Every release in the open, with the screen to prove it.
Novi ships continuously. Here's what's landed recently: the new capability, where it lives in the product, the nature of the change, and a look at the actual screen. Newest first.
The month Novi learned to run the part of planning that is about people. The headline is a redesigned Contributions: the budget round, from the first department to the last approval, inside the model. Under it, links that match codes by rule, a register grid that groups, thirty formula functions, and forecasting that holds up on real data. Read the full release note →
Contributors and approvers share one task tree, with every department’s state and a progress bar for each approver. A task opens read-only until someone takes ownership; the form set takes over the tab bar with Submit, Send back and Lock in the toolbar. Numbers go straight into the model, there is no copy to consolidate, and state changes reach everyone as they happen. With the Contributions module.
When the codes of two systems do not line up, each side of a mapping takes a key: code, name, alias, attribute, a register’s column text, or an expression such as LEFT(Product, FIND(" ", Product) - 1). The panel counts what matched as you type, and next month’s new elements follow the rule on their own.
T-Tables and S-Tables group by up to three levels, including a calculated column or a date by month or quarter, over every row that matches the filters. A summary row totals each column over the filtered rows, a search looks across all columns, and the download saves every matching row, however many.
IF nests anywhere, with AND, OR, NOT and BLANK. MIN, MAX, powers, logarithms, dates that behave as they do in Excel, the start and end of a cell’s period, DATETOPERIOD, and BALANCE(Opening, In, Out): a roll-forward in one function. Time functions now work over calculated measures.
Seven statistical models compete on every series, the winner writes its forecast with 80% and 95% ranges, and the run reports its score against a seasonal-naive baseline and which series it could not forecast, and why. No cap on the number of series. With the Forecasting add-on.
Views that total by weighted average across many aggregates can open more than a hundred times faster. A link from a register into a planning table runs about sixteen times faster.
For standard users, an object appears in the explorers, search and every list only when an Access List grants it, and cannot be opened any other way. Modellers are not restricted. Models, Matrices and Maps are modelling areas, closed to standard users by default.
A model has a description, which CLARA drafts in your language for you to accept. The workspace shows every model as a card with its description, size and last change. Hover over a model, a dictionary or a row header and it says what it is.
CLARA says so when an answer was cut short and explains a failed step instead of stopping. It answers “is anything waiting for my approval?” across the whole workspace, and when docked beside a model or a dictionary it opens on that object.
Fifty articles written step by step in the running product, from orientation to the first two levels of building models, and three multi-part guides that build a whole model from an empty workspace. Orientation and the guides are also in Polish.
A long month. The headline is the Novi Gateway: a way to read the systems that run your company without opening your network to anyone. Under it, a spreadsheet loader, CLARA working by proposal, links that carry calculated results, and a dozen smaller things. Read the full release note →
A small service on a machine inside your network that connects outwards to your tenant. No port to open, no database to expose, no inbound firewall rule. It runs your extracts against a read-only account Novi never sees, and sends out only the result. SQL Server, PostgreSQL and MySQL directly, or the output of the pipeline you already run.
A CSV, Excel or Parquet file straight into a table from the toolbar. Novi shows how the columns line up with the dimensions, how many members matched, what it did not recognise and exactly how many cells will change. Nothing is written until you have seen what will happen.
CLARA works inside a context you choose, the models and tables a conversation may see, and stays there. Every change arrives as a proposal with a preview, and nothing is applied until you approve it. Models are hidden from CLARA by default; a modeller decides what it may see.
A text column can now key a link: map it onto a dictionary by code, by name, by both or with manual pairs, and filter which rows are read. A list of transactions with a cost-centre code becomes a planning table with no intermediate step. Links also carry calculated results now, so a weighted average or a margin arrives as the number you see in the source. Large loads are hundreds of times faster than in July.
REF() reads a value from another table at a coordinate you project. PARENT() climbs the hierarchy. ROUND, ABS, FLOOR, CEILING, FINDFIRST, FINDLAST and a set of text functions for calculated columns. The editor autocompletes all of them.
Column headers with a filter funnel whose values cascade, three-state sort, hide this or hide others, copy and download of distinct values. Create a dictionary from a column in one step. A T-Table can be built on any dictionary type, and its columns follow the dictionary as it changes.
A link that builds a dictionary can be bound to a saved view of its source, and multilevel hierarchies come from Level columns. Before a run deletes anything, the plan says what will go and asks. Source links run from an Automation, and picklists can be bound to any subset of any dictionary.
A new time dictionary can choose its fiscal year and take half-years, hand-built codes are normalised, and when a dictionary cannot be extended Novi explains why rather than failing quietly.
Drag the corner to fill a block. Drag to resize columns. Cells shown in thousands or millions are typed in those units. Picklists and dates copy as text and paste back as values. Export the current view, totals and formulas included, or all data, to a CSV that opens correctly in Excel.
The audit trail sits beside the model canvas and says what changed. A model records who created it and who deleted it. Tables and dictionaries show their last update date. Every Gateway event, from sign-in to access granted to each synchronisation, is written there too.
Building structure is modeller work, so for standard users the buttons that would only be refused are gone. A modeller can still let users edit a particular dictionary.
A dictionary with hundreds of thousands of elements opens in seconds.
The work that built the surface August landed on: scenarios, forecasting, time analysis, mappings and the theming system. Dictionaries, Tables, Scenarios, Forecasting, Dashboards and Stories, the whole surface, moving.
One-click variance analysis on any table: actuals against budget or forecast, with the absolute and percentage differences generated for you.
Scenarios that compute from other scenarios: build a variance, a blend or a what-if as a derived layer instead of re-entering data.
Classify scenarios by type (actual, budget, forecast, what-if) so plans organise and behave consistently across the model.
Built-in time intelligence on a table: period-over-period, year-to-date and growth measures available as standard, with no formulas to write.
Relative time references expressed directly on the time dimension (prior period, same period last year, year-to-date) as reusable modifiers.
Splice members together to stitch segments of a dimension into a single working view. Draft description: confirm what Splice does.
Set the as-of "turn" date for a table so actuals and forecast split at the right point in time automatically. Draft description: confirm what Turn Date does.
Define how members of one dictionary map to another: the basis for translating ERP codes and statutory charts into management views.
A matrix view across dictionaries to see and manage how members and hierarchies relate, in one grid.
One-click statistical forecasting straight from the grid: project a series forward without leaving the table.
A deeper forecasting toolset: more methods and controls for statistical projection, beyond the one-click baseline.
Coordinated colour palettes and styling applied across dashboards and Stories, so a whole reporting pack stays consistent and on-brand.
Ready-made time dimensions (years, quarters, months, periods) so every model starts on a consistent, properly structured time axis.
We ship in the open, and we'd rather show you than tell you.
These are live in the product. Get early access to put them to work, or read how they fit together in the docs.