Drill down path design that follows overview first

The screenshot arrived in a message with two words above it: this looks wrong. A bar chart of exhibitor counts by category, and the bars were about a third of the size everybody expected.

They were correct. The person had drilled the chart down to a single edition and then screenshotted it, and the title still read Exhibitor categories because the title had been typed in by hand two years earlier. Nothing in the image said which edition, which show, or that any filter was applied at all.

Drill down path design is the work that prevents that message. It covers the order the reader moves through the data, how many levels there are, and what the interface tells them about where they currently are, and most reports get the first of those right and the third wrong.

What order should a reader move through the data?

Ben Shneiderman settled this in 1996, in a paper called The Eyes Have It: A Task by Data Type Taxonomy for Information Visualizations, presented at the IEEE Symposium on Visual Languages and running from page 336 to 343. The line everybody quotes is the visual information seeking mantra: "Overview first, zoom and filter, then details-on-demand."

The paper sets out seven tasks in total, of which those four are the first four, with relate, history and extract following. It is worth reading the mantra as a sequence with a reason behind each step rather than as a slogan.

Overview first exists because a subset has no meaning without the population it came from. A category holding 61 exhibitors is a large category in a show of 412 and a rounding error in a portfolio of 3,900. A reader shown the 61 without the 412 will supply their own denominator and it will be whatever they last had in their head.

Zoom and filter come next because narrowing is a decision the reader makes, and it works only if they can see what they are narrowing from. Details on demand come last because record-level rows are expensive to render, expensive to read, and needed by perhaps one reader in ten.

The order also happens to match how a show director thinks, which is portfolio, then the show they own, then the part of it that looks odd. A drill path that matches the reader's own sequence needs no explanation.

One hierarchy per page

The failure that produces unnavigable reports is not too many levels. It is two hierarchies sharing a page.

An event portfolio has at least three natural hierarchies. There is the organisational one: portfolio, show, edition, day. There is the commercial one: exhibitor category, tier, account, contract line. There is the audience one: registration type, job function, seniority, individual. Each is sensible. Mixing two on one page produces a state where the reader has drilled to a category within a show but the page also holds a chart drilled to a job function, and no single breadcrumb can describe where they are.

Pick one hierarchy per page, name it in the page title, and let the other hierarchies appear as slicers or as separate pages. On a portfolio overview the hierarchy is almost always organisational: all shows, one show, one edition. On a commercial detail page it is almost always the exhibitor hierarchy. Two pages, two hierarchies, no ambiguity about what a drill means.

Depth is a smaller decision than it looks. Three levels covers a portfolio: all shows, one show, one dimension within it. Four levels usually means somebody has bolted a second hierarchy onto the end of the first.

Two clicks from portfolio to a category

Work an actual path. The portfolio holds eight shows and 3,900 exhibitor accounts across the current editions.

Level one is the overview: eight rows, one per show, each with exhibitor count, net square feet against target and a variance flag. The reader sees the whole portfolio and can compare, which is the only thing this level is for. The Chicago show shows a variance flag.

Level two is one show. Click the Chicago row and the page filters to Chicago's current edition: 412 exhibitors across nine categories, with the same three measures broken out by category. One click, and the reader has gone from 3,900 accounts to 412 while keeping the measures identical, which is what makes the comparison legible.

Level three is one category. Click the category showing the variance and you get its 61 accounts with a per-account view. Two clicks total, and at every step the measures on screen are the same three, so the reader learns one page rather than three.

Count what the reader has to hold in mind. At level three they are looking at 61 accounts out of 412 in a show out of eight, which is three facts. A breadcrumb reading Portfolio, then Chicago 2026, then Fluid Handling carries all three in about forty characters. The eight shows across five editions and nine categories give 360 reachable states, and a three-token breadcrumb names any of them unambiguously.

Notice what is absent. There is no fourth level down to contract lines. If somebody needs those, they belong on a record detail page reached deliberately, which is a different mechanism with different rules about what gets shown.

Why does the reader lose track of the filter?

Because in most tools the drill state lives in the visual and not on the page, and the reader has no reason to know the difference.

Microsoft's documentation for drill mode in the Power BI service, updated in June 2026, is direct about it: "By default, drilling one visual doesn't filter the other visuals on the report page." The report designer can change that, and many do not. So a reader who drills the exhibitor chart into one category is looking at a page where one visual is filtered to 61 accounts and the four tiles beside it still describe all 412.

The same documentation distinguishes two drill actions that look similar and behave differently. Go to the next level in the hierarchy moves the whole visual down a level. Expand all down one level keeps the current grouping and adds the next level under it. A reader who uses expand when they meant go to next level ends up with a chart showing both categories and accounts at once, which is a shape most people read as a bug.

Three things fix most of this. Put a text box on the page bound to the current filter state, so the breadcrumb is rendered rather than remembered. Set visual interactions so that drilling one chart filters the page, if that is what your readers expect, and if it is not, make the unfiltered tiles visually distinct. And put the filter state into the chart title dynamically rather than typing a title by hand in 2024 and leaving it there.

Drill inside a visual, or a page per level?

Two implementations, and I have a preference.

Drill inside the visual keeps the reader on one page and costs nothing to build, since the hierarchy is defined once in the model. It suits an analyst who is exploring and wants to move quickly. It has the filter-state problem above, and it means every level is displayed with the same chart type, which is rarely right: eight shows want a bar chart, 61 accounts want a table.

A page per level costs more to build and gives you a designed screen at each depth, with the chart type, the tile set and the labelling chosen for that level. It also makes the breadcrumb trivial, because the page itself is the breadcrumb.

For a report read by operating staff on a weekly rhythm I would build a page per level, three pages, and accept the extra work. For an analyst surface I would use in-visual drill and spend the saved effort on the model. The mistake is building a page per level and then also enabling in-visual drill on each of them, which gives you two navigation systems whose states can disagree.

Where this stops

A drill path assumes the reader's question decomposes along a hierarchy. Plenty of real questions do not.

Why is Chicago behind is a question about the interaction of several dimensions at once: a category that shrank, a region that stopped travelling, a price tier that moved. Drilling one dimension at a time finds the largest single contributor and hides the case where three medium-sized effects combine. A reader who drills into the biggest category, finds it down 4 per cent, and concludes that is the story will have missed the two categories down 3 per cent each that together matter more. Decomposing variance across dimensions is a different tool with its own post in this cluster, and a drill path is a poor substitute for it.

The second limit is that every level you add is a click, and clicks are where readers stop. A number the director needs weekly and can only reach three clicks down has been buried rather than organised, and it belongs on the overview itself even if that costs you a slot. Which numbers earn that promotion, and where they sit on the page, is the decision the drill path exists to serve rather than replace.

Take your main report this week and screenshot it after drilling two levels down. If the image does not say which show, which edition and which subset it describes, add a filter-state text box to the page. It is a ten minute change and it removes the single most common category of question your event business intelligence inbox receives.