Virelli

As a car enthusiast, I noticed that a huge number of online auto listings used photos that ranged anywhere from ordinary to awful. Images were often dark, overexposed, cluttered, or even featured dirty cars. And for many dealers, these images serve as their primary feature photos for all social media advertising and their website.

Research shows that polished online photos gain more attention, garner greater click-through rates, and ultimately drive more sales. And yet, dealers are not equipped to retain professional photographers to adequately position, light, and photograph each vehicle within their large, fast-turning inventory.

So, I came up with a solution: an advanced, AI-driven photo editing process that requires nothing more than an online upload of each ordinary feature photo, which then turns those photos into scroll-stopping, showroom-quality marketing visuals.

I didn't just build it, I took it to market. First, I manually created a highly customized cold-outreach email campaign beta run with 60 select dealers. This provided valuable learning and revealed areas of interest as well as client objections. I then hired a 3^rd-party email list supplier to scale the campaign to 3,000 dealerships.

Following poor CRM results from that campaign that seemed strange (the supplier claimed 32% "open" rates, yet other metrics put that in doubt), I undertook my own data analytics effort and diagnosed why it failed. (Bounces, bot farm and spam rates were sky high; the true "open" rate was closer to 2%.) So, I rebuilt the entire sending stack from scratch. Among other corrective initiatives: I created a fully authenticated sending domain and target audience list with verified emails; I configured sender personas on Instantly and executed a proper multi-week warmup instead of just blasting cold; I set outreach rules from first-hand testing: link-free and image-free first-touch, one personalized line per dealer, quality of fit over list size. I segmented the contact lists to strip non-genuine targets (parts stores, lube chains, accessories suppliers) before a single send, and handled positioning against the real incumbents - the processing platforms, not photographers.

Architectural build: AI still has its limits - this is something us creative techs, and everyone else, keep relearning every day. While AI can do a lot, my earliest iterations reminded me that hallucinations are still present in every AI design-and-build project. No client-facing solution should ever be built or deployed without adequate "human-in-the-loop" quality controls built in.

Penetrating existing workflows: We humans are creatures of habit. And our existing business workflows are just that: habits; habits that are hard to break. For an AI solution to truly be workflow-enhancing it must reflect a deep understanding of that workflow, and be designed to clearly show an efficiency AND output quality improvement. Because of the existing end-to-end photo/VIN services companies, Virelli's offering has struggled to break into a market saturated with existing infrastructure.

Go-to-market: Third-party data metrics should not be taken on face value. Had I relied on such metrics here, without doing my own data analytics, I would have hastily concluded that the Virelli business proposition was flawed when, in fact, the email execution was flawed. Conducting my own analytics, I discovered that the third-party email vendor was providing misleading metrics. Among the red flags: over 90% of so-claimed "opens" tracked through Chrome; most of the traffic was kicking over to known datacenter hubs. It turns out the true human open rate was only around 2%. The lesson here: be vigilant in selecting suppliers, and verify everything.