IPS Group’s 2026 AI Attribution Challenge

Listen to this article · 10 min listen

In mid-2025, the execs at IPS Group had a huge problem: they couldn’t figure out the ROI on their very expensive AI marketing campaigns. After Nayax bought them out in late 2023, the combined company was pushing hard into new markets, using AI for everything from finding leads to personalizing emails. Getting leads wasn’t the issue. The real challenge was untangling which AI touchpoints actually led to a sale, a messy problem known as AI referral attribution. Scaling their AI spend was just gambling without a clear view of what was actually working.

Key Takeaways

  • You need a multi-touch attribution model. It has to give credit to AI’s direct hits and its assists all along the customer’s path.
  • Use unique tracking codes for every single AI campaign and platform. That’s the only way to isolate where referrals are coming from.
  • Pull all your data, from your CRM, marketing automation, and AI tools, into one analytics dashboard to get the full picture.
  • Audit your AI model outputs and the attribution rules themselves all the time. Data gets stale, and your tracking will become useless if you don’t.
  • Set up clear KPIs for AI-generated leads that are tied to conversion rates and customer lifetime value, not just how many people clicked a link.

The Acquisition and the Attribution Abyss

The Nayax acquisition brought in a ton of tech, especially around data processing and machine learning. IPS Group, a company known for smart parking tech, suddenly got its hands on Nayax’s massive stream of data from unattended retail devices. This data fed new AI models that could predict demand, tweak pricing, and spot potential clients in ways they never could before. “We could see the AI flagging opportunities that our traditional sales channels missed,” Sarah Chen, Nayax’s VP of Marketing, said in an internal review in Q3 2025. “But when a deal closed, attributing that success solely to the AI’s initial nudge versus a human salesperson’s follow-up became a constant debate.”

Their first try at attribution was way too simple: if an AI found a lead that later converted, the AI got 100% of the credit. That fell apart fast. For example, an AI might flag the city council in Phoenix, Arizona, as a perfect target for smart parking meters and then fire off a personalized email sequence. But if a sales rep spent the next two months working that lead and finally closed the deal, how much credit does the AI really deserve for that initial flag versus the rep’s hard work? Without a solid AI referral attribution framework, trying to optimize their marketing budget was like flying blind.

Building a Multi-Touch Attribution Model for AI

The team’s first big move was to get away from last-touch or first-touch attribution. Those old models are easy to set up, but they completely misread the messy, non-linear buyer journeys that AI creates. “You can’t just give all the credit to the last click,” warned Dr. Alistair Finch, a data scientist they brought on as a consultant. “Especially with AI, the influence can be subtle and distributed.” He pushed for a weighted multi-touch attribution model, specifically a time-decay model that they adapted for AI interactions.

This new model gave more credit to AI touchpoints that happened closer to the sale, but it didn’t ignore the earlier nudges. So, if a predictive analytics engine flagged a high-value client six months before they signed a contract, it got some credit. But if an AI chatbot answered a critical question a week before the deal closed, that interaction got a much bigger piece of the pie. Of course, this meant they had to track everything carefully. Every single AI interaction, every AI-generated email, every AI-managed ad bid, needed its own unique tracking ID.

Implementing Granular Tracking Parameters

To get this done, IPS Group and Nayax had to completely redo their tracking protocols. They started sticking specific UTM parameters on every single link, email, and ad that an AI touched. An AI-powered email campaign, for example, might have a URL with utm_source=AI_PredictiveEngine&utm_medium=email&utm_campaign=SmartCity_Q3_2026. They went even deeper by adding unique IDs for the specific AI models involved. So if their lead-scoring AI, nicknamed “Project Chimera,” was the one that first surfaced a prospect, that fact was logged right alongside all the other touchpoints. This was the only way they could analyze the performance of individual algorithms instead of just looking at “the AI” as one big black box.

The real headache was getting this to work consistently across their different platforms. Nayax was using Salesforce Marketing Cloud for emails and Google Ads for search, and both have their own built-in AI. IPS Group, meanwhile, was running a custom-built AI that generated leads from sensor data. The only way to make sense of it all was to build a central data warehouse where every piece of customer interaction data, now tagged with AI attribution markers, could be dumped and organized.

Data Integration and Visualization: The Unified Dashboard

That data warehouse became their single source of truth. They piped data from their CRM (Salesforce Sales Cloud), marketing platforms, web analytics (Google Analytics 4), and internal AI logs into an Amazon Redshift warehouse. From there, they used Microsoft Power BI to build out dashboards that finally let them visualize the attribution data. For the first time, the marketing and sales teams could see a customer’s entire journey and exactly where and how AI had an impact, complete with its assigned attribution weight.

One dashboard they called the “AI Impact Tracker” was particularly useful, showing the percentage that different AI models contributed to closed deals. They saw that “Project Chimera,” the lead-scoring AI, was consistently responsible for about 18% of the initial pipeline, while their AI-driven email sequences were adding another 12% by nurturing those leads. This was about identifying where AI was most effective and where it was falling short. For example, they saw AI was amazing at finding public sector prospects, but the conversion rate on those leads was low, a clear sign that a human salesperson needed to get involved much earlier in that particular sales cycle.

Challenges and Iterations: The Reality of Attribution

Attribution is never perfect. A constant issue was the “dark funnel,” where a prospect might see an AI-generated presentation, then tell a colleague about it, who then makes an untracked phone call. You can’t track that directly. The team worked around it by looking for correlations. If AI-driven marketing activity surged in a specific territory and was followed a few weeks later by a rise in organic, untracked inbound leads from that same place, they could infer an indirect AI influence.

Another big issue was attribution decay. The AI models themselves were always learning and changing. An algorithm that worked great in Q1 2026 might get tweaked by the data science team and behave differently in Q3, changing its real-world impact. This meant the attribution model couldn’t be static. Dr. Finch’s advice was spot-on: “You have to treat your attribution model like another AI model,” he said. “It needs continuous training and validation against real-world outcomes.” This meant they had to do quarterly reviews of the attribution logic itself, auditing the weights in their time-decay model and adjusting them based on what was actually closing deals.

As a concrete example, they discovered their AI-powered pricing optimization engine, which suggested real-time pricing for parking spots, had a far more direct and immediate effect on conversions than they’d initially assumed. After digging into a few months of data, they significantly increased its attribution weight for deals where dynamic pricing was a core part of the sales proposal. You can’t just set this kind of system up and walk away. That constant refinement is what makes it work.

The Impact: Smarter Investments and Sharper Strategies

By the fourth quarter of 2026, their refined AI attribution framework was delivering real results. IPS Group and Nayax could finally reallocate their marketing budget with confidence. They moved money into the AI models and campaigns that the data showed were contributing most to actual revenue. A December 2026 internal marketing report showed they increased investment in their AI content personalization engine by 30% after seeing its direct effect on engagement and, more importantly, downstream conversions. They also figured out which AI models were great for lead gen but bad at nurturing, which let them build a smarter process for handing those leads off to the sales team sooner.

The framework also got the marketing, sales, and data science teams talking the same language. Data scientists got direct feedback on which models were actually helping close deals, so they could prioritize their work. The sales team started to trust the AI’s insights because they could see how it was setting them up for wins. The old, pointless debates over AI’s true contribution were replaced by data-driven conversations about how to make the whole system better.

The whole experience at IPS Group and Nayax shows that AI’s real power comes from our ability to measure its actual effect on the bottom line. If you’re using AI anywhere in your customer journey, a strong AI referral attribution framework isn’t a “nice to have.” It’s fundamental to smart, strategic growth.

Putting a full AI referral attribution system in place requires serious planning, detailed tracking, and a commitment to constantly tweaking it. When you can see exactly which AI touchpoints drive success, you can finally make decisions based on evidence, optimize your AI budget, and get much better results.

What is AI referral attribution?

It’s the method for identifying and giving proper credit to specific AI-driven touchpoints that help convert a customer. This means tracking how AI influences different stages of the customer’s journey, from the first time they hear about you to the day they sign a contract.

Why is multi-touch attribution important for AI referrals?

Because AI often influences a customer through many different interactions over a long period of time. A multi-touch model, like a time-decay or U-shaped model, gives a more honest picture by spreading credit across all those AI touchpoints instead of just giving it all to the first or last click.

What data points are important for tracking AI referrals?

You absolutely need unique IDs for each AI campaign and algorithm, specific UTM parameters on all AI-generated links, timestamps for every interaction, and customer journey data from your CRM and marketing platforms. Getting all of this into one unified system is the main goal.

How can businesses integrate AI attribution data from different platforms?

The standard way is to use a central data warehouse (like Amazon Redshift or Google BigQuery). This warehouse pulls in data from all your different sources, CRM like Salesforce, marketing automation tools, Google Analytics 4, and your own internal AI logs. Then you use a tool like Power BI or Tableau to build dashboards on top of that unified data.

What are the common challenges in AI referral attribution?

The biggest headaches are trying to track interactions across a mess of different systems, dealing with the “dark funnel” (influences you can’t track, like word-of-mouth), and managing data privacy. You also have to fight attribution decay as your AI models evolve and maintain strict data hygiene so your tracking codes don’t become a mess.

Editorial Team

The Editorial Team at AI Answer Growth is a group of writers and editors who cover technology, products, and how teams ship. They research each story and publish under a shared desk byline — experts in their fields, working as one team.