The short answer
The first thing to automate isn't a workflow at all. It's getting your company's data out of six apps and into one place an AI can actually read. Do that, and the automations almost pick themselves, because the AI can see where the hours go. Skip it, and you're bolting robot arms onto a filing cabinet nobody can open. After the data step, the single most valuable automation we've found is turning recorded sales calls into finished proposals: minutes instead of days, with a human approving every one before it goes out.
Everyone asks for the automation. Almost nobody is ready for it.
The typical call goes like this. An owner has seen what AI can do and wants the version of it that answers their leads and writes their follow-ups. Fair. That exists and it works.
Then I ask where the customer history lives, and the answer is: some in the CRM, some in email, some in spreadsheets, some in the project software, and a chunk of it in one employee's head. That company isn't ready to automate anything yet, and if they pay somebody to try, they'll get something fragile sitting on top of a mess.
So this article covers what we actually do, in the order we actually do it, with the numbers from our own company, including the part where things break. Because they do.
Step one: gather the data. That's the whole step.
The first project for most companies is figuring out what data they have, where it all is, and consolidating it into one place an AI can access. That's it. No automation yet.
I know that's a boring answer next to "the AI writes your proposals." But here's what it unlocks. Once the data is in one place, you can ask questions about your business that were never answerable before. Which jobs actually made money. Which lead source produces the customers who pay on time. What your best estimator does differently. That information was always in your company, scattered across apps that don't talk to each other. Nobody could read it because it wasn't anywhere.
And the second thing it unlocks is the answer to "what should we automate?" You stop guessing. The AI can look at the real data and show you where the repetitive hours actually go, which is frequently not where the owner thought they went.
At Adapt we run this on ourselves with a system I built called the Eye. It's a live feed of everything happening across all the apps in the company, organized by team member, so I can see what's going on anywhere in the business in real time. When a lead comes in, when an appointment gets scheduled, when it's time to talk to somebody, it's all just visible. Companies our size never had that before. There's no magic in it, just plumbing done right. The data got gathered, so the data can be seen.
The most valuable single automation we've found: call to proposal
Once the foundation exists, this is the workflow I'd point most companies at first, because it costs almost no extra effort and the payoff is absurd.
I record my sales calls. Before the call, AI has already done the reconnaissance on who I'm talking to. During the call I do my actual job: ask the right questions, cover the bases, present the information a proposal would need. That's the only discipline required, and it's a discipline good salespeople already have.
When the call ends, AI takes the transcript and turns everything I said into a proposal. About five minutes later I'm looking at a document that looks better than proposals that used to take three days to build. I review it, fix what needs fixing, and send it. The customer gets a detailed, professional proposal while the conversation is still warm in their head, not two weeks later after it crawled out of somebody's backlog.
Two kinds of companies win here. Companies that already do proposals compress days into minutes. And companies that never did proposals, because who has the time, suddenly get to be the outfit that sends a polished document the same afternoon. In trades and services where the competition follows up with a text message and a number, that's not a small edge.
The same shape works past sales. A friend of mine in sales was doing about six calls a day and spending 45 minutes to an hour summarizing them and tracking follow-ups. An AI pipeline took that to about five minutes of review. Call it five hours a week handed back to one person, from one workflow.
The pattern behind every automation that works: AI does the legwork, a human signs
What stayed human in that proposal story was the judgment. The AI drafted. I approved.
We built it that way on purpose. Every pipeline we build at Adapt works the same way: the AI does the legwork, organizes the output, and queues it up for an expert human who approves, denies, or edits before anything real happens. Websites, competitor research, Google Ads management, SEO work, all of it runs through that pattern in our shop. It's the same promise on our workflow automation service: nothing goes out the door without a human signature.
AI gets you most of the way fast, and the last stretch is where expertise and responsibility live. A proposal with a wrong number in it isn't 90% good. It's a problem with your name on it. As far as I'm concerned, that human review at the end is the product.
Run this pattern across a company and the arithmetic gets loud. We do about ten times the work with the same people we had before, and profitability is up around 22%. Nobody works harder than before. The hours that used to go into legwork now go into the judgment calls only people can make.
Things break. We plan on it.
Now the part every vendor leaves out: failures happen all the time, in our company and in every company doing this honestly. And that's actually fine, because of how building software works now.
It used to be that custom software meant months or years of planning, building, and debugging before anyone used it, and it had better be right because you only got one shot. That world is gone. Now you build version one fast, use it for real, find where it breaks and where it falls short, and build version two. By version three or four the thing works well and rarely fails. Then six to twelve months later you build the next version anyway, because your company grew and your needs changed.
That's why "set it and forget it" automation is a fantasy, and why anyone selling you a one-time AI solution that's good forever is selling you version one and a handshake. It's also why what we sell has turned into a long-run seat, a fractional AI officer who keeps iterating the systems with you as the business changes.
Where to start, in order
If you're a company between $1M and $50M and this all sounds right but you don't know what's first, the sequence is:
- Inventory the data. What exists, where it lives, who can get at it. This is a conversation, not a purchase.
- Consolidate it. One place, organized, AI-readable, with permissions so people only see what they should.
- Ask it questions for a month. You'll find automation candidates you'd never have guessed, backed by your own numbers.
- Automate the loudest one. For a lot of companies that's the call-to-proposal pipeline. For yours it might be reporting, or scheduling. The data will tell you.
- Keep a human at the end of every pipeline that touches money, customers, or commitments.
We do the whole sequence, and step one starts as a conversation. But even if you never call us, do steps one and two. Everything else stands on those two.
Questions owners actually ask
What should a business automate with AI first? Nothing, until your data is consolidated somewhere an AI can read it. That's the real first project. After that, pick the workflow where the most repetitive hours go. The data will show you, and for many sales-driven companies it's turning recorded calls into proposals and follow-ups.
How much time does AI workflow automation actually save? The honest answer is "measure one workflow." Our reference points: a salesperson's 45 minutes to an hour of daily call admin dropped to about five, and our own proposals went from days to minutes. Across our whole company it all adds up to about ten times the output with the same headcount.
Do AI automations replace employees? In our experience they replace the legwork inside jobs, not the jobs. The people stay. They just stop spending their days summarizing and formatting and spend them deciding and talking to customers. That ten-times number above happened with the same team.
What happens when an automation makes a mistake? A human catches it at the review step, which is why every pipeline we build ends in approve, deny, or edit. And the mistake feeds version two. Iteration is the maintenance model: nothing is built once and left alone, because businesses don't hold still.
Do I need new software for all this? Usually less than you'd think. The work is mostly connecting and organizing what you already run, then building the pipelines on top. That's also why quotes vary so much company to company. The automation itself is the cheap part. Getting a company ready for it is where the price swings.