How we make Kaz better

Run experiments and get way more done with Kaz!

Kaz has been our flagship product for a long time because he can make running an ad setup far easier (and quicker). Over the past few weeks, we've invested lots of effort into further refining the Kaz workbench and making it easier for anybody to interact with.

Running experiments

We introduced new functionality, allowing Kaz to query your analytics and run experiments.

Experiments are an impactful feature we shipped earlier this year, which helps you with testing out new placements, rules, or really any other part of your configuration by serving both the prior version and the updated version (each to a comparable cohort of visitors). They let you set up A/B testing and compare the data with no coding required.

Running experiments with Kaz makes this even easier, since Kaz can guide you through coming up with ideas, creating the placements to try, or even looking into your analytics reports and existing ad setup to suggest improvements that are worth experimenting with.

It has been a priority of ours to make sure that Kaz has a strong understanding of your website, which is what makes our agentic harness so powerful. It is imperative that Kaz gets direct insight into your existing ad setup, both with us and with other monetization partners.

Alongside these changes, we launched an overhaul to the Kaz design, which brings it closer to modern expectations and increases the smoothness of working with Kaz overall, with cleaner animations, fewer edge cases, faster response times, lower network load, and really all kinds of improvements across the board. Plus, we made it clearer to see what changes Kaz tries to make, and the research he conducts along the way. We improved things like inline citations (that open directly to the report or piece of data Kaz saw), and more accurate reporting of when there were empty results or queries Kaz needed to submit again.

One-click undo

Sometimes you might not review the change Kaz has proposed fully, or change your mind later. With the added changes tab and one-click undo control, that becomes very easy to fix – you can instantly preview recent changes, when they happened, and revert them, bringing them back to a pending state (or even sending them to Kaz directly to make the edits you're looking for).

Checking his work

We introduced additional tools that help Kaz be more careful and make sure that proposals will match your site and work correctly from the get-go.

We already had the rule simulation area (for our human users), but it might be a little slow to configure. You have to think about the various characteristics of a single visitor before you can see which rule will actually reach them.

Kaz is far more efficient at writing configurations than a human is, so simulating a proposed rule against a real visitor or a visitor archetype makes it very quick and cheap to verify changes that you or Kaz make.

Another useful piece for verification is confirming whether a newly created rule or placement is actually touching users. For this, we gave Kaz another tool: he checks the recent history for that placement or rule, then confirms to you whether it matches a sensible number of visitors and has been successfully deployed.

We also upgraded Kaz's memory system, so when you reject a proposal Kaz makes, or tell him to keep track of something you like or dislike, Kaz will hold on to this idea for future conversations and make sure to take it into account.

How you benefit

Ad ops is a pile of small, high-stakes tasks, but they're difficult to automate in the traditional computing sense. Machine learning can help, but you generally need a model that's based on language to understand human intent and why a website is structured the way it is. An assistant that can help you with these topics makes the difference between a morning spent in the dashboard and a sentence you type out, or even dictate, in chat.

You do not have to be an ad ops expert to run a serious setup that is well optimized, and you don't need to engage ad ops experts for every task either. We still have many humans available to help support you, but Kaz allows us to provide a high-quality service to a larger number of publishers, including smaller ones, for whom it would be less cost-effective to provide 24/7 human support by experts.

Plus, this allows you to achieve speed without the usual cost of speed. Things get faster because the blast radius is small, and because it is easy for Kaz to collect high-quality information. You still get to review things, and we implement strong validation on the backend too.

Another advantage is that we focus on making Kaz work against the stack you already have. Regardless of whether you've implemented tags on your own from our platform or also work with other monetization platforms, Kaz is capable of understanding where ads are on the page. He will avoid unsuitable placements, ads touching each other, and mangling placements in unsuitable locations.

Our business model is always about providing a good-quality service where you stay because you want to. We don't use lock-in contracts or any kind of exclusivity.

If you're doing this by hand, the next experiment is the one you have been postponing because setup is annoying. That is the kind of thing to try with Kaz. Whether you are already our customer or just interested in learning more, feel free to speak to our team through the contact page about our offering and how Kaz can help with improving your website's performance.