Viveka I have opened plenty of AI tools because somebody showed me a clever demo and I thought, oh, I could use that. Sometimes I could. Sometimes I spent two hours on setup, watched three tutorials, made an account I would later forget existed (which is its own small tragedy), and still had no idea where the thing belonged in my business, and I am 94% sure I am not the only woman whose credit card statement has a small monthly mystery charge on it that she cannot identify.
The women I work with are not refusing AI. Most of them are already in ChatGPT or Claude or Perplexity or Gemini or Magai, poking around, getting a decent result here and there. The stall happens somewhere else — their curiosity runs straight into a business that is already full of client commitments, follow-up, content, ideas, and the operational work nobody sees — and the next tool does not fix that, it adds a login.
So this is the piece I wish somebody had handed me three years ago, before the mystery charge.
Why does adding an AI tool usually make the overwhelm worse?
Adding a tool makes the overwhelm worse because a tool is an answer, and most of us go shopping before we have written down the question. Technology arrives first, the business job stays fuzzy, and the setup time gets spent on something you were never going to keep.
What my client needs is not another list called “100 AI Tools You Must Try.” (Please. I beg you.) She needs to know which part of her business deserves the attention first, what a good result would even look like when she sees it, and how much of the work has to stay hers no matter how good the software gets. Those three decisions take about twenty minutes and they save you the two hours, the tutorials, and the forgotten account.
Over the next few weeks I am publishing fifty practical AI use cases for an expertise business, organized around work you already know you need to do better, and this piece is the one that makes the other fifty usable. Read this one first, then take what fits.
What kinds of work is AI useful for in an expertise business?
Five kinds: marketing and visibility, building assets out of your method, running the business, client delivery and the intellectual property underneath it, and visuals and media. Those five cover nearly everything an experienced founder does in a week, which is exactly why I organized the series that way instead of by tool.
Marketing and visibility
AI can find the ideas buried in work you already published, research what your audience is asking, shape a first draft, help you protect your voice instead of sanding it off, plan a newsletter, follow up with people who raised a hand, and tell you which of it produced an actual response. You probably do not want to become a full-time content creator, and you do not have to — what you need is a visibility system that helps the right people understand what you know and why you are worth trusting.
Building assets out of your method
Your expertise can become an assessment, a guided planner, a client portal, a calculator, a workshop companion, a resource library, or a small app that does one useful thing. Some of those bring in leads, some make delivery better, some become part of a paid offer — and all of them start with your method and the result you want to create, because software has never once rescued an idea that was still vague. I wrote about that gap in the difference between using AI and building with it, and about what it looks like when a founder realizes her IP is already an app.
Running the business
AI can take on recurring prep, summaries, reminders, reporting, onboarding, organization, and handoffs. This category demos terribly and pays off enormously on a Wednesday afternoon when the client work is finished and the administrative work is still sitting there looking at you. (Rude.)
Client delivery and the IP underneath it
Your questions, your frameworks, your exercises, the way you explain a hard thing, the order you make decisions in — that is what clients are paying for, and AI can organize it, prepare customized resources from it, support clients between sessions with it, and turn teaching you have repeated two hundred times into an asset you own. The software should make your expertise easier to use without pretending to replace the judgment underneath it.
Visuals, video, and media
AI can make diagrams, article covers, workshop graphics, presentation visuals, video plans, captions, and clips. It can also produce synthetic sameness at breathtaking speed, so if your brand guidance is vague you will get handed the same grey-blond woman in beige that AI seems determined to assign to every woman over fifty. We were punk rockers, dammit.
How do you find the pressure point to start with?
Look at the last two weeks of your work and find where the business leaned too hard on your memory, your time, or your physical presence. That is the pressure point, and it is a much better starting question than “where could I use AI,” which is broad enough to eat an entire afternoon and leave you with nothing.
The candidates are usually work that repeats, information you keep hunting for, material you rebuild for every client, ideas that never get published, and follow-up that only happens if you personally remember at the right moment. Write the task down in plain English, the way you would say it to a friend: after every discovery call I need to send a personal follow-up and remember the next action. That sentence is the whole brief. (If you cannot write it in one sentence, the task is not one task yet, and that is useful information too.)
How do you describe the result before AI touches it?
Tell yourself what “done well” looks like, specifically, before the AI sees the task. For the discovery-call follow-up, done well might be a short email that refers to her real situation, includes the resource you promised her, and proposes the right next step — sounding like you, containing nothing the two of you did not discuss.
This is the step everybody skips and it is the one that does the most work. “Write a follow-up email” is an invitation for the model to fill in the gaps with whatever it has, which is how you end up with a warm, competent, slightly creamy email that could have been written for anyone by anyone (I have sent that email, and I could feel it going out). A described result tells it what to produce, and it gives you something concrete to hold the draft against when you review it.
What information does AI need, and what should you never hand over?
It needs the smallest safe pile of material that can do the job, and nothing beyond that. The follow-up email might need your call notes, the offer details, two emails you have written before so it can hear you, a line of context about the relationship, and your rules about what you never promise — a content workflow might need your audience file, your voice guidance, a few approved articles, and your current point of view on the topic.
Do not upload an entire client archive because one task needs three paragraphs of notes. I say that gently, as someone who has done the enthusiastic-oversharing version and then had to think hard about what exactly was now sitting in a vendor's system. If you want the version of this that is specifically about keeping your own voice intact while you do it, I wrote that one here.
Where does the human boundary go?
Decide up front what AI may prepare, what it may suggest, and what only you may approve. The more sensitive the information or the more consequential the action, the tighter you draw that line.
In marketing, it might draft the language while every claim and every client story goes through you. For client delivery, it can organize what the client submitted, and the recommendation still comes from you. On the operations side it gets to prepare a read-only report long before it is allowed to change a record or send anything to a human being. Write these down once, per category, and you stop relitigating the question every time you sit down. (Mine live in a file called RULES, in caps, because subtlety was not working.)
What is the smallest version you can test today?
Run it by hand, inside the AI you are already paying for, before you build anything. Paste in the material, give it the job and the boundary, look hard at what comes back, then fix your instructions and go again — and if you expect to repeat the task, save the prompt or drop the instructions into a Project so it is waiting for you next time.
Do not build an agent for a process you have never once run by hand. Do not move your whole business onto a new platform to avoid copying one set of notes into a chat window. And do not buy the annual plan because the monthly plan is “less of a bargain” — ask me how I know.
How do you know when the workflow is dependable?
Run it several times with different inputs and watch specifically for where it fails, because one good result can just be luck. Five clean runs on five different real inputs gives you something you can rely on.
Watch for the three failures that matter: it invents details when your notes are thin, it smooths your voice until the thing could have come from anyone, and it misses the same category of information every single time. Each of those has a fix in the instructions, and a surprise you do not like is a reason to tighten the brief or take a permission away. Once it is producing dependable drafts, then you can ask whether automating it would give back enough time to be worth the setup — which is a real question with a real answer, and often the answer is no.
What should you measure?
Measure the thing the pressure point was costing you. If the problem was warm leads going cold, count follow-ups that went out; if it was your method evaporating after every session, count the client resources that exist now and did not before; if it was time, count the hours.
The number of AI tools involved is not a measure of anything, and neither is how impressive the workflow sounds when you describe it to another founder at a conference. (I have been the person describing the impressive workflow. It was not making me any money.)
What does this look like all the way through?
Here is the discovery-call follow-up, start to finish, because a worked example is worth more than another framework. The pressure point: I get off a good call, I mean to follow up that afternoon, the afternoon fills with client work, and three days later I send something generic or I send nothing at all — and a warm lead cools for no reason except that my memory is a terrible system.
The result I want: a short email that names her actual situation in her words, hands her the specific resource I promised on the call, proposes one clear next step, and sounds like a person she just talked to for forty minutes. The material it needs: my notes from that call, the one-paragraph description of whichever offer we discussed, three follow-up emails I wrote myself so it can hear my rhythm, and my standing rule that I never promise a result or a timeline I have not seen someone hit. The boundary: it drafts, I read every line, I send. It never sends.
The test: I ran it on five real calls, with notes of varying quality, and the first two drafts were too smooth — competent, warm, could have come from any coach with a decent email template. So I added two of my scrappier emails to the material and told it to keep my actual sentence length, which is long and a little tangenty, and the third one came back sounding like me on a good day. It also invented a resource I do not offer, once, when my notes were thin, which is exactly why the boundary says I read every line.
Total setup: about forty minutes, most of it deciding things I had never written down. It now takes me four minutes to send a follow-up I would have taken twenty minutes to write, on the days I wrote it at all. That last clause is the real win.
When is AI the wrong answer?
When the work is rare, when the problem is a decision you have been avoiding, or when the real bottleneck is that you have not decided what you sell. I say this as someone who uses AI every single day and still tells clients no on this regularly.
A task you do twice a year does not deserve a workflow — do it twice a year and go outside. And if your positioning is still fuzzy, an AI content system will help you produce fuzzy content faster and in greater volume, which is a worse problem than the one you started with, and I have watched smart women spend a month building the machine so they would not have to make the decision underneath it. Fix the decision. The tooling gets easy after that.
What do you do with the other forty-nine?
Skim each edition for the ideas that connect to a pressure point you have right now, and ignore the rest with a clear conscience. A founder losing warm leads because she never follows up has a completely different first use case from a founder whose calendar is full but whose method disappears the second a session ends, and a speaker sitting on nine years of recorded teaching has a different opportunity from a new consultant still deciding what her core offer does.
The twenty-minute version
Find the pressure point from the last two weeks. Write the task in plain English. Describe what “done well” looks like. List the smallest pile of material it needs. Decide what stays yours. Run it by hand five times in the AI you already have. Then, and only then, ask whether it is worth automating.
This gets easier, and faster than you would think. The first one takes longer than just doing the task yourself, because you are making decisions you used to keep in your head and never had to say out loud — and the payoff shows up the next time that work comes around, already decided. If you want help picking the first one for your business specifically, that is most of what I do with clients.
Pick the one that has been annoying you the longest. Start there.