AI Spotlight

Import format detected Partner 07

Column C → net revenue ✓

Column F → territory ✓

Column H → billing month ✓

Approve Edit 12 columns

Run 02 · no model

Run 03 · no model

Run 04 · no model

The licensing and royalty platform of a global media group Read a Layout Once, Import It Forever Every streaming partner bills in its own layout. The AI reads it exactly once and a person approves — after that, deterministic code imports every subsequent statement, reproducibly.

AI agent MCP Engine

asks carries decides

probabilistic round

deterministic square

A global media group's licensing and royalty platform The Agent Asks, The Engine Decides Partner lists with hundreds of titles, resolved in a single MCP call. The AI agent phrases the request; the mapping itself is decided by the company's deterministic engine.

Chain of methods

exact match 12 %

spelling variants 11 %

similarity score 7 %

the rest needs meaning

Language model 70 %

In production The production-finance team at a major studio It Says Director, It Means Directing About 30 percent of a film budget can be matched automatically by comparing text. The rest only yields to meaning — and that falls to a language model at the end of the chain.

Proposal · title 4711 Diff

− The X Case (2019)

+ The X Case

+ Year: 2019

Approve Reject +34 in batch

In production A catalog platform for media distribution Every Change Gets an Approval Card Catalog work at deal scale: the assistant prepares every change as an approval card. The AI never writes to the catalog directly — approvals happen one by one or in batches.

What sets part 1 apart from the sequel?

Crew +12 % ▸ 41 rows

Equipment −4 % ▸ 18 rows

Post-production +31 % ▸ 22 rows

every figure sourced

Concept The production finance team at a major studio Analysis on Demand “What sets part one apart from the sequel?” An assistant translates questions like this into analyses of mapped budgets — with answers you can recalculate.

New message automatic

Cost report rev. 7 — 3 findings

Equipment +18 %

Shooting days unchanged — unusual.

Crew +34 %

Second unit from week 30 — plausible.

Concept The production finance team at a major studio The Cost Report That Explains Itself The moment a revision arrives, an agent analyzes the cost report and emails its findings: push instead of dashboard, context instead of rigid thresholds.

Attachment structured rows

Concept The production finance team at a major studio From Inbox to Finished Budget One agent pulls budgets from the inbox, imports and maps them — and a second one checks the result. What comes out is a draft by design, not a fact.

Import pipeline 22 projects

01 detect structure 22/22

02 map line items >90 %

03 validate result 2 open

no figure leaves the building

In production A regional film fund Every Budget Is a Different Spreadsheet Cost budgets arrive in every Excel format imaginable. A three-stage pipeline reads them in and maps more than 90 percent of line items automatically — without a single number leaving the building.

2 discrepancies

Model A Model B Difference

In production A regional film fund The Invisible Second Opinion A second model checks every import result without seeing the first one's reasoning. It warns, it never overwrites: a displayed 100 percent has to be a real 100 percent.

Cover

one source six formats

Concept A catalog platform for media distribution Every Channel Wants Its Own Assets Every channel demands its own image set. The AI generates into the specification — from cover to title treatment — and humans sign off on every asset.

Portal

AI client

events.read()

same answer, one tool

In development A German film industry trade association The Portal You Never Have to Open The entire members’ area as an AI interface – reachable in the portal chat or via MCP from a member’s own assistant. One toolbox, two channels, writes only with approval.

2 lines 1 page Detail

Market report Q2 · key finding

In development A German film industry trade association Reports Members Actually Read A stream of PDF market reports becomes readable: summaries at three zoom levels, trends over time, and a search that understands meaning, not just words.

interviews.dekkode.com/r/ 8fK2

“… and what was new about that?”

Server Question 3 of 7

Our voice interview platform Your Next Interviewer Is a Link Open the link and you are talking to a voice agent. The model carries the conversation, the server decides the next question — and audio never touches our servers.

Scoring · proposal 3 criteria

Depth 4/5

“… we solved that with Mercure”

Concreteness 2/5

“… went pretty well, really”

Decision deterministic

Our voice interview platform The Voice Never Grades Good listeners make lenient judges. So a second model scores every answer with cited evidence for each criterion — and deterministic code has the final word.

Interview #2417 14 min · DE

Speech → text 4.210 T

Conversation model 18.930 T

Scoring 6.400 T

Total 0,41 €

Our voice interview platform Every Interview Comes With a Receipt Every run carries an itemized cost receipt: which model, how many tokens, for what. Across all providers, priced only at display time.

Cinema Production Group

Concept A German film industry trade association One Report, Read Three Ways Three persona agents read the same report through the eyes of different member groups, then debate it like a newsroom. Every readership gets its own findings.

Portal PR News Trade feed + 12

Collector agent

Digest pipeline already in place

Briefing

Concept A German film industry trade association The Market Briefing Writes Itself A collector agent turns scattered public industry news into a weekly market briefing — powered by the digest pipeline the association already has.

PDF List Report Table

Agents

Extraction

Funding database queryable

Concept A German film industry trade association From Funding Reports to a Funding Database Funding data is public, but scattered across reports and PDFs. Agents and Document Extraction turn it into a queryable funding database.