Chris Puype

The through-line

Across his posts, Chris Puype keeps returning to one architectural idea: that credit risk, IFRS 9, Basel RWA and stress testing should live on the same governed platform instead of being stitched together through data movement between systems [1][9][15]. That conviction expands over time from a platform-design argument into a broader claim about how risk functions should operate under uncertainty, whether that uncertainty comes from climate and nature-related risks [5], from regulatory delay debates like FRTB and the Output Floor [8][12], or from the rise of agentic AI in banking [10]. Running alongside this is a second, sharper thread: impatience with claims that aren't backed by production-scale proof, whether that's a bank's capital buffer, an AI demo, or a regulatory "level playing field" argument [8][12][13].

On unifying the platform

Puype's core pitch is that asking a new question of a risk system should not require an ETL cycle. "For years, asking your credit risk system a new question meant moving data. Extract it. Transform it. Load it into another system. Run another batch. Reconcile the results." [1] His alternative: "If Credit Risk Modelling, IFRS 9 ECL, Basel RWA and Stress Testing & Forecasting operate on the same governed platform, you can keep the portfolio where it is and simply change the assumptions." [1] He frames this explicitly as a shift in mindset: "Stop moving data. Start moving risk assumptions" [1]. The same instinct shows up in his description of internal and regulatory models: "Internal models and regulatory models don't have to live in separate worlds." [9]

On proof over promises

Puype is consistently skeptical of claims that haven't been tested under real volumes. "In technology, it has never been easier to look impressive. AI can generate content in seconds. Demos can be carefully controlled. Screenshots can show only the best-case scenario." [13] For him the real test is production scale: "Can the platform process millions of contracts? Can it scale as demand grows? Can it continue delivering results when complexity increases?" [13] He backs this with concrete figures: "reduced processing times from 24 hours to one," "processed three million contracts in 12 minutes," and "maintained calculation times within an hour despite 900% volume growth" [13], summarized as "Claims are easy to make. Results are harder to fake." [13] The same challenge-us-directly posture appears when he invites clients to test the platform on their own data: "Give us the question and the data. We'll run it." [1]

On regulatory shortcuts

Puype pushes back hard on what he sees as regulatory retreat dressed up as pragmatism. On the EU's FRTB delay, he quotes Alain Papiasse approvingly ("We know there is a wake-up call. But everyone has a different clock." [8]) before delivering his own verdict: "Delay FRTB if you must. Just don't mistake 'following the US' for a strategy. The rules will land. The banks that are ready will have the advantage." [8] On the Output Floor debate he is even more direct: "Blaming the Output Floor for Europe's banking competitiveness gap is like blaming the scoreboard for the result." [12] He argues the real fixes lie elsewhere: "Fix Pillar 2 overlap. Complete the Banking Union. Strip out reporting that's purely operational cost with zero supervisory value." [12] and warns that "Weakening the floor doesn't make European banks more competitive. It makes them less comparable" [12].

On emerging and underestimated risks

Puype treats stress testing as only as good as what it targets. "Stress testing only works if you're stressing the right things." [5] He cites the ECB's view that "physical and nature-related risks are 'very likely being underestimated'" [5] and notes that "many banks are still building the data, models, and governance frameworks needed to assess them properly" [5]. He extends the same forward-looking lens to AI, describing agentic systems as a shift from insight to execution: "For years, AI has helped banks make decisions. The next step is letting it execute them." [10] He flags architecture as the binding constraint: "An AI agent is only as effective as the systems it can access." [10], while acknowledging "new challenges around governance, explainability, control, and accountability" [10].

From the stage

On the ElysianNxt Podcast, Puype goes into specifics that his written posts don't cover. In Episode 4 he discusses a white paper co-authored with Avantage Reply on "the risk function of the future," built from a CRO survey identifying pain points for chief risk officers [14]. In Episode 13 he walks through the Dragon Release (R10) in product detail: a lifetime PD at origination calculator for IFRS 9 that compares probability of default at reporting date versus origination, improved overlay functionality, Basel 4 revisions support with flexible CRM configuration, a new credit spread risk in the banking book component required by EBA guidelines, and enhanced reporting with data lineage tracking [15]. In Episode 14, hosting Pete Milla, the discussion surfaces the view that "stress testing means different things to different organizations" and that "banks have historically underinvested in proper stress testing capabilities" [16], a diagnosis more granular than the general stress-testing critique in his written posts.

Takeaways

• Push vendors and internal teams to justify why credit risk, IFRS 9, Basel RWA and stress testing sit on separate systems rather than one governed platform [1][9].
• Treat any AI or platform performance claim as unproven until it's shown under real production volumes, not demos [13].
• Don't treat regulatory delays (FRTB, Output Floor) as strategy-neutral; ask whether the underlying risk has actually decreased [8][12].
• Assume nature and climate risks are currently underestimated in stress testing frameworks and check whether data, models and governance are mature enough to capture them [5].
• When evaluating agentic AI initiatives in banking, examine the surrounding system architecture and governance controls, not just the model's decision-making capability [10].
• For IFRS 9 practitioners, watch for tooling that compares lifetime PD at origination versus reporting date as a way to isolate credit risk migration [15].