Christian Stender is the global head of AI for Tax & Legal at KPMG International. He specializes in tax organization and process consulting as well as on the setup and support of tax organization structures. In addition, Stender advises both medium-sized companies and globally networked corporations on the development and implementation of technical solutions in the tax environment.
Greatest achievement / personal milestone
For me, the single greatest milestone isn’t a specific product launch – it’s the cultural shift we’ve achieved across KPMG’s Tax & Legal practice, and the fact that we’ve achieved it together. Three years ago, “AI in Tax” was a research topic. Today, our Tax & Legal professionals around the world work with Digital Gateway every day – building personas, running agents, and embedding AI directly into client delivery.
What makes me most proud is that this is a genuinely shared success. On one side, our Tax & Legal colleagues globally – partners, directors and professionals – have embraced the change rather than resisted it, bringing forward use cases, challenging the status quo, and driving adoption from the front line of client work. On the other side, we have a technology team that is truly leading in this space – translating those ideas into production-grade capabilities on Digital Gateway at a speed that would have been unthinkable in our industry just a few years ago.
And we did all of this without compromising on defensibility. Every capability sits inside our Trusted AI Framework, with curated content, governed data ecosystems and human oversight built in from day one. Convincing a profession built on precedent and precision to genuinely trust AI – and doing so as one global team of business and technology – that, for me, is the real achievement.
Algorithmic liability / accountability
For us, the answer is clear: AI is an enabler for us – accountability stays with the acting tax professional. A model can assist, accelerate, and enrich the work, but it never produces a final work product on its own. No AI output is ever used directly or unchecked as a deliverable – every consequential result is reviewed, validated, and signed off by a qualified expert.
That principle is the foundation of our governance: We operate AI-enabled. The professional remains in the loop, in control, and accountable – and our processes, documentation, and quality framework are designed to make exactly that defensible.
Solving the “black box” problem
Our approach is to engineer the black box open, so that our tax and legal experts can follow and document all necessary steps.
Concretely, this means three things. First, we make the reasoning visible, which we call “Show Thought process.” For every answer generated through Digital Gateway, we expose the thought process, so the user can see how a conclusion was reached, not just what it concluded. Second, we make the evidence visible. Every source used to produce an answer is cited and can be opened in full text – no hidden retrieval, no untraceable claims. Third, we make automated workflows visible. Wherever we automate via our “Steam” functionality, a transparency principle applies to every individual step, not just to the final output – so each handoff, decision point, and sign-off is auditable.
Alert fatigue in mass transactional analytics
Mass data analytics only creates value if the signal is louder than the noise – otherwise you simply replace manual spot-checking with industrial-scale alert fatigue. Our approach is therefore to introduce the analytics wave by wave, not just a big bang.
In practice, we don’t switch on hundreds of rules at once. We onboard the rule set incrementally: A defined wave of rules is applied to the data, the resulting findings are analyzed jointly with the client, false positives and systemic data issues are isolated, and the rules are refined before the next wave goes live. That way, what reaches the risk team in the end is not raw noise, but a curated, prioritized pipeline of genuinely structural compliance issues.
Stress-testing against algorithmic audits (such as GoBD)
Tax authorities are no longer waiting for the annual audit. Under frameworks like Germany’s GoBD, they are already shifting from sample-based reviews to process and system audits powered by data analytics. Compliance officers should not wait until that machinery flags their flaws – they should run it on themselves first.
With today’s combination of AI and data analytics, this is very achievable. You can stress-test your own filings and positions by reconciling them against multiple internal and external data sets – ERP data, sub-ledgers, master data, contracts, transactional data – and let the algorithms surface inconsistencies before the tax authority does. The same logic that an auditor will apply to your data, you should apply yourself, continuously.
A book or film recommendation
If I had to pick one, I would recommend Co-Intelligence by Ethan Mollick – the most pragmatic and honest take I’ve read on how to actually work with AI: not as a replacement for human judgment, but as a partner that amplifies it. Which, as you might guess, is very close to how I think about AI in Tax & Legal.