaaS tie for second. But the honest finding is the last bar: in roughly 3 of every 10 requests, the client's industry genuinely wasn't identifiable from the listing. Businesses describe the workflow they want fixed far more readily than they describe their own business — which matches what we'd expect from buyers who think of automation as an operations problem, not an industry-specific one.

Geographically, demand is concentrated: the US alone accounts for about a third of requests, and the US plus Germany, the UK, Australia, and Canada together account for 64% of everything we tracked.

VWhat this means if you're evaluating automation help

A few practical takeaways fall out of this, whether you're buying automation work or building it:

"AI automation" is not a specialty anymore — it's the baseline. If a request needs an LLM in the loop, that's normal scope now, not a premium add-on.
Budget for maintenance, not just the build. A meaningful share of this market exists because someone's first automation broke. Plan for an owner, not just a launch date.
Most of the value is unglamorous. Lead routing and CRM sync beat voice agents and chatbots by a wide margin. The flashy use cases get the attention; the plumbing gets the budget.
Vague scope is common, not a red flag. A large chunk of requests read as "we know we need this, we don't know exactly what 'this' is yet." A short discovery step before committing to a build is normal, not a stall tactic.

None of this is exotic. It's what happens when a technology moves from novelty to infrastructure: the requests get more ordinary, not less — which is exactly the point where it's worth having a partner who treats automation as an operations discipline instead of a demo.