AI consulting
You tried AI, and it just made more work.
The tool was never the problem. Nobody defined what “good” looked like, so it guessed, and handed you all the checking. We build AI workflows with the rules and the scoring built in, so the boring 80% runs itself and your team just approves the output.
Sound familiar?
You’ve already had this letdown
“We tried AI, and I spent more time editing the output than it would’ve taken to write it myself.”
“The content is generic and repetitive, the tone is always off, and it makes up facts and ‘customer wins’ that never happened.”
“Nobody trusts the output. Every piece needs heavy fact-checking, so the bottleneck just moved from writing to reviewing.”
“I had to do more editing than actually using what it generated.”
So the team quietly stopped using it. You’re not alone, and the tool isn’t the reason.
Diagnosis
The tool was never the problem
AI did exactly what any tool does with no standard: it guessed. Nobody had defined what “good” or “relevant” meant, no rubric, no scoring, nothing to filter against. So it produced plausible-looking output and handed you the entire job of judging it, which is the part that eats the time.
The fix
AI does the boring 80%. You check the rest.
We build workflows that use AI for the filtering and the first draft, with a person making the call. What makes one stick instead of getting abandoned like the last tool:
- An explicit scoring rubric, not a clever prompt. The workflow knows what “good” and “relevant” mean, so it filters before it drafts.
- You review fewer things, not more. It surfaces the 20% worth your attention instead of dumping 100% back on you.
- It runs on tools your team already maintains (Feedly, Zapier, Make, your CMS), so it doesn’t need us to keep it breathing.
- A non-technical person can change it without calling us.
It will not replace your team, and it will not run unattended. That is the point, not a limitation.
Proof
We built this inside a federal health agency
A federal health agency needed the week’s AI news filtered from the hype. Traditional PR monitoring was too expensive and too vague. We built it on Feedly, Zapier, and Make, deliberately no-code, so their team maintains it without a developer. The hard part was the rubric, and iterating it cut API usage by about three quarters. The result the team can restate themselves: fewer articles to review, higher quality, and no developer needed to keep it running.
For an enterprise sales team at a B2B software company, reps couldn’t personalize outreach at quota, and the experienced ones resisted every new tool. We built drafting on their CRM data with tailored follow-ups, and it beat their existing control sequence. We didn’t measure the exact lift, so we won’t pretend to a number.
Where to start
Start where it’s eating someone’s week
Straight talk
When we’ll tell you not to use AI
This is where we part ways with every consultancy promising the opposite. We’ll tell you no when:
- You’re trying to cut headcount. That isn’t what this does, and we won’t pretend it is.
- You want fully autonomous publishing. Nothing we build ships without a human approver.
- It’s regulated content with no one to sign off. The risk isn’t worth the time it saves.
Knowing when not to reach for AI is most of the job. It’s also why what we build is still running a year later.
The alternatives
Who owns it when the model changes?
| Build it yourself | Hire an AI agency | Common People | |
|---|---|---|---|
| Who maintains it after launch | You, on top of your day job | Them, on a meter | Your team, because we build it to hand off |
| When the model changes | You debug it | Another invoice | Swappable by design; the rubric stays |
| Cost | Your time | An open-ended retainer | From $3,000, scoped |
| Time to a working version | Weeks of trial and error | Slow discovery first | A working v1 fast, then iterate |
FAQ
What you’re worried will go wrong
We build tool-agnostic where we can. The rubric and the logic outlast any single model, so swapping it out doesn’t mean starting over.
You do. It runs on your accounts and your tools, documented, no lock-in.
Only what the workflow needs, and you can revoke it after. We design for the least access that works.
Yes. We build it no-code wherever possible, so a non-technical person can adjust it.
Ready when you are
You keep the judgment. The machine does the sorting.
That’s the only version of AI worth paying for. Talk about your workflow.