AI, Trained on Your Campaigns
Trophyst AI pairs a natural-language assistant with predictive models built from your own data. Ask questions, run campaigns, and generate reports by talking — and let machine learning forecast volume, price, and margin before you launch.
Two Kinds of Intelligence, One Platform
Trophyst AI works on two fronts — a conversational assistant that puts the whole platform a question away, and predictive models that turn your campaign history into a forecasting advantage.
Trophyst Assistant
A natural-language interface to your campaigns, products, and sales. Ask, act, and generate reports — available today via the CLI and through any AI tool over MCP.
Explore the AssistantPredictive Models
Specialist machine-learning models forecast volume, find the profit-optimal price, and predict margin — retrained automatically on every campaign you run.
Explore Predictive AITalk to Your Campaign Data
The Trophyst Assistant turns natural language into real work. Ask about performance, search products, create and update campaigns, adjust prices, and produce finished reports — all without clicking through a single screen. Write actions are permission-checked and always confirmed before anything changes.
It ships today as a command-line assistant, and it speaks MCP — so you can drive Trophyst straight from the AI tools you already use.
Ask Anything, in Plain Language
"How many campaigns ran in Q4?" · "What was our best campaign by ROI?" · "Which products have the highest margin in electronics?" The assistant answers from your live data — no reports to build, no exports to wrangle.
Get Things Done by Asking
Create draft campaigns, add products, and update offer prices through conversation — "Change the offer price for article 12345 to 299 SEK." Write actions are permission-checked and confirmed before anything changes.
Reports & Decks, Generated for You
Ask for a campaign wrap-up and get a finished deliverable — PowerPoint, Word, PDF, or Excel — built from real numbers. The "how did it go" report writes itself the moment a campaign ends.
Permission-Aware by Design
The assistant only ever sees and does what the signed-in user is allowed to. Every action runs through the same fine-grained permission model as the rest of Trophyst — and is logged for audit.
Works With the AI Tools You Already Use
Trophyst ships a built-in Model Context Protocol (MCP) server. Point Claude Desktop, Claude Code, or any MCP-compatible client at your Trophyst instance and it can read your campaigns, products, and sales — and take action — through the same secure, permission-checked tools.
No bespoke integration, no data exports. Standards-based, secured by an activation token you generate inside Trophyst.
Your Trophyst Instance
Live campaigns, products & sales — permission-checked
Forecast Every Campaign Before You Launch
Trophyst trains specialist machine-learning models on your own campaign history — so every campaign starts with a data-backed view of how much it will sell, what to charge, and the margin you'll actually achieve.
And with the AI Campaign Wizard, you can hand the whole job over: pick a campaign, scope the products, set a target margin, and the AI fills it with optimized products and pricing.
AI-optimized price · volume · profit
Three Specialist Models, Each Trained for One Job
Volume Forecaster
Predicts how many units a product will sell during a campaign before it launches — using campaign duration, store and channel reach, marketing investment, seasonality, and price.
Price Suggester
Evaluates the price–volume curve for every product to find the profit-optimal campaign price — protecting margin without sacrificing sales velocity.
Margin Optimizer
Goes beyond revenue to predict achieved margin, accounting for returns, settlements, and mixed pricing across HQ and store channels.
Trains on Your Own Data
Models learn exclusively from your historical campaigns and sales. Your data is the moat — predictions get sharper as you run more campaigns.
Retrains Automatically
Nightly scheduled training picks up your latest campaign results — or retrain on demand with one click. No data-science team required.
Never Gets Worse
Each new model is scored against the one in production. If it isn’t actually better, the previous version is kept automatically — every model is versioned with full rollback.
Fully Transparent
See accuracy scores (R², MAE, MAPE), feature-importance rankings, and a complete training log for every run — so you always know how much to trust a prediction.
You Stay in Control
Run each model Off, in Training-Only mode (learning quietly without surfacing predictions), or fully Active. Ease AI into your process at your own pace.
Knows When It’s Ready
A live data-readiness score (insufficient → minimum → good → excellent) tells you exactly how much history you have and when predictions become reliable.
Bring Your Own AI — Your Keys, Your Rules
Trophyst doesn't lock you into one AI vendor and never sits between you and your provider. Choose the model that fits your contracts and compliance — your prompts and data flow directly between your Trophyst instance and your provider.
Claude (Anthropic)
Best-in-class tool calling
OpenAI
GPT models via your account
Azure OpenAI
In your own Azure tenant
Self-hosted
Runs on your hardware, fully offline
No Markup, No Metering
You pay your AI provider directly at your own rate. Trophyst adds nothing in between.
Your Data Privacy Terms
Your existing agreement with your AI provider governs the data — no third party added to the chain.
Run It Fully Offline
For strict data-sovereignty needs, point Trophyst at a self-hosted model so nothing ever leaves your network.
See Trophyst AI on Your Own Campaigns
Book a demo and we'll show you the assistant, the MCP integration, and the predictive models working on real campaign data.