The Path to Validity Engage: Three Ways to Think About AI, Before You Think About Tools

Every AI initiative I’ve watched fail didn’t fall flat because the technology broke. It failed because the organization around it wasn’t ready. The data was dirty, the teams weren’t aligned, or a pilot that looked great on a slide couldn’t survive at scale. 

I’ve spent the last couple of years leading Validity’s own AI initiative and building Validity Engage alongside some of the world’s biggest brands.  

Here’s the thinking that got us there without headaches. 

Three ways to think about AI, before you think about tools 

When I talk to enterprise leaders trying to figure out where to start with implementing AI into their workflows, the first mistake I see is jumping straight to adding new tools. I know tools are exciting, but for the best results you have to start with finding the right approach. One customer of ours, a large financial institution, had identified 180 possible AI use cases in marketing alone, and sorting through that kind of volume only works if you’ve got a strategic framework for AI use first.  

I generally see three implementation approach buckets. 

The first bucket is leveraging foundational LLMs sitting right on the desktop—the ChatGPT, Claude, or Copilot most organizations are already rolling out broadly. They’re genuinely useful for smaller, single-step tasks, but they come with real governance questions: shadow usage, data leaking out of the organization, and a ceiling on how much complexity they can handle, even as MCP integrations start to close that gap. 

The other two buckets hold agentic approaches. You can build your own agent internally, which still takes real engineering expertise. It also raises its own questions about how that agent gets access to your data securely and requires ongoing care. It’s not something you deploy and walk away from.  

The other option is looking for agentic capability embedded directly inside the SaaS tools you already run. That’s exactly where Validity Engage lives: an agent built into our platform that automates tasks and surfaces intelligence without asking your team to stand up and maintain something new from scratch. 

The people problem shows up before the first line of code 

Long before we wrote any code for Engage, we had to rethink how our own organization worked—not just the product, but the process behind it, from design through product management.  

Our first move was to spin up a dedicated AI team, incubated slightly apart from the rest of the org, so they were free to figure out new ways of working without the weight of old workflows. Then they can bring those lessons back to everyone else. We’re still largely in that structure today. 

One shift stands out: our design process used to mean wireframes, then high-fidelity mocks, then customer review, one round at a time.  

Now we can generate ten or fifteen high-fidelity design options at once and get customer feedback fast. That only works, though, if people bring real creativity to how they deploy the technology. 

The bottleneck moves, it doesn’t disappear 

Here’s the catch we ran into almost immediately when building our first AI solution: making one part of a process fast doesn’t make the whole process fast if a human still has to review everything coming out the other end.  

In terms of marketing: just because you can agentially create tons of novel marketing materials doesn’t mean you should. We saw the identical pattern with code—AI can generate it far faster than any team can review it. Our answer was to get creative about using an LLM as a judge, filtering down what genuinely needs human eyes before it gets there.  

It’s an easy trap: assuming that because you’ve accelerated one stage, you’ve accelerated the outcome. 

Why your data has to be more pristine than it’s ever been 

We know from our own research that close to half of marketers don’t believe their CRM data is ready for AI.  

That number doesn’t worry me—it excites me, because it’s a solvable problem, and it’s central to what Engage was built to solve.  

A human working with messy CRM data can look at a poorly configured parent-child record and think, “oh, that’s actually a subsidiary of that company,” filling in the gap from memory.  

An agent doesn’t have that context. If you’re going to let agents act on your behalf, triggered by your own data, that data needs to be cleaner than it’s ever been—structured and free from bias, at minimum.  

Sometimes that means simplifying the dataset itself so an agent can query it reliably, on top of just cleaning it up. 

Business needs first, technology second 

Whenever we’re deciding whether something is a valuable product versus a vanity project chasing a trend, we go back to a simple question: is it solving a real business need?  

Take our own pre-send email optimization work inside Engage. Before, a marketing specialist building a single campaign might spend eight to ten days gathering feedback, running QA, and manually checking screenshots across 120 different email clients.  

Engage automates that entire chain, using computer vision to catch broken rendering a person might miss on a first pass. With this level of automation on your side, it can also cut campaign production by 20-40 percent.  

That’s the model we look for: an agent combining reasoning, good data, and the right tooling to string together a sequence of steps nobody had the bandwidth to do manually before. 

Measuring success without losing sight of the customer 

When it comes to reporting the results of AI implementation to stakeholders, the first step is always establishing a baseline—cycle times for a campaign before AI touches it—and then being honest about the gains, including where a human is still in the loop and whether that review step is eating into the efficiency you thought you’d captured. But efficiency by itself isn’t the whole scorecard. A 20 or 30 percent efficiency gain means nothing if engagement metrics start sliding at the same time. Those two measures have to move together, or you’re not actually ahead. 

Where this leaves us 

Every customer we’ve built Engage alongside shares the same starting point: a mandate from leadership to adopt AI. What differs is the job AI should do, whether that’s a subprocess inside a bigger workflow or something as specific as catching broken links. I see a long runway ahead for these agents, and not just on efficiency—there’s real opportunity around legal and compliance risk, staying on brand, and avoiding the kind of subject-line misstep that’s landed real companies in real regulatory trouble. 

If there’s one thing I’d want a team starting this work to take away, it’s that AI implementation rewards whoever does the unglamorous work first: clean data, the right team structure, and clear metrics for what success actually looks like. Get those foundations right, and everything downstream gets easier. 

Want to see how Engage puts these ideas into practice? Take a look today. 

And if you want to see how this looks from the very top, this on-demand episode of our AI Executive Briefing webinar features Validity’s CEO, Mark Briggs, to talk about how executives set realistic AI expectations with their boards—and how leaders avoid the trap of overpromising and underdelivering.