Improvado’s AI Cuts CPL 35% in 2026

Using an AI agent in marketing isn’t just a theory anymore. For a lot of us, it’s a practical tool we need to make sense of data and get campaigns to perform. This breakdown of a recent project for Improvado, which is a marketing data aggregation platform, shows how applying AI in a targeted way can completely change your results. So how do these systems actually affect the numbers when you’re in the trenches of digital advertising?

Campaign Overview: Improvado’s AI-Driven Lead Generation

Our goal for Improvado was straightforward: get them high-quality marketing-qualified leads (MQLs) for their enterprise data platform. We were going after marketing VPs, directors, and data analysts at mid-market to large companies, mostly in North America. The entire campaign ran for 12 weeks, between January and April 2026, and we had a total ad budget of $75,000 to work with. The whole plan was built around using an AI agent to manage everything from bid optimization and audience building to swapping out ad creatives on the fly.

We started by setting a baseline. We got campaigns up and running on Google Ads (both Search and Display) and LinkedIn Ads, hitting keywords like “marketing data integration,” “BI for marketing,” and “marketing analytics dashboards.” Right out of the gate, our Cost Per Lead (CPL) was sitting around $85. Calculating Return on Ad Spend (ROAS) was tough at this stage because of the long sales cycle for this kind of enterprise software. We were seeing Click-Through Rates (CTR) of about 1.2% on Google Search and 0.6% on LinkedIn from roughly 2.5 million impressions across the board.

Strategy Breakdown: AI Agent at the Core

The AI agent we built for this campaign had a few specific jobs. It pulled real-time performance data straight from the Google Ads and LinkedIn Ads APIs, looking at things like impression share, conversion rates, time-on-site from Google Analytics 4, and the lead quality scores coming from Improvado’s own CRM. Then, its machine learning models got to work predicting the best bid adjustments based on what had worked before and what the competition was doing. For example, if a group of keywords on Google Search was bringing in leads that went deeper into the sales funnel, the AI would automatically bid more for those exact terms.

It also took care of dynamic audience segmentation. On LinkedIn, the agent watched engagement signals (likes, shares, comments) on all our ads and content, which helped it spot new interest groups we hadn’t thought of. It would then recommend new segments to target, like “marketing operations specialists” or “data strategy consultants,” and spin up lookalike audiences from our best lead profiles automatically. This process created continuous, micro-segments based on actual user behavior, which is a world away from standard A/B testing.

The AI also helped with creative optimization. It couldn’t write new copy or design images, but it did watch the performance of every single ad variation we threw at it. If a headline or image wasn’t working, the AI would pause it and move that budget over to the ads with higher CTRs and conversion rates. This freed up our creative team to work on new ideas instead of getting bogged down in spreadsheet analysis.

Creative Approach: Emphasizing Data Cohesion

Our creative plan was built around showing the pain of having your marketing data all over the place and positioning Improvado as the fix. We ran with two main creative ideas. “The Data Maze” used infographics and short animated videos to show how messy disconnected data sources can be. “Unified Insights” presented clean, actionable dashboards as the solution. All our ad copy hammered home benefits like having a “single source of truth for marketing data” and being able to “accelerate reporting by 50%.”

On Google Search, the ad copy was very direct and packed with our main keywords to capture problem-aware searchers. For Display and LinkedIn, we leaned on strong visuals. A split-screen ad on LinkedIn that showed a chaotic mess of logos on one side and a clean dashboard on the other really hit home with our audience, since they’re the ones who deal with data silos every day.

Targeting Evolution: Beyond the Obvious

We started with the usual suspects for targeting: job titles like Marketing Director or CMO, industries like SaaS and E-commerce, and company sizes of 500+ employees. But the AI agent spotted something interesting pretty fast. While those people were fine, the engagement and conversion rates were way higher for people who were also interested in “marketing automation platforms,” “business intelligence tools,” and “data visualization.” That finding changed our whole approach.

The AI suggested we broaden our LinkedIn targeting to include “growth-focused marketing managers” and people in specific groups for martech and data science. We were a little hesitant at first because it felt too broad, but it turned out to be a great move. We even ran some Reddit ads in subreddits for marketing analytics and data engineering. The CPL for a whitepaper download there was surprisingly low, even though those leads didn’t convert to MQLs as well as the ones from LinkedIn. The lesson was that Reddit was a cheap way to fill the top of the funnel, but LinkedIn was still the king for getting bottom-of-funnel MQLs.

What Worked: Data-Driven Agility

The AI agent’s ability to make real-time, granular adjustments was the real win here. In the first month alone, our CPL fell 20%, from $85 down to $68. That was almost entirely because the AI was shifting budget to the keywords and audiences that were producing higher-quality leads (as scored by Improvado’s sales team). Bids for “marketing dashboard software” on Google Search went up 15% in some regions where intent was higher, while we pulled back a bit on broader terms like “marketing analytics.”

The dynamic creative rotation was also incredibly effective at fighting ad fatigue. When we saw the CTR for our “Data Maze” video on LinkedIn tank after two weeks (dropping from 1.8% to 0.7%), the AI had already started shifting budget to a newer “Unified Insights” infographic ad that was pulling a 1.5% CTR. This constant refresh kept engagement from falling off a cliff and made sure our ad spend was always on the best-performing assets. By the end of the 12-week campaign, we had generated 1,200 MQLs at an average CPL of $62.50, a 35% reduction from where we started.

Impressions and Conversions

When all was said and done, we’d served 8.3 million impressions. The conversion rate from an impression to an MQL came out to 0.014% on average, which sounds tiny but is actually quite strong for a pricey enterprise B2B product. After the initial optimization phase, the cost per MQL held steady in the $60-$65 range.

What Didn’t Work: Integration Hurdles and Attribution Gaps

The campaign wasn’t without its problems. Getting the AI agent to talk to Improvado’s Salesforce CRM was a real pain. There was a data lag, which meant the lead quality scores the AI needed for optimization were sometimes delayed by up to 48 hours. For the first few weeks, the AI was making decisions on slightly old data, which cost us some efficiency. We ended up building a nightly batch update to patch the problem, but it just goes to show how complicated it is to connect an advanced AI to older systems.

Multi-touch attribution was another tough nut to crack. The AI was great at optimizing for the final click, but figuring out the whole customer journey for an enterprise sale with a bunch of decision-makers was still a very manual process. Did a lead see a LinkedIn ad, then do a Google search, download a whitepaper, and finally convert after a sales call? Trying to figure out the true ROAS for each platform in that messy journey is difficult, and while the AI helps, it can’t solve the inherent limitations in most ad platforms’ attribution models. It’s a common headache, a 2025 IAB report found that it’s still a top challenge for 60% of marketers.

Optimization Steps Taken: Iteration and Refinement

After hitting those early roadblocks, we made a few key adjustments. The first fire we had to put out was the data lag, so we changed the connection between the CRM and the AI agent from a pull to a push system for lead quality scores. That got the delay down to under 6 hours and let the AI react much faster to performance changes, which was a critical fix for fine-tuning bids and audiences.

Next, we put a “negative audience” strategy in place on LinkedIn. The AI started to identify job titles and company types that looked good on paper but never turned into good leads or just had crazy long sales cycles. We then excluded those segments from our targeting to stop wasting money on them. Marketing agencies, for instance, were interested in the tech but rarely bought it, so the AI learned to stop bidding for them.

We also started feeding the AI better data from the sales team. Instead of just flagging a lead as an “MQL,” sales gave us feedback on “SQL” (Sales Qualified Lead) and “PQL” (Product Qualified Lead) stages. This gave the AI a much richer dataset to work with, allowing it to optimize for leads that were actually likely to move through the pipeline, improving the overall quality of leads, not just the count.

Finally, we got more aggressive with A/B testing our landing pages, with the AI watching bounce rates, time on page, and form submission rates. If one landing page variant was performing much better for traffic from a certain ad, the AI would start sending more people there automatically. This kind of continuous testing meant we were optimizing the whole funnel, not just the ads themselves, which is exactly what eMarketer’s 2026 Marketing Automation Trends report says more people are doing.

This campaign showed that while AI agents are powerful marketing tools, you get the most out of them when you’re careful about integration, set up strong feedback loops, and are willing to change your plan based on what the data tells you. It’s definitely not a magic button. It’s a partnership between your strategy and the machine’s intelligence.