DS Agent is the conversational interface of the DemandSens suite. Editors, collection heads and marketing leads ask questions in plain language — DS Agent pulls from market panels, your history and our models, and answers in seconds.
Every week, millions of sell-out, stock and trend data points flow into publishing houses. None of it was designed for the people who actually use it: editors, collection heads, marketing leads. They still reach for Excel, Outlook and intuition.
Every question an editor asks, typed as a feature. Pricing, reprint triggers, segment scans, portfolio alerts: each one is a question your teams ask every week, returned as an answer in seconds — with sources and confidence levels.
Pricing, reprint triggers, segment scans, portfolio alerts: each one is a question your editors ask every week, returned as an answer in seconds.
Behind every DS Agent answer runs a single prediction engine, tuned for publishing and contextualized per market. It surfaces in three places where editorial decisions actually happen.
Before a cover exists, before a print run is decided, editors already reach for comparables: titles close enough in shape and intent to anchor a forecast. DS Agent turns that reflex into a content-based filtering engine, running on the full metadata graph your data already carries.
Identify the most similar titles across the market portfolio (same publisher, same group, external) before a single copy ships.
Content-based filtering on master data and content description. Every field you carry (Thema · BIC · BISAC, publisher ID, collection, pagination, keywords, extract, price) becomes a similarity axis.
A scored ranking of comparables with minimum distance to the new title: the anchor for every downstream forecast.
Three modes of work (predicting a print run, anchoring a new title to its comparables, scanning a segment or a backlist), all accessed the same way: a question typed in plain language, an answer returned in seconds. The editor never leaves the conversation.
A 52-week print run projection per ISBN, with confidence band and sell-through envelope. Actuals compared to model live.
A ranked shortlist of comparable books from the market portfolio, scored on genre, theme, audience, price and format, feeding the frontlist model.
Segments accelerating or saturating, emerging themes, concentration shifts: 15 lenses, every one accessible by plain-language query.
Weekly feeds, normalized into a single schema, BISAC/Thema-aligned across markets. DS Agent consumes — never republishes.
LangChain-orchestrated models: velocity detection, reprint estimation, segment scoring. Tuned per-market, not a generic LLM.
Natural-language routing to 15 analytical features, multi-language (FR · EN · DE · ES · IT), ISBN-aware. Every answer traces its sources.
« For the first time, we're asking the data the same questions we ask each other at the morning stand-up. And getting answers before the coffee gets cold. »
A fully functional prototype runs today on simulated data. Five markets, five languages, fifteen features. Every element on this page is backed by working software.
The next generation of editorial workflows won't live in dashboards. They'll live inside the AI copilots publishers already use every day. DS Agent ships in two modes from day one: a hosted portal for the 80% who need an experience, and an MCP + REST surface for the 20% with their own AI stack.
We configure 2-3 lenses on your catalogue. You ask your real questions. You see the answers, the sources, the confidence levels. No generic pitch.
