Where this comes from
From 2022 to 2024 I worked as a Pega architect on RijkZaak — the case-management platform that Dictu, the Dutch government’s IT service, runs for its agencies — building the claims process for the Instituut Mijnbouwschade Groningen. Households damaged by gas extraction file a claim, the case moves through assessment and decision, letters go out, payments follow, and sometimes an objection comes back.
Two years inside that machine teaches you something no course does: most of the pain in government case handling isn’t in the case — it’s in the communication around the case. A citizen who understands where their claim stands is patient. A citizen who doesn’t calls the helpdesk, files a complaint, or submits a formal objection. Each of those costs the agency far more than the letter that would have prevented it.
That lens is how I look at everything “next-gen” Pega now ships. Not is the technology impressive — but would it have made that machine kinder and cheaper to run? Here are four places where I believe the answer is yes.
1. Decisioning, but not for selling — for communicating
Pega’s Customer Decision Hub was built to pick the “next best action” for bank customers. Strip away the marketing vocabulary and what remains is a machine that answers: for this person, right now, what is the most helpful thing to do — and on which channel?
That question is exactly the one a compensation agency faces every week, for every open case. Should this claimant get a proactive status update? A letter, or a portal notification, or a text? Is this case quietly heading toward an objection, and would a phone call from a caseworker this week prevent a legal procedure next month?
Today those judgments are made by rules of thumb and capacity. A decisioning layer on top of the case system could learn — from real response patterns — which proactive contact actually prevents escalation, and prioritize caseworker attention accordingly. The arbitration logic CDH uses for offers (propensity × context × value × levers) works unchanged; you just redefine “value” as avoided objection cost and citizen trust instead of revenue.
2. The Toeslagenaffaire test
Any Dutch reader will have had the same reflex just now: algorithms deciding things about citizens — haven’t we been here? Right reflex. After the childcare-benefits scandal, every algorithm a Dutch agency runs must survive scrutiny: explainability, bias monitoring, registration in the Algoritmeregister.
This is precisely where Pega’s adaptive models are an unusually defensible choice, and it’s worth being technically precise about why: they are naive Bayes classifiers, not black boxes. Every prediction decomposes into inspectable pieces of evidence — “cases with this damage type and this waiting time historically escalated more often.” You can print that on paper and hand it to an auditor, an ethics committee, or a judge. Try that with a deep neural network.
I’d still draw a hard line: models may prioritize help (who gets proactive contact first), never decide outcomes (whether a claim is honored). That line — and knowing where to draw it — is the actual skill in bringing decisioning into government.
3. Constellation is an accessibility program wearing a UI costume
Government portals in the Netherlands are legally required to be accessible (Besluit digitale toegankelijkheid). On the older Pega UI stacks I built with, accessibility is re-earned application by application, screen by screen. Pega’s Constellation architecture centralizes the design system, so accessibility and NL Design System alignment are solved once, at platform level — and every agency that onboards afterwards inherits it.
For a shared platform like RijkZaak, serving many agencies, that changes the economics of compliance. The migration itself is honest work — I’ve built on both generations, and the differences run deeper than looks — but it’s the rare modernization where the business case writes itself.
4. GenAI belongs in the letters (with a human holding the pen)
The functionality I’m proudest of from IMG is unglamorous: automated generation of case correspondence, with a rich-text editor so caseworkers could adapt letters to real decisions in real time. Letters are where a case system touches a human being.
That is exactly where generative AI should land first — and where Pega has pointed it. Drafting letters in B1-level plain Dutch (a standing government commitment that most correspondence still fails) is something a language model is genuinely good at, provided it works inside guardrails: approved templates, case facts it cannot invent, and a caseworker who approves every word before it ships. The newest Pega releases add document-processing and knowledge-assistant agents on the same principle — AI does the reading and drafting, people keep the authority.
Same rule as with the models: in government, AI proposes, humans decide. Design for that from day one and GenAI stops being a risk item and becomes what it should be — the end of incomprehensible government letters.
The honest summary
None of this is science fiction; every capability I’ve named ships in Pega Infinity today. What’s scarce is people who have carried both halves: the platform depth to build it, and the citizen-facing scar tissue to know where it helps and where it must never be allowed to decide. That combination is what I bring — and if your team runs case management for a government organization, I’d genuinely enjoy comparing notes.