Artificial intelligence promises companies more than just automation: it is supposed to make business decisions predictable. Predictive analytics platforms now enable specialist departments to derive concrete forecasts from data without in-depth programming knowledge. But how far has this development actually progressed, and what role does Europe play in this market?
Founded in Tel Aviv in 2018, Pecan.ai is developing a low-code platform for predictive analytics that aims to help business and data teams derive actionable forecasts from existing data more quickly. Among other things, the technology is used to predict customer churn, upsell, conversion, claims and optimise marketing measures and better plan demand trends.
In an interview with thebrokernews, Dan Goldenblatt (Managing Director, EMEA & APAC at Pecan.ai) explains how companies are putting predictive AI to practical use today, what challenges exist in its implementation, and why European companies in particular have great potential to integrate data-driven decisions more strongly into their business processes.
Dan, congratulations, Pecan.ai was just announced as a winner of this year’s Webby Award. The company is expanding rapidly in Europe. What differences do you see between European and US companies when it comes to the use of predictive analytics?
There is significant interest on both sides of the Atlantic. The key differences could be divided into three areas: regulatory, data strategy and risk philosophy.
Regulatory
In Europe explainability of predictive models is key. This is particularly important when using AI for claim management but even when predicting which customer will lapse or which customer is most likely to upsell, explainability is important. In that respect, Pecan AI offers a very high level of explainability and transparency with the added advantage of enabling non data-science teams to build models.
The US market is more complex because of the state level regulations that differ from state to state. But from what we are seeing, the US market focuses on fairness and non-discrimination rather than the technical audit trail required in Europe.
Data strategy
The EU is much more stringent when it comes to protecting privacy and, for example, use of non-insurance data from a variety of sources. In Europe the consumer must explicitly opt in to allow use of their data. In the US, on the other hand, carriers have traditionally used more sources of 3rd party data and predictive analytics is often used to create hyper-personalized marketing and on the spot underwriting.
Risk philosophy
Europe approach is “Predict-and-Prevent”: Because of regulatory restrictions on how they price risk, European insurers are using predictive analytics to prevent losses. They use IoT and satellite data to warn clients of floods or fires before they happen. On the other hand, the US’s approach is “Straight-Through Processing” (STP), whereby companies use predictive analytics to remove humans from the loop entirely with a goal of “Zero-Touch” claims and underwriting—using AI to approve a policy in seconds based on massive external data lakes.
Many European companies have large data sets but are still relatively cautious in their use. Where do you currently see the greatest untapped potential for predictive AI in Europe?
The real potential isn’t just predicting the future, it’s using those predictions to fundamentally lower the cost of risk for society, turning the insurance industry into a proactive safety net rather than a reactive payer. The EU AI Act is great for a company like Pecan AI because it requires companies to use models that are transparent and explainable, which is something that Pecan does “out of the box”. Switzerland is considered an innovative but also regulated market. What opportunities do you see specifically for predictive analytics in the Swiss financial and insurance sector?
Since the Swiss Federal Council emphasizes “accountability” over broad bans, the untapped potential lies in “White-Box Models” ensuring users can explain exactly why an AI recommended a specific asset, maintaining the human-in-the-loop is maintained.
In Switzerland, the greatest potential isn’t in ‘disrupting’ the traditional bank or insurer, but in digitally fortifying the Swiss brand of trust. In 2026, the winners will be those who use predictive analytics not just to increase margins, but to provide ‘Predictive Certainty’—protecting client assets and health before a risk even manifests.”
Many companies talk about AI but struggle with implementation. In your experience, what are the three biggest obstacles when companies want to actually integrate predictive AI into their day-to-day operations?
In our experience, the gap between an AI “pilot” and daily operational integration usually comes down to three specific friction points. In 2026, we’ve moved past basic “data quality” issues; the hurdles are now more structural and behavioral.
The “Data Silo” vs. Operational Context
Most companies have enough data, but it isn’t “AI-Ready” for real-time operations. Predictive AI requires a continuous stream of high-velocity data, but many European firms still store data in departmental silos (e.g., Claims data doesn’t talk to Policy-writing data).
The Obstacle: The model might work in a lab, but it fails in the “day-to-day” because it lacks the live context of the business.
The Solution: Shifting from static “Data Warehouses” to Data Mesh architectures where data is treated as a product, owned by the business units that actually use it.
The “Black Box” Trust Gap (Explainability)
In a highly regulated Swiss or European context, a predictive model that says “Reject this loan” or “Increase this premium” without an explanation is a liability.
The Obstacle: If a Relationship Manager or an Underwriter doesn’t understand why the AI made a prediction, they will revert to their gut instinct and ignore the tool. This is the “Shadow AI” problem—tools are bought but never used.
The Solution: Implementing Explainable AI (XAI) frameworks. By 2026, the gold standard is providing “Local Interpretable Model-agnostic Explanations” (LIME), which gives the human user 2-3 clear reasons for every AI prediction.
Change Management: From “Assistant” to “Agent”
The biggest hurdle isn’t the code; it’s the workflow redesign. Most companies try to “bolt-on” AI to their existing manual processes rather than reimagining the process around the AI.
The Obstacle: Employees often view predictive AI as a threat to their expertise or an extra step in an already busy day.
The Solution: Moving toward Agentic Workflows. Instead of the AI just giving a prediction (e.g., “This client might churn”), the AI acts as an agent that prepares the retention email, suggests the discount, and puts it in the employee’s inbox for a “one-click” approval.
The three hurdles are Context (siloed data), Confidence (lack of explainability), and Culture (failure to redesign workflows). To succeed, a company must stop treating AI as a software upgrade and start treating it as a new type of digital colleague.”
Pecan.ai relies heavily on low-code and automation. Is this development changing the role of traditional data scientists in companies?
The short answer to this is: No. In a similar way to how spreadsheets did not make mathematicians or accountants or bookkeepers redundant but rather allowed them to focus on more complex tasks, while allowing less highly technical folks to build sophisticated spreadsheets, Pecan AI does the same. We enable business users without technical know-how to build predictive models, freeing up valuable time for data scientists to address much more complex tasks. Having spoken to numerous insurance executives (as well as executives from other industries) we know that the less complex and more mundane modeling tasks are often at the end of the Todo list of data scientists. Even when those are addressed, getting the data scientists to maintain the models is always a challenge and often the downfall of the models. And yet, these use cases deliver significant value to the carrier. We see this with every new insurance company we onboard. Pecan helps business users quickly and easily build and deploy predictive models, reduce some of the huge demand for data scientists and improve the corporate standing of the data teams by reducing frustration of internal stakeholders not getting the modeling services they need. This is a win-win-win-win solution.
In practice, how long does it take for a company to achieve measurable business results after implementing your platform?
Nothing speaks better than an actual customer, an insurance customer, who in less than 3 months from kickoff, is already willing to give a testimonial. This is the case with the American Fetch Pet Insurance. Akash Gupta, the company’s Chief Analytics and Strategy Officer had this to say about Pecan:
“The Pecan platform is extremely useful for the data team and helps them accelerate their model building work manifold. Our team can now answer a lot more business questions much faster.
One very impressive aspect has been Pecan AI’s customer onboarding and support program. Pecan team of experts not only helped Fetch Data Scientists onboard on the tool, but also assisted with our data science models, making our team better. That kind of support is unparalleled in the industry.”
Predictive models are only as good as the data on which they are based. How do companies deal with incomplete or fragmented data when they want to use predictive models?
Pecan has made serious inroads when it comes to the quality of data required to build models. With proprietary IP for data ingestion and preparation, Pecan can use data that is BI ready and not what is traditionally known as “AI ready”. Naturally, a certain level of data quality is required but if a company has a data lake or data warehouse and is capturing data in a reasonable way, Pecan AI can work with that data and enable the users to generate valuable future facing insights from that data.
Many organisations are still sceptical about automated decisions. How do companies build trust in AI-based forecasts?
A healthy level of skepticism is good. It is cynicism that is destructive. And fully automated decisions by AI can indeed be problematic. There is a now (in)famous MIT report that talks about the fact that 95% of all AI projects implemented on the corporate level fail. The 5% that succeed are the ones that have a human in the loop. That is exactly Pecan AI’s model. We have a human in the loop, which is ensuring that predictions are accurate, valuable, reliable and executable.
With the EU Artificial Intelligence Act, Europe has developed one of the world’s strictest regulatory frameworks for AI.
What impact does it have on predictive analytics platforms such as Pecan.ai?
Pecan does not require any PII to be used to generate predictions. Knowing the name, addresses, phone numbers, birthdates or any other PII does not help the modeling. The platform enables users to determine exactly which data to share and which not and we highly recommend not to use any PII.
How can companies ensure that their predictive models remain transparent and explainable, especially in highly regulated industries such as insurance or banking?
Transparency and explainability are built into the Pecan AI platform. We have spent a lot of time and effort ensuring that this is the case and have been audited and vetted by a variety of companies in different industries that have confirmed this. Pecan AI is anything but a black box. More like a glass box showing the users how things occur, what the process was, the models, the hyper parameters that were chosen and what were the features that impacted each prediction on an individual customer level.
Insurers traditionally use data for risk models. How is predictive AI changing the classic actuarial logic in the industry?
Actuarial logic prices risk using historical loss data and predefined variables. Predictive AI breaks that constraint by ingesting thousands of signals – telematics, imagery, behavioral data, IoT – and finding relationships actuaries never thought to model.
The impact shows up across the value chain: sharper individual-level underwriting, dynamic reserving instead of static development triangles, fraud and litigation scoring at first notice of loss, and retention models that optimize for customer lifetime value, not just loss ratio.
The core tension is explainability. Regulators require defensible models, so ML hasn’t replaced actuarial logic – it’s pushed actuaries up the stack toward model governance and interpretation.
The remaining gap for most insurers isn’t insight – it’s operationalization. They have data science teams and proof-of-concepts, but predictions aren’t running in production or changing real decisions. That’s where Pecan sits: closing the distance between a model that exists and a prediction that drives action.
Beyond churn prediction and cross-selling, what new areas of application for predictive AI do you see in the insurance industry over the next five years?
Pecan AI is not an Insurtech company and we don’t have a specific insurance expertise. We are data experts and wherever there is tabular, structured data, the platform has and continues to prove itself as being very effective in building accurate and reliable models at record speed and at low cost. The proliferation of AI in the insurance (as well as other) industry, is expected to grow in the coming years.
The AI discourse is currently heavily influenced by generative AI. How do you see the interaction between generative AI and predictive AI in a business context?
That is exactly what Pecan AI did. We took a long existing process of predictive analytics and built a machine that automates it. Then we added an agentic component that allows for interaction between the user and the machine in a highly effective way that significantly expedites processes that used to take many weeks and often months and reduced it to hours or a few days.
In the long term, will generative AI enable business teams to develop complex predictive models without in-depth data science knowledge?
Partially, but not fully – and the distinction matters.
Generative AI will meaningfully lower the barrier for business teams to explore data, frame problems, and prototype models. Natural language interfaces already let non-technical users query datasets, generate feature ideas, and interpret outputs without writing code. That’s real and it’s accelerating.
But complex predictive modelling still requires things GenAI can’t substitute for: clean, well-governed data, sound problem framing, understanding of model assumptions, and judgment about when a model is wrong in ways that matter. Business teams tend to underestimate all of these until something breaks in production.
The more likely outcome is a new division of labour – business teams own the problem definition and outcome validation, technical teams own reliability and governance, and AI handles the heavy lifting in between. The analyst who understands the business and can work fluently with AI tooling becomes the critical profile.
For Pecan specifically, this is a tailwind: the harder problem was never building models, it was connecting them to decisions at scale. That remains true regardless of who builds them.
The questions were asked by Binci Heeb.
Dan Goldenblatt is Managing Director for the EMEA and APAC regions at Pecan.ai, where he leads regional sales and is responsible for the company’s insurance division. With over two decades of experience selling SaaS and AI solutions to large customers in utilities, insurance, fintech and other industries, Dan has a track record of building markets from the ground up and closing complex deals with multiple stakeholders. He built Pecan’s EMEA team and insurance division, closing deals with companies such as Fiverr, Markel, Fetch Pet Insurance and Savills (to name a few). Previously, as VP Business Development at App Orchid, he achieved over $2 million in annual sales in the AI/ML/NLP enterprise software space. Dan is based in Berlin and brings legal expertise, an MBA degree and deep industry expertise to every business engagement.
Read also: Pecan: Generative experience in the blink of an eye
Search:
Sponsors:
