My First AI Feature

What a fractional actually needs

After lockdown, the real gap wasn’t time tracking or invoicing. Fractional Tools lets you keep track of your time and manage your invoicing. But the reality is what a fractional really needs is the ability to prove their business value week over week.

If you’re charging $300 to $500 an hour, at some point someone’s going to ask, should this be an employee? And if you don’t want to become an employee and you want to be fractional, you need to make sure that the value prop you have — the absolute emotional connection between the value you bring to the company and the value the company perceives — is crystal clear.

Invoices don’t tell a story

I looked at my invoices that were going out to clients and I realized these don’t tell a story. There’s no narrative here.

But I often use Wispr Flow or Handy to dump fairly descriptive outlines of what I did in a given two or three hour period. These two AI tools (paid versus free) are powerful. Once I started using audio to text, the utility of my work product notes went way up.

So we decided: what if we could turn an invoice into a statement of business value? Not a task list for the CEO or CFO — a short narrative aligned to their business value, something that could flow into a board slide to justify the fractional spend.

You don’t want the CEO to be saying, I think he brings value, I think she brings value. You want them to have an emotional connection to you because a fractional is not an employee.

Talk to twenty fractionals first

But we didn’t do it by just jumping into the feature. I went out and talked to 20 different fractionals and asked specific questions around how they defend their value prop. And not surprisingly, not a lot of people had a strong description for it.

I took three months to figure out what I should build. If I overbuild a feature, assuming that I understand what a customer wants, then I’ve committed to a pathway without getting validation.

A founder needs to be critical of their own ability to fit a solution to a customer. If the customer doesn’t say it, it doesn’t matter — it doesn’t exist. Believing something about your customers is not the same thing as having real feedback.

The smallest slice

Once the job was clear, the next problem was shipping AI without owning the blast radius.

There are a lot of pitfalls. Do I give a feature out that can potentially get prompt attacked? How do I make sure I know what goes out the front door? If I can’t confine the conditions, then I end up with an unknown. And I can essentially add a cost that builds a token spend that I can’t justify.

So I reversed the equation and decided to build the exact smallest amount possible.

First, the customer has to bring their own API key. That means that if you’re not savvy enough to figure out how to get an API key out of Claude or out of OpenAI, you won’t use this feature. I didn’t think that was a bad idea.

Second, I used my existing encryption to lock down those keys so they don’t leak — so I didn’t inadvertently create a problem.

And then I gave a very simple way to provide a description of the value that the client has. And then we built a simple feature to let you say, take this invoice and create an AI narrative. It takes that initial seed context prompt and takes whatever tone you want and runs a very, very simple OpenAI completion that returns back a narrative that can fit in an email.

The goal isn’t a report. One of the first things we heard from most fractionals is getting a CEO to read a report is kind of hard. Getting them to read something they might turn around and use immediately for the board of directors — way easier.

Something that was so light and tight and small that if we had to delete it a week later, we wouldn’t care. By keeping it light, tight and small, we also limited the blast radius if the AI feature impacted negatively. It could be slid out of our feature set and UI without making customers feel like we were bait-and-switching them on a feature.

Avoid the feature hole

When I use other people’s solutions, AI is peppered everywhere. You can see the icon. But I often find that I don’t understand why AI was the tool that solved this. Most problems can be solved without a prompt.

I am not an AI luddite, but I question the need to put AI everywhere.

Every founder is going to have a different experience with this. But there are too many founders who built themselves into a feature hole where they have tons of features but no clients. That feels like a trap that multiple founders stumbled their way into — and that I’ve been trying to break.

If I had to explain this to a co-founder, I would describe this as the most brutally resourced, restricted, thin-slice experiment I can run — where I deploy the fewest number of hours, the least amount of risk, and the fewest assumptions about a client possible.

One feature. Brutally smallest budget.

Do you want that same Darwinian filter on the next feature you ship — sit down, walk one scenario, and bring it to market with the brutally smallest budget possible? That’s the filter I’m using on Fractional Tools — discover the job, then ship the smallest slice.

See Fractional Tools

Related: Fractional Tools tech stack