v7.0.0

Fri Oct 30 2026

This release introduces the Express Model Builder, a much larger Alviss AI agent, reference periods for planning, variable limits, two-factor authentication and SSO, and a long list of speed and reliability improvements.


✨ Highlights

Express Model Builder

A new, faster way to build models. It runs alongside the existing advanced builder and does not replace it.

  • Build, fit and save models in the browser. Fits are quick, and Save is instant.
  • Run predictions, simulations, response curves, attributions and optimizations in your browser for an instant preview, or send them to the server as regular jobs.
  • Model versions are numbered across the whole model lineage, so two models saved from the same parent no longer share a version number.
  • In the Models table, model types are now shown as Express and Standard.
  • More than 15 rounds of bug fixes covering saving and reloading sessions, constraints, shared parameters, hidden KPIs, sampler fits and error messages.

Alviss AI agent

You can now ask the Alviss AI agent business questions in plain language, and it carries out the whole workflow for you.

  • It covers every type of job: predictions, simulations, optimizations, attributions and attribution sets, response curves, prior vs. posterior, and backtests.
  • It can create and refit models, validate and save model graphs, update variable metadata and project settings, and extend datasets.
  • Search tolerates typos and understands synonyms and acronyms ("forcast", "MMM", "what if we cut TV", "CPX per channel").
  • It can find and filter models and jobs by who created them and when ("all models I built today"), then rank and compare them.
  • You can describe a rough media mix in one request, e.g. "10M, ~30% TV, ~30% online, the rest elsewhere".
  • Results link straight to the matching page in the web app.

Reference periods for planning

  • Fill variables for simulations, predictions and optimizations with values from a past period you choose, instead of entering them by hand.
  • Shape future plans, including flighting and phasing, from a chosen past period.
  • Each value records where it came from (manual entry or reference period), and job detail pages separate the two.
  • "New from" carries reference-period inputs over to the new job.
  • Events are now filled from the reference window too.

Variable limits

  • Set minimum and maximum limits at project and dataset level. A dataset with values outside its limits is marked Blocked, and you can see exactly which values cause it.
  • A new Strict variable limits project setting chooses whether limits block jobs or only warn.
  • Negative values are now always refused for variables that cannot be negative, such as Media and Distribution.

Security

  • Two-factor authentication: authenticator app (TOTP) with backup codes, and passkeys.
  • Per-team MFA enforcement: team owners can require MFA for all members.
  • Self-service enterprise SSO: team owners and admins can manage SSO connections from team settings.

🆕 New features

Data and datasets

  • Derived variables: create new variables from existing ones when building a dataset.
  • Dataset window and combination filter: limit a dataset to a date range and a set of combinations. These settings are saved with the dataset.
  • Interpolation settings per variable, and a new Interpolation Lab tab.
  • Raw vs. interpolated toggle in the Data Explorer, with interpolated points marked on detail dashboards.
  • Data Quality report for uploads and datasets, with configurable checks, correlation metrics and a timeline.
  • Dedicated file editor page for uploads.
  • New Preparing sources dataset status while source uploads are processed.
  • Event data is shown as a days-active heatmap.
  • Country and region codes are shown with their full names.

Optimization

  • Revenue and profit targets..
  • You can now choose whether the target includes the pre- and post-period.
  • Goal types are renamed: Change versus baseline → Incremental, Absolute value → Total business outcome.
  • Boundary constraints can be edited directly in the table, and thresholds can be edited inline or in bulk.
  • Conflicts between constraints are detected before you submit.
  • The boundary section has a clearer layout: sorted by name, full names shown, and a sticky bulk-action bar.

Model building

  • Graph editor overhaul: new layout options, proper left-to-right graphs, and multiple connections per node input.
  • Save, name and reopen graphs, and insert an existing model's graph into the canvas.
  • Graphs are checked against the dataset before training starts, and you see a clear reason when a graph is invalid.
  • New holiday options: one-hot holidays, and start/end years per holiday (Denmark's Store Bededag now ends in 2023).
  • The training chart opens zoomed in on the meaningful part of the run.

Across the app

  • Lineage tab and explorer: see what each object is built from and what depends on it, with cascade delete.
  • A single bulk selection bar on every list page.
  • Variable Display Names now appear on every dashboard, comparison page and quality page.
  • Project notes..
  • Hide variables from the input pickers on simulations, predictions and optimizations.
  • Values you have entered on a simulation or prediction are kept when you change the period.
  • A new Source column on attribution, simulation and prediction lists shows whether a job came from a model fit, an optimization or another source.
  • Banner messages support markdown and different styles.
  • Accessibility and design-system improvements across the app, including dark-mode fixes.

⚡ Performance

  • Common pages load faster thanks to caching and fewer database round trips.
  • Data imports, datasets and results are stored and read as Parquet in object storage.
  • Large uploads are streamed instead of being held in memory.
  • The Express Model Builder is faster to use: fits warm up ahead of time, sampler fits run in parallel, and no work runs in the background after a fit.
  • Requests that are abandoned or exceed their time limit are now cancelled, so they no longer slow down everyone else.

🐛 Notable fixes

  • Baseline Variables on optimizations are now applied. Previously they never reached the optimizer.
  • Profit targets now subtract cost for each unit sold, not once per date.
  • Optimizations no longer crash when a variable's minimum and maximum bounds are equal.
  • Holidays that have been discontinued, such as Queen's Day (NL), no longer disappear. Christmas and New Year are no longer missed in one-hot encoding.
  • The "skip attribution" setting on holidays is now respected.
  • Performance metrics (CPX/ROI) now only count spend inside the attribution's window.
  • Attribution waterfalls now match the source job's target value.
  • Inserting a Basic Graph on data without Sales now shows a clear error instead of a job that fails later.
  • Prediction results now load for models trained on the whole dataset.
  • Many crash fixes found through error monitoring.
  • Security updates to dependencies across all services.

⚠️ Behaviour changes

  • Live Model Builder is now the Express Model Builder. Update any bookmarks to the builder page.
  • Profit optimizations now optimize a different quantity (cost per unit sold). Existing runs keep their results.
  • Goal targets now count only the optimized period. Direction goals act only as a tie-breaker.
  • Relative goals now work the same way in the AI agent as in the optimization engine.
  • Negative values for variables that cannot be negative are refused on every project, not only on strict projects.
  • The Group column has been removed from exported job CSVs.
  • Superusers now start in member mode and must switch on elevation explicitly.