InnovateNow’s AI Boosts ROAS 18% in 2026

Getting a complete well-rounded customer view is what every marketer wants, and it’s only getting harder as AI agents start handling more customer chats. In this campaign teardown, we’re looking at how a B2B SaaS company, “InnovateNow,” used some pretty advanced AI agent attribution to connect all their customer data. The result? They boosted their return on ad spend (ROAS) by 18% in Q3 2026.

Campaign Overview: InnovateNow’s AI-Driven Q3 2026 Push

InnovateNow sells enterprise project management software, and they had a problem we all know: their data was a mess, making it impossible to see a clear customer journey. Their old attribution model, a basic mix of last-click and linear, was completely ignoring the work done by their AI sales assistant, “ProjectPilot.” That AI was busy talking to prospects on the site and in emails but getting zero credit. The whole point of the Q3 2026 campaign was to fix that by finally tracking ProjectPilot’s influence.

The campaign ran for three months, from July 1 to September 30, 2026, on a $250,000 budget spread across Google Ads, LinkedIn, and programmatic. Our main goal was to get more qualified leads and a better ROAS by properly crediting conversions that the AI agent had a hand in. For us, a qualified lead was any company with more than 50 employees that filled out a demo request form.

Strategy: Beyond Last-Click

Our big strategic change was to get away from old-school attribution. We saw that ProjectPilot was an active participant in the sales funnel. It was answering tough questions, pointing people to case studies, and even booking initial calls. Ignoring all that meant we were just guessing with huge chunks of our budget. So, our core strategy was to build a custom machine learning attribution model that could actually understand ProjectPilot’s conversational data.

This new model gave fractional credit to every single touchpoint, including specific AI agent interactions like “answered pricing question,” “provided case study link,” or “scheduled demo.” We went with a Markov chain model, which is particularly good for mapping out a sequence of events and seeing how likely each step is to lead to a conversion. To make it work, we fed it our standard ad platform data (impressions, clicks), website analytics, and, the real key, the detailed conversation transcripts and logs from ProjectPilot.

Creative Approach: Contextual Conversations

Our creative strategy went deeper than ad copy. It was about how ProjectPilot actually talked to people. For Google Ads, we created super-specific ad groups for long-tail keywords like “enterprise resource planning software comparison” or “agile project management tools for large teams.” The landing pages for those ads put ProjectPilot front and center, ready to handle the next question.

Over on LinkedIn, we pushed out a bunch of thought leadership, whitepapers and webinars on “AI in Project Management,” for example. We trained ProjectPilot to know when a user was coming from one of those LinkedIn campaigns, so it could offer smarter follow-ups based on what they just read. If you downloaded a whitepaper on AI, ProjectPilot might ask, “Did you find the section on predictive analytics useful? I can provide more examples of how InnovateNow uses it.” This kind of contextual awareness was miles ahead of generic chatbot scripts.

The ads and landing pages all had a consistent look that felt clean and efficient. We also tested ad creative that explicitly mentioned the AI, with copy like “Get instant answers with our AI assistant” in the ad extensions. Those ads did really well, which tells me people are getting more comfortable talking to AI.

Campaign Performance and Analysis

The Q3 campaign got great results, mostly because we could finally see what was actually working. Here’s the raw data:

• Budget: $250,000
• Duration: 3 months (July 1 to September 30, 2026)
• Impressions: 12,000,000
• Click-Through Rate (CTR): 1.8% (up from 1.5% in Q2)
• Total Leads: 25,000
• Qualified Leads: 1,800
• Cost Per Lead (CPL): $10.00 (Total Leads)
• Cost Per Qualified Lead (CPQL): $138.89
• Conversions (Closed-Won Deals): 90
• Average Deal Value: $5,000
• Total Revenue: $450,000
• Return on Ad Spend (ROAS): 180%

Just to compare, our ROAS in Q2 2026 was 152% on a similar budget, so this was an 18% increase in ROAS. The cost per qualified lead dropped from $180 in Q2 to just under $139, a 22.7% improvement. That’s a direct result of smarter targeting and better lead nurturing from ProjectPilot.

What Worked: The Power of AI Agent Attribution

The new attribution model was easily the biggest win. By actually tracking ProjectPilot’s impact, we found a bunch of touchpoints we’d been undervaluing. For instance, the model showed that prospects who spent more than 5 minutes chatting with ProjectPilot and asked at least three specific product questions were 3.5 times more likely to convert. That insight alone made us completely rethink our ad placements and landing pages to encourage more of those deep AI conversations.

We also found out that some of our programmatic display campaigns, which looked like duds based on last-click, were actually starting great conversations with ProjectPilot that converted later through other channels. Once the AI attribution model showed us this, we shifted about 15% of the budget away from high-volume, low-intent search terms and into programmatic and our LinkedIn content campaigns that were teeing up these AI chats.

A/B testing ProjectPilot’s chat flows paid off, too. Just by tweaking the prompts and response logic, we got a 7% lift in the rate at which ProjectPilot qualified a lead for sales. It turned out that asking an open-ended question like “What’s your biggest challenge with project collaboration?” worked way better than a closed one like “Are you looking for a project management tool?”

What Didn’t Work: Over-Reliance on Initial AI Engagement

We had one experiment that fell flat. It was a Google Search ad with creative promising “instant AI-powered solutions.” The ad had a great CTR (around 2.5%), but the landing page bounce rate was over 70%. People clicked expecting a magic button, and when ProjectPilot started asking normal qualifying questions, they just left. It showed us that you have to be realistic when setting user expectations for AI. We tweaked the ad copy to frame ProjectPilot as an assistant, not an oracle, and it fixed the problem.

We also ran into headaches trying to get ProjectPilot’s data to play nice with our older CRM. Our attribution model itself was solid, but getting clean data feeds from the AI logs into our data warehouse was a heavy lift for the dev team. It just goes to show you need a decent data infrastructure in place before you can do this advanced AI stuff.

Optimization Steps Taken

Based on what we saw in the data week-to-week, we made several changes on the fly:

1. Dynamic Budget Reallocation: Our model showed strong early-stage AI engagement coming from channels that last-click ignored, so we moved 15% of the ad budget to them. Our programmatic display spend, for instance, went up by 20%.
2. AI Agent Script Refinement: We read ProjectPilot’s conversation logs every week to find where people were dropping off. We then refined the scripts to be more helpful, like proactively offering a human contact option if the user seemed frustrated.
3. Landing Page Adjustments: To lower bounce rates, we rewrote the text on our landing pages to be clearer about what ProjectPilot does (and doesn’t do). This one change dropped bounce rates by about 10% on our main pages.
4. Enhanced Sales Handoffs: Sales reps started getting much richer pre-call briefings. ProjectPilot would summarize the prospect’s questions, pain points, and any content they’d viewed, which made the first sales call far more effective and improved the sales team’s closing rate on AI-qualified leads by an estimated 5%.

These weekly tweaks, all driven by the granular data from our AI agent attribution model, completely changed the campaign’s direction. It was a series of small, calculated moves backed by hard data.

From my perspective, too many marketers treat AI agents like a fancy website widget, not a real part of the customer journey. If you aren’t giving those AI conversations any credit in your attribution, you’re flying blind and leaving money on the table. The shift to a well-rounded customer view is about intelligently connecting your data, not just hoarding more of it.

This campaign proved that understanding AI agent attribution isn’t an academic exercise. It has a direct, measurable impact on your marketing budget and effectiveness. When you can see exactly which AI interactions help lead to a sale, you can target and optimize with a precision that was impossible with older models. This kind of insight is table stakes for competitive marketing in 2026.

According to a recent IAB report, more and more ad spend is going to channels where AI is the primary point of customer contact. If you ignore that trend, you’re missing huge chunks of data for your attribution. On top of that, a 2026 eMarketer study found that companies that successfully wire AI into their attribution models see their marketing ROI go up by an average of 15-20%.

Conclusion

InnovateNow’s Q3 2026 campaign shows that solid AI agent attribution is critical for building a well-rounded customer view. It lets you find the touchpoints that really matter and seriously improve your ROAS. Building a custom machine learning model that factors in AI agent interaction data is how you’ll find the hidden value in your marketing funnel.