We spent months building AI marketing tools. Then realised the tool was the problem.
When we started Svay, the opportunity seemed pretty obvious. AI was making it dramatically cheaper and faster to create good marketing, so we thought the answer was to build better tools for doing it.
Then we spoke to 50+ businesses.
And the thing we kept hearing was that they already had tools for almost everything.
Canva for design, ChatGPT for writing, etc...
But someone still had to sit in the middle of all of them, decide what should actually be made, move the work around, check the quality, publish it and then figure out what to do next.
That changed how we thought about the product.
We stopped thinking about Svay as another tool a marketer would use and started trying to build Svay like the marketer itself.
Around 10 companies are now interested in trying it. The product still isn't really self-serve, so right now we're using it internally and with a few early pilots.
One thing we got pretty wrong was the quality bar.
We assumed that if a business was barely posting anything, getting them from zero to consistently decent content would already be a huge improvement.
Turns out, even businesses doing almost nothing have a very high bar for what goes out under their name.
They might not have the time to make the work themselves, but they know immediately when something feels average. In hindsight this is pretty obvious, but it changed the way we're building a lot of the product.
If Svay is going to behave like an employee, it needs judgement, not only speed.
For video, it can now go from research → idea → script → video → review → publish. It can also join meetings and take context from places like Slack and WhatsApp. The biggest time saving we've seen so far was taking something that used to take around 10 hours down to about 5 minutes.
Another interesting thing has been the reaction when we show businesses what it can already do.
A few times we've explained features that literally already exist and people have responded like what we're describing sounds unrealistic.
Basically, they didn't realise a system could already work this way.
But I think the harder problem is becoming more obvious as we build more of it.
Svay currently has around 20 different touchpoints where it can learn something about a company. Meetings, messages, research, previous work, feedback, performance and so on.
Making each of those things work is one problem. Making all of them feel like the same brain is much harder.
If Svay researches a competitor at 9am and finds something important, then joins a meeting at 3pm, it should already know what it found, understand why it matters and have something useful to say about it.
That, I think, is where the difference between an AI tool and an AI employee starts becoming real.
The bet we're making is that businesses eventually want one system they can trust to take responsibility for marketing, rather than another set of tools someone still has to operate.
The thing I'm trying to figure out right now: when you hear "AI marketing employee", what would it have to do before you'd trust it with your marketing without checking everything it creates?
If anyone wants to see what we're building: svay.ai