At most AI conferences, unstructured data and audit trails are side‑notes. At Alteryx Inspire 2026, they quietly became the main event. Between Andy MacMillan’s keynote, Ben Canning’s product roadmap, and a live recap with “Classification Guru” Susan Walsh and Alteryx’s Stephen Archut, a clear theme emerged: the next phase of enterprise AI will be won not by the flashiest model, but by those who can turn messy, unstructured content into deterministic, explainable answers that finance, audit and regulators can actually sign off on.
This is where Alteryx is trying to position itself: not just as an AI‑enabled analytics tool, but as a “logic and data engine” that grounds probabilistic systems in governed truth.
From Confidently Wrong Answers to VURA
CEO Andy MacMillan opened the day with a now‑familiar scene: a CFO asking a generic AI tool a mission‑critical question and getting a confident, beautifully phrased answer that is “probably wrong”. Down the hall, the CEO is emailing every department head demanding “20% more productivity” from AI, and nobody has a credible plan for how to get there.
His diagnosis is blunt: large language models are good at understanding intent and producing some answer, but often not the business‑correct answer. They lack the internal rules, policies and calculations that define “right” inside a given enterprise.
MacMillan’s proposed test for enterprise‑grade AI is VURA: answers must be visible, understandable, repeatable and auditable. If you cannot see where an answer came from, reproduce it on demand, and walk an auditor through every step, it doesn’t belong anywhere near a financial close, a regulatory filing or a strategic decision. That’s not a prompt‑engineering problem; it’s a problem of missing business logic and governance.
ALSO READ: Alteryx Announces New Enhancements to Alteryx One, Enables AI Operationalisation at Scale
The Logic Layer: Owned by Analysts, not Models
The keynote made a distinction that ran through the rest of the day:
- Data: access to warehouses, lakes, SaaS platforms.
- Governance: permissions, privacy, security, lineage.
- Logic: the way your business calculates revenue in California, commission for a given deal, or “AI‑ready” status for a dataset.
That logic layer, MacMillan argued, belongs with the people who understand the business — analysts, finance, rev‑ops, supply‑chain teams — not with IT alone, and certainly not with the model vendor.
In the live recap, Susan Walsh seized on this immediately. She has spent years “bridging the gap between tech and the business”, and her takeaway was simple: business logic finally has first‑class status. In her world, tools like Alteryx act as force multipliers, but only when the domain logic is explicit, consistent and applied to clean data. Without that, you get different numbers in every meeting and no‑one trusts the output.
ALSO READ: What to Expect at Alteryx Inspire 2026: Field Guide for CDOs, CTOs
Unstructured Data: From Liability to Signal
Against that backdrop, Chief Product Officer Ben Canning’s roadmap session put some hard numbers on the scale problem. Since last year’s Inspire, Alteryx users have processed hundreds of trillions of records and hundreds of petabytes of data through the platform. Increasingly, that isn’t just structured tables; it’s invoices, PDFs, call recordings, free‑text fields and other “unstructured” sources that have historically sat on shared drives and mailboxes.
Canning’s LiveQuery demo showed what in‑place analytics looks like for structured data: 10 million rows processed in around a minute directly in BigQuery, without pulling a single row down to a laptop — preserving cloud‑side governance and performance. But more interesting was what came next.
Using new capabilities for unstructured data, the team:
- Ingested a large batch of audio files,
- Transcribed them to text within the workflow,
- Ran sentiment analysis to classify positive/negative interactions,
- Joined those results with tens of millions of transaction records and tens of thousands of customer records in BigQuery,
- Surfaced “high‑value, highly unhappy” customers at full, millions‑of‑records scale.
That pipeline did two important things. It turned messy, previously under‑used content into structured signals and it kept the entire process inside a visual, versioned workflow that can be inspected, reproduced and governed.
ALSO READ: Q&A with Alteryx Chief Product Officer Ben Canning
For organisations drowning in call recordings, email trails, document archives and scanned invoices, this is a blueprint: treat unstructured data as a goldmine of risk, customer and operational insight — but always bring it into a governed workflow rather than letting it drive black‑box models on the side.
Deterministic Data in a Probabilistic World
In the recap session, Senior Director of Solution Marketing Stephen Archut put a name to what many finance and risk teams are chasing: deterministic data. In practice, that means taking inherently probabilistic AI systems — LLMs that can be creative and approximate by design — and grounding them in a layer of logic and data that behaves deterministically for critical questions.
It’s the difference between:
- “Here’s a plausible explanation of your margin dip” versus
- “Here is the reconciled, SOX‑signed‑off breakdown of exactly what drove that margin change, with a traceable workflow.”
Walsh’s own “COAT” framework (making data Consistent, Organised, then pushing it towards being Accurate and ultimately Trusted) aligns neatly with that view. Before any agent or copilot is allowed near a decision, the underlying datasets need consistent naming and categorisation; only then can you talk about accuracy and trust at all.
She also pointed out something often glossed over: accuracy is contextual. Finance may reasonably demand 100 per cent accuracy for statutory reporting, while sales and marketing can live with 5–10 per cent tolerance in pipeline or campaign forecasts. That nuance needs to be encoded into workflows and controls, not left to an LLM’s judgement.
Auditability, Explainability and The KPMG Moment
Several times, the conversation returned to audit and regulatory pressure — the “KPMG is coming tomorrow” moment. If an external auditor or regulator asks, “How did you arrive at this number?”, “ChatGPT told us” is not an acceptable explanation.
Here, the visual nature of Alteryx workflows becomes more than a usability feature. Archut noted that flows are inherently visible, understandable, repeatable and auditable — they encode every transformation and rule in a way that can be inspected and signed off. For corporate finance teams, that means being able to walk internal audit or external regulators through exactly how a number was produced, and by whom, with version history and approvals to match.
New platform features — versioning, labelling, approvals, and “AI‑ready” flags — are designed to make that audit surface explicit rather than accidental. The promise is that as AI features proliferate (Ask Alteryx, Agent Studio, embedded agents in BI tools and chat interfaces), the governance story keeps pace.
ALSO READ: The AI Infra Deals That Defined Q1 2026
In‑place Analytics and Data Sovereignty
Another thread that resonated strongly with both speakers was in‑place analytics: working with data where it already lives, without copying it into yet another platform.
Canning’s LiveQuery for BigQuery demo was one example: the Alteryx canvas drives transformations, but the data never leaves Google Cloud, remaining behind existing security and governance controls. For organisations dealing with data residency rules, multi‑cloud estates or tight security postures, that “zero‑data‑movement” stance can be the difference between AI projects being approved or blocked.
Walsh pointed out that many organisations are actively reluctant to move data out of primary systems or across borders. A pattern where tools like Alteryx orchestrate logic and workflows against those systems allows them to keep sovereignty and compliance concerns in check while still exploiting that data for AI and analytics.
Human in the Loop, Human on the Loop
Crucially, none of the speakers framed AI as a substitute for analysts. The consistent language was augmentation: “use it wisely, not blindly”.
- Ask Alteryx accelerates workflow creation but keeps outputs inside governed Designer flows.
- Unstructured data capabilities automate the heavy lifting while analysts decide what to include, exclude and how to interpret the results.
- New UX and Agent Studio experiences aim to make it easier for non‑coders to build, share and supervise workflows and agents without rewriting engines or forcing migrations.
Walsh highlighted curiosity and attitude as the real differentiators in analyst roles: skills can be taught; the instinct to question, to probe outliers and to understand domain nuance cannot. That is precisely what the logic layer depends on — and why, as both MacMillan and Canning argued, analysts and domain teams will remain central in an “agentic” future.
What this Means for Enterprise AI Leaders
Taken together, the keynote and recap point to a pragmatic agenda for organisations trying to move beyond AI pilots:
- Invest in the logic layer: treat business rules, definitions and calculations as first‑class assets, owned by domain experts and encoded in visible workflows.
- Clean and connect data — especially unstructured: invoices, calls, PDFs and free‑text fields are where much of your operational reality lives; bring them into governed pipelines before you let models near them.
- Demand deterministic answers for deterministic domains: finance, audit, regulatory reporting and critical risk use cases need explainable, repeatable outputs grounded in workflows, not just prompts.
- Keep humans in and on the loop: use AI to accelerate modelling, pattern‑finding and pipeline building, but keep domain experts responsible for what gets shipped into production.
At Inspire 2026, Alteryx’s answer to the AI hype cycle wasn’t another model announcement. It was a bet that the real AI transformation will happen where unstructured data, business logic and governance meet — and that deterministic answers built on that foundation will ultimately matter more than the shiniest new agent.
ALSO READ: Inside IBM’s 11 Billion Dollar Bet: What the Confluent Deal Reveals About AI’s Investment Paradox
AI & Data Insider is tracking these threads from Orlando — platform bets, leadership conversations and the reality of customer deployments — and will follow up with deeper coverage on the CPO’s product vision, Executive Summit field notes, and the key bets that will matter most for AI‑driven enterprises over the next 12–18 months. Register here for Inspire 2026 by Alteryx.
Organising an AI or analytics summit and looking for an independent media partner to translate it for the boardroom? Reach out to AI & Data Insider to explore a press collaboration. Write to us at editorial@aidatainsider.com
