Christopher Drum published this retrospective on 12 October 2026. He uses International Spreadsheet Day, the 15th anniversary of Steve Jobs’s death, and the first anniversary of his blog as the occasion for examining Lotus Improv.

The test system is a NeXTstation Turbo running in Previous 4.4 on Windows 11. It has a 68040 processor clocked at roughly 40 to 50 MHz, 128 MB of RAM, a 2 GB hard drive, a monochrome Megapixel display, NeXTSTEP 3.3, and Lotus Improv Rel 1.0, identified as v108r0.

NeXT and Improv were both attempts to rethink established computer interfaces. NeXT made UNIX accessible through a graphical system, Display PostScript, and unusually strong development tools. Improv treated spreadsheet data as a multidimensional model and provided formulas that read closer to ordinary language. NeXT hardware failed to reach a broad market, while NeXTSTEP and OPENSTEP later became part of the technology lineage behind macOS and iOS. Objective-C and Interface Builder also shaped the author’s programming career. Their influence remains visible in Apple’s NS-prefixed Foundation classes. DOOM and Tim Berners-Lee’s first web server were also created on NeXT systems.

Improv’s central idea is to replace anonymous coordinates such as B1 or K12 with named categories and items. A worksheet can define a Months category with January through December and a Days category with numbered entries. A new sheet begins with one cell and grows only as data is added. The model therefore stores the relationships the user defines, not a vast empty grid.

Formulas live in a dedicated panel and apply at the selected level of the model. A rule can target every December item or a single intersection such as December:15. The PREV, NEXT, and THIS keywords provide limited relative references. In one example, each January value is calculated from the previous day, while December values are set to twice their matching January values. The whole worksheet can be driven by two visible formulas, with no hidden copy in every cell.

Improv also exposes conflicts. If an item named March exists in more than one category, a reference can specify Months:March. When two formulas address the same result, the interface marks the overlap and asks the user to choose the winning rule. The development team once considered this problem unsolvable. Steve Jobs reportedly argued that making conflicts visible was a strength because the user could resolve them explicitly.

The model becomes harder to use as dimensions multiply. Drum builds a baseball dataset with Year, League, Division, and Record categories. Adding teams creates a structural problem because a new category applies across every existing dimension, even though teams belong to particular leagues and divisions. Groups provide a partial solution. They collect sequential items inside a category and can be nested. The groupsum() function calculates subtotals, while SKIP prevents a formula from applying to selected groups, items, cells, or categories.

This structure also explains Improv’s relationship with pivot tables. Pito Salas developed the concept at Lotus Development’s Advanced Technology Group. His early prototype, called Back Bay, demonstrated English-like formulas, multidimensional data, and movable views. After roughly a year of development, the prototype toured the United States and impressed financial modelers. Improv probably deserves credit for inventing the pivot table in its modern form, although the author reserves judgment because another historical product may have an earlier claim.

Modern spreadsheet software can rearrange pivot views, but its pivot data usually exists as a separate snapshot. Changes to the source table require an explicit refresh. In Improv, each view operates directly on the underlying model. The comparison uses LibreOffice Calc 26.8, Apple Numbers, Microsoft Excel, and Google Sheets. Their filter and panel interfaces are familiar, yet they require more duplicated data and more manual linking for multidimensional calculations. Improv can express a complete win-ratio calculation as Win Ratio = Win / (Win + Loss). A comparable modern model may require several SUMIFS expressions and additional lookup logic.

Improv’s help system is another strong point. It offers an index, search, linked documentation, and a Point for Help mode that turns the cursor into a question mark. Clicking an unfamiliar control opens the relevant explanation. The NeXT version also has an item dispenser for generating sequences such as weekdays or years. The tool requires a starting item and was replaced by a more conventional list generator in the Windows release.

Charts open in a separate application called Presentation Builder. Users can place several charts, text, shapes, lines, and sound annotations on a PostScript page. NeXT’s Display PostScript engine preserves the output as high-quality vector data, although the interface spreads formatting controls across too many palettes. Exported files can contain separate vector objects and awkward masks, which makes later editing difficult.

The first release has serious limitations. It offers no undo. Formula order matters, but formulas can only be rearranged through cutting and pasting. PREV, NEXT, and THIS can work relatively along only one chosen category. There is no scripting, no cross-project worksheet system, and no shared category definition that automatically stays synchronized across worksheets. A project may contain up to 100 worksheets, with references limited to that project.

In 1993, NeXT abandoned its hardware program and moved to Intel, becoming NeXT Software before releasing OPENSTEP. Improv also moved to Intel and became a Windows 3.1 product. Version 2 added undo and LotusScript, revised the list-generation interface, and received a Japanese localization. It retained the limits on formula movement, relative references, three-dimensional worksheets, and category synchronization. Lotus also offered several competing spreadsheet and productivity products, including 1-2-3 for Windows, Lotus Symphony, Lotus Works, and Lotus 1-2-3 for DOS Release 4.

Reviews in Compute in July 1994 and PC World in February 1994 praised Improv’s modeling power while warning that it was a poor general-purpose spreadsheet. PC World still named it the winner of its Applications category in the magazine’s 10th Annual Awards for Technical Excellence. The product beat Quattro Pro 5.0 for Windows Workgroup Edition and WordPerfect 6.0a for DOS.

Lighthouse Design, founded in 1989 by a group led in part by Jonathan Schwartz, built a NeXT software suite after Microsoft declined to port Office to the platform. Its products included development and project-management tools, the presentation program Concurrence, and two spreadsheets. ParaSheet followed a conventional design. Quantrix closely followed Improv’s multidimensional approach and could import Improv documents.

Quantrix addressed many of Improv’s weaknesses. It added undo, reorderable formulas, scripting, synchronized categories across worksheets, inline item creation, multidimensional relative references, color, and a single Inspector for chart editing. It ran on OPENSTEP for Intel and NeXTSTEP on Motorola 68K systems. Lighthouse was later acquired by Sun, and its applications were placed in storage. A 1997 license notice offered most of them free to longtime users with a two-user license, excluding FrameMaker, OmniWeb, and JavaPlan.

Quantrix survived as a commercial product. Pete Murray, formerly Lighthouse’s chief engineer, is reported to have rebuilt it as a Java application through Subx in 2003. IDBS acquired Quantrix in 2010. The current product targets high-end financial modeling and remains sold through a quote-based model. Drum’s conclusion is blunt: Improv is an impressive modeling environment and a superb technology demonstration, but its strict structure made it a poor everyday spreadsheet for most users. For people who want its ideas in a more complete package, Quantrix is the practical successor.