Dogfooding Dynamic Ad: what running ourselves on our product taught us

Dogfooding is a term from software engineering: eating your own dog food, running your product on your own problems before asking clients to run it on theirs. It is supposed to ensure that you find the hard edges yourself, before your clients do.

We run Dynamic Ad's entire client acquisition through Dynamic Agent. This post is an honest account of what we learned when we became our own client: what the Intelligence phase found, where the Workshop phase surprised us, and what we now understand about our own product that we did not understand when we built it.

This is the first entry in what will be an ongoing public log. Small company, early stage, no pretense that we have everything figured out.

What the Intelligence phase found

When we ran the Intelligence phase on Dynamic Ad, we gave it the same inputs we would give for a client: our website, our offer (the free growth audit), our competitive context (other AI marketing agencies, traditional agencies, AI point tools), and our stated target audience (SMB and mid-market performance brands, $5K–$100K/month in ad spend).

Three things came back that we did not expect.

The messaging was fighting itself. Our website simultaneously positioned Dynamic Ad as "better than a traditional agency" and "different from AI point tools." Those are two different arguments for two different buyer objections. A buyer who is currently working with a traditional agency needs a different argument than a buyer who has tried AdCreative.ai and found it insufficient. The Intelligence phase flagged that our messaging was too broad. We were trying to win both arguments at once and not landing either cleanly.

This was not a new insight once we saw it. It was a thing we had been aware of and had not resolved. Seeing it surfaced by the system we use for clients made it harder to ignore.

Our target buyer definition was too vague. "SMB and mid-market performance brands spending $5K–$100K/month" is not an ICP. It is a spend range. The Intelligence phase mapped the competitive landscape and found that the accounts most likely to see value from a fully AI-operated model share more specific characteristics: they have tried running ads in-house, found it requires more specialization than they can hire for, and are spending enough that the cost of a bad month is significant. The spend range is a filter; the profile is the target.

Our audit offer had a positioning problem. We were presenting the free growth audit as a lead generation mechanism. "Get an audit" as a conversion event. The Intelligence phase analysis suggested it is more accurately a proof-of-work: the audit is a demonstration of the Intelligence phase's capability, and framing it as such ("watch our Intelligence phase run on your business") is a stronger argument than "get a free audit." The difference is subtle but real.

We changed the framing on the website as a result.

What the Workshop phase taught us about our own creative instincts

The Workshop phase produces creative variants based on the angle library built from the Intelligence analysis. For our own acquisition, it produced a first batch of angles and copy that was partly different from what we would have written intuitively.

Our intuitive angle: position around speed. "We test ads faster than any human team." Clear, defensible, relevant.

The Workshop angle, built from competitive analysis: position around certainty. "You'll know in 72 hours if your current campaigns are leaving money on the table." This speaks to a specific buyer fear (the worry that the current agency is underperforming and nobody is telling them) rather than to a product feature.

We tested both. The certainty-framed angle significantly outperformed the speed angle in early testing.

Our intuition was not wrong. Speed is a real differentiator. But the Workshop analysis found a buyer emotional context we had underweighted. The system was not smarter than us. It was less biased by our own enthusiasm about the product.

This is the specific value of dogfooding: your own creative instincts are shaped by your proximity to the product. You think about what is impressive about what you built. The Intelligence phase thinks about what the buyer fears.

What we are still figuring out

Honesty requires a list of what is not resolved.

We are still early in building the angle library for our own acquisition. Four weeks of testing is not enough to make confident claims about which positioning wins. We have early signals, not conclusions.

The audience segmentation is not complete. We know the profile of the buyer most likely to see value; we have not yet translated that into precise Meta and Google targeting parameters that produce a cost-effective reach into that audience. That work is ongoing.

The audit delivery workflow (the 72-hour human-assembled audit) is the current bottleneck. We deliver on the 72-hour commitment, but it requires significant operator time. The automated version is on the roadmap, but not yet built. We are transparent about this: the intelligence pipeline that will eventually automate audit production requires more engineering work than we have completed.

We are running this log in public because the discipline forces rigor. If we publish that we are testing a specific angle, we have to actually measure it and report back. The log is accountability infrastructure, not just content.

Why transparency is a pricing signal

Publishing what we find (including what is not working) communicates something that a polished case study cannot: that we have direct experience with the gap between what the product promises and what it delivers at this stage of development.

Most agencies publish only the wins. "We grew ROAS by 40% for Client X." That statement is unverifiable, often cherry-picked, and usually missing the context of what happened before and after the number was captured. It produces a credibility signal roughly equivalent to five-star reviews on a vendor's own website.

Publishing the Intelligence phase findings, the Workshop surprises, and the current open questions communicates something different: that we are running the product in the actual conditions a client would experience, and we are willing to describe it honestly.

For a buyer evaluating whether to trust us with their ad budget, the honest account is more informative than the polished one. It demonstrates that we have judgment (we can distinguish what is working from what is not), and that we value the relationship enough not to mislead them.

That is the thing dogfooding produces that no other exercise does: an honest reckoning with your own product, in public, with your name on it.

Amit Harari is the founder of Dynamic Ad. He writes about AI-driven performance marketing operations at dynamicad.ai/insights. This is the first entry in an ongoing public log of Dynamic Ad's own acquisition operations.