TL;DR
AI automation work isn't prompt engineering. It's plumbing, process design, and knowing where automation breaks. Here's what the job really involves and what it's worth.
→ See how this applies to your business (free 30-min call)AI automation work sounds like it should involve models. In practice, on a normal engagement, the model is maybe five percent of the job. The other ninety-five is figuring out how a business actually works, connecting systems that were never designed to talk to each other, and deciding where a machine should stop and a person should start.
That gap between the perception and the reality is worth understanding, whether you're hiring for it, buying it, or considering doing it.
What the Work Actually Is
An AI automation project has four phases, and their time allocation surprises people.
Phase 1: Process archaeology (30 percent of the work)
Before anything gets built, you have to find out what actually happens — not what the org chart or the SOP says. Where do leads really come from? Who really answers the phone after 5 PM? What does the office manager do with the sticky notes?
This phase produces the most valuable finding of most engagements, and it isn't technical. In over half the businesses we audit, we find at least one live lead source that nothing is connected to. Google Business Profile messages going to an unwatched inbox. A tracked number on a truck wrap that rings a disconnected extension.
Fixing that finding often produces more revenue than the automation does.
Phase 2: Plumbing (40 percent of the work)
Connecting systems. Forms to CRM, calls to CRM, ad platforms to CRM, CRM back to ad platforms. Handling the source that has no native integration. Deciding what to do when two systems disagree about who a contact is.
This is unglamorous and it's where projects live or die. An agent can only act on what it can see, and the most common cause of a disappointing AI deployment is that it was connected to a third of the leads.
Phase 3: Agent and workflow design (20 percent)
Now the AI part. Writing the qualification script — which is really just encoding the five questions your best closer already asks. Setting guardrails: what the agent must never say, what price bands it can quote, when to hand off. Building the escalation path. Designing the follow-up sequence.
The skill here is restraint. Narrow jobs work. "Qualify inbound leads and book the good ones" works. "Handle customer relationships" does not.
Phase 4: Tuning (10 percent, forever)
Reading transcripts weekly. Finding the question that confuses people. Discovering that a segment you were disqualifying is actually profitable. This never ends, and it's where the difference between a mediocre system and a good one accumulates.
The Skills That Matter
Ranked by how much they determine project success, which is roughly the inverse of how much they're discussed.
1. Business process literacy. Can you sit with an owner for an hour and correctly identify where money is leaking? This is the highest-value skill in the field and it is not technical.
2. Systems integration. APIs, webhooks, authentication, data mapping, error handling. Practical plumbing rather than computer science. What happens when the webhook fails at 3 AM and nobody notices for a week?
3. Platform depth. Knowing one automation and CRM platform genuinely well beats knowing six superficially. We standardize on GoHighLevel for client builds specifically because depth in one object model compounds — you stop rebuilding and start deploying.
4. Conversation design. Writing scripts that sound like a person, handle interruptions, and disqualify politely. Closer to sales writing than to engineering.
5. Measurement discipline. Knowing that cost per lead is a trap and cost per booked job is the number. Being able to build the report that proves the work paid for itself.
6. Prompt and model knowledge. Genuinely useful, genuinely not the bottleneck. It's the last five percent, and treating it as the whole job is the signature of someone who hasn't shipped.
The hard part of AI automation work has never been the AI. It's knowing which twelve minutes of a business's day are worth automating.
What Businesses Actually Pay For
This is the part that determines whether the work is a job or a career.
Businesses do not pay for automations. They pay for outcomes they can name:
Notice that none of those are "we deployed an AI agent." The deliverable is a number that moved. Practitioners who can name the number before they build get hired repeatedly; practitioners who demo tools do not.
What It Pays
Roughly, across the market:
The consistent pattern: pay tracks proximity to revenue. Someone who automates internal reporting is a cost-center improvement. Someone who moves contact rate from 38 percent to 68 percent is defending a revenue number, and gets paid accordingly.
What a Week Actually Looks Like
To make it concrete, here's a normal week on live client accounts. Very little of it resembles what the job title suggests.
Monday: review weekend call transcripts across accounts. Find that one agent is mishandling a specific question — customers giving a cross-street instead of an address — and rewrite that prompt segment. Check that overnight workflows fired; find one webhook that silently failed on Saturday and repair it.
Tuesday: onboarding call with a new client. Ninety minutes of process archaeology. Discover their Google Business Profile messages have been going to an inbox nobody has opened since March. That single finding is worth more than the next two weeks of build work.
Wednesday: build day. Pipeline stages, follow-up sequences, calendar routing, reminder cadences. Mostly configuration and testing, punctuated by one integration that doesn't behave the way the documentation claims.
Thursday: measurement. Push conversion events back to ad platforms for two accounts, rebuild a cost-per-booked-job report, and discover a client's best-performing channel by lead volume is their worst by revenue.
Friday: client reporting and a difficult conversation about why cost per lead went up and why that's good news.
Note the ratio. Roughly one day of building, four days of understanding businesses, connecting systems, and proving results. That ratio is the job.
How to Get Into It
The path that works, in order:
Pick one platform and go deep. A consolidated CRM and automation platform, not six tools. Build twenty real workflows in it.
Automate something in a real business. Your own, a friend's, a first client at a discount. Theoretical knowledge in this field is worth close to nothing because the difficulty is always in the specifics.
Learn to measure. Build the before-and-after report. Response time, contact rate, cost per booked job. This is what turns a build into a case study.
Add voice. AI callers are the highest-value layer right now and the one most practitioners skip because it's harder than text.
Specialize by vertical. Home services, med spa, legal, dental. The second build in a vertical takes a third of the time of the first, and you can talk to owners in their own vocabulary.
The people doing well in this field are not the ones with the deepest model knowledge. They're the ones who can walk into a plumbing company, find the four hours between a lead and a callback, and close it.
The One-Sentence Version
AI automation work is business process consulting with a technical delivery layer — and the practitioners who understand that get paid multiples of the ones who think it's about prompts.
If you're a business owner reading this to figure out whether to hire for it or buy it: [book a free strategy call](/book) and we'll show you what's actually leaking before you decide.
Free Weekly Briefing
One AI Marketing Tactic.
Every Tuesday. Free.
What's actually working across our client accounts right now — ROAS moves, follow-up sequences, creative angles. The stuff that isn't in any blog post yet.
No spam. Unsubscribe anytime. 1,200+ business owners already in.