Enterprise UX in the Age of AI, Part 3: Insight at Speed
Part three of a series on designing the tools behind the business. This time: Insights Over Information and Reliable at Scale.
Part two argued that agents are joining the org chart and deserve job descriptions. This post takes the next pair of principles, and they share a theme: trust. Can you trust what the system tells you, and can you trust it to be there when the whole business leans on it at once?
Principle 5: Insights Over Information
Enterprise users don’t want dashboards. They want decisions.
Raw data isn’t help; it’s homework. A screen that shows forty metrics and leaves the user to work out which three matter has outsourced the hard part. Good enterprise design does that work up front: it surfaces the anomaly, shows the trend against the target, and pairs the chart with the underlying table so the number can be interrogated, not just admired.
The best data tools follow a simple discipline. Every view answers a specific question someone actually asks in the course of their job. An infrastructure monitor pairs usage charts with alerts and the log lines behind them. A revenue view shows the delta, why it moved, and where to click to see the accounts that moved it. Big picture and drill-down, both essential, one click apart.
The AI twist
Gen AI looks like the endgame for this principle. Ask a question in plain language, get the answer in a sentence. No query language, no waiting for the analyst.
For thousands of people who could never write SQL, that is a real unlock. The query box replacing the query language is the biggest widening of data access since the spreadsheet.
But an answer with no provenance is an opinion with good grammar. When a model tells you revenue dipped because of seasonal churn, the first question is the old auditor’s question: says who? Confidence and provenance now belong in the interface itself. Where did this number come from, how sure is the model, and what would change the answer. The “show your workings” path from part one applies with full force: every AI-generated insight needs a route back to the source data in one click.
Treat the model as a brilliant junior analyst. Fast, tireless, occasionally wrong in fluent English. You’d never let that analyst present to the board without their sources. Don’t let the model.
Principle 6: Reliable at Scale
Speed isn’t cosmetic in enterprise software. Delays block operations.
Consumer apps get judged in calm conditions. Enterprise tools get judged at the worst hour: Monday morning login, month-end close, open-enrolment week, the outage when every support queue fills at once. Design for that hour. Paginate the big tables, cut the decorative animation, minimise the round trips, and remember the field worker on a weak connection.
A beautiful interface that buckles under load is a broken interface. Users don’t experience your design and your performance separately; slowness is a design property.
The AI twist
Model latency is the new page load.
For twenty years we’ve engineered pages toward instant. Now we’re inserting model calls that take two, five, ten seconds into workflows we spent a decade shaving milliseconds from. Sometimes that trade is worth it. The point is to make it a decision, not an accident. AI features need a latency budget: what runs live, what runs ahead of time, what streams progressively so the user can start reading while the rest arrives.
Cost scales the same way. A clever prompt applied to every row of a ten-thousand-row table is a bill, and a queue. Pre-compute what repeats, cache what’s stable, reserve live inference for what’s unique.
And reliability now includes the model itself. Models get rate-limited, degraded and deprecated. If your tool stops working when the AI does, you’ve built the AI as a load-bearing wall. Build it as a layer instead: the workflow still functions without it, just with less assistance. The old discipline holds. Workflows can’t wait, and “the model was slow today” is the new “the system was down”.
Where This Series Goes Next
Part four takes Total Ecosystem Coherence and Smart Onboarding: why agents that cross tool boundaries make consistency machine-critical, and how AI becomes the always-on onboarding layer.
The thread through this pair: trust is earned in the interface. Show the source, hold up under load, and the AI stops being a demo and starts being infrastructure.
Patrick Mooney is a senior design leader based in Ireland, with more than 25 years’ experience in UX, product and enterprise software, and the founder of Dublin UX. He writes at patrickmooney.me..