I hired an AI coworker in May. Three days in, it told me my Facebook ads were reaching an audience I hadn’t chosen.
Not underperforming, not “needs optimization,” wrong audience. I ran Meta’s Advantage+ targeting on my ad account, the version where you hand Meta a seed audience and let its algorithm expand from there, and Viktor pulled the campaign data and found that only 48% of clicks were landing inside the audience I’d actually set up. The other half went wherever Meta decided to send them.
I’d been running that campaign for months, reading the weekly reports, and I never once cross-referenced click destination against my own targeting settings, because why would I. The dashboard shows spend and clicks, and hands you a CTR that looks fine on its own, but none of that tells you that half your budget went somewhere you didn’t send it.
That’s the first thing Viktor did that I couldn’t have done myself in the time I had, and the reason wasn’t that the analysis was hard. I wasn’t going to run it, because I had a business to write books for.
What Viktor actually is, for anyone who hasn’t set one of these up
Skip this section if you already know it. Viktor is an AI agent connected to my actual accounts, not a chatbot I paste numbers into, and that distinction is the whole point. It has API access to my Amazon Ads console, plus Meta, MailerLite, and my KDP dashboard, and it runs on a schedule, a Meta campaign check every morning at 10am, a full attribution report every Friday, alongside just answering when I ask it something in Slack. Setting up the Amazon side alone meant navigating three separate Amazon portals that don’t talk to each other cleanly, since Amazon splits developer access from advertising access, and support access sits behind a third wall entirely. Viktor did that once, and it stayed done.
A chatbot tells you what to check, and you have to remember to ask. An agent with account access checks it on a schedule, whether you remembered or not.
The finding that actually changed how I spend money
Every indie author I know assumes their ad spend is doing something, even when they can’t say exactly what, and I assumed the same thing. So when Viktor reconciled my Meta ad spend against Amazon Attribution data for my best month of the year, I expected to see paid ads driving a meaningful chunk of revenue, even 15%, even 10%.
$5,065 in revenue that month, and $52 of it traceable to Meta ads.
I want to be precise about what that number means, because it’s easy to round it into something dramatic and wrong, and it doesn’t mean advertising is worthless. It means my specific Facebook campaign, on my specific books, in that specific month, produced almost nothing measurable next to organic and Kindle Unlimited discovery, and I’d been treating that ad spend as central to my results for over a year when it turned out to be a rounding error.
I turned the campaign off within the week, and nothing in my sales moved.
The gap I didn’t notice for 44 days
The third thing worth mentioning isn’t a diagnosis, it’s an alert. Viktor’s weekly scorecard pulls ad spend and MailerLite subscriber counts together with Amazon Attribution into a single view, and one Friday it flagged that I’d gone 44 days without sending a newsletter. I run three separate lists under different pen names, and I was sure I’d sent something in that window. I hadn’t.
Nobody catches that by feel, because it isn’t a feeling, it’s arithmetic someone or something has to actually run.
Where I’d push back on the hype
I don’t want to write the version of this post where the AI coworker is magic and I just sit back, because that’s not what happened and it isn’t honest. Every one of these findings still required me to decide what to do with it, kill the campaign, fix the newsletter cadence, question an assumption I’d been carrying for a year. Viktor surfaces the problem, and it doesn’t run the business. Two guardrails I built in from day one, and would tell anyone setting this up to build in too: nothing customer-facing ships without me approving it first, and no number goes into a report without a source I can trace back to the raw export.
Of the roughly 91 hours Viktor’s own usage report says it saved me over 30 days, most of that came from me actively working with it in conversation, not from unattended automation running while I slept. The scheduled jobs matter, and the ad monitor and the weekly scorecard do run without me. But the bulk of the value came from directing an agent that could act, not from a system I could ignore.
If you want to try this yourself
I’m not neutral here, so I’ll say it plainly: the link below is an affiliate link. If you subscribe through it, you get $50 off your first purchase, and I receive a commission.
I wouldn’t put my name on it if the audience-leak finding and the ad reconciliation above hadn’t been real, but you should know the incentive going in.
Start with whatever account is bleeding money you haven’t looked at closely, since mine was Meta and yours might be somewhere else entirely.