An auctioneer publishes a catalogue for every sale. Photos, captions, estimates, everything has to fit a consistent layout, and the sale date never moves. At Giquello, a Paris house based on rue La Boétie, that production took several days of layout work per catalogue. The system delivered by Tandem now generates around 80% of the pages from a spreadsheet.
The remaining 20% is deliberate. It is reworked by hand by the designer, in her own tool. That split drove every technical decision on the project, and it explains why the system outputs an InDesign file rather than a PDF.
Why auction houses spend so long on their catalogues
The auctioneer's trade carries a rare constraint. Every sale produces a complete editorial object, printed and distributed, with a non-negotiable publication date. The sales held by the Giquello house cover twentieth century art, prints, drawings and design, with hundreds of lots to present per sale.
The unit cost of a catalogue looks bearable. Multiplied by the number of annual sales, it becomes a budget line of its own, and above all a calendar bottleneck. Two risks hide behind it. Dependence on graphic skills being available at the right moment. And loss of know-how, since the composition rules live in the heads of the people who apply them.
The real work, taming a catalogue full of special cases
The technical part took less time than the business part. A sale catalogue rests on decisions nobody ever writes down. When a photo deserves a full page. When two neighbouring lots share a double page. When a dark background showcases an object, and when it swallows it.
When the project opened, every sale and every speciality had its own habits. The same word covered different intentions depending on the page and on the person. As long as that vocabulary floated, every generation produced a surprise.
The work ran as weekly sessions with the business team, over about two months. Each session started from real, already published pages, to name what they were doing. The result is a written lexicon, then frozen, then turned into a dropdown in the spreadsheet. Nobody types a term by hand any more, so nobody gets its meaning wrong any more.
- Start from existing catalogues, page by page, rather than from a list of theoretical needs.
- Name each layout once, with a written definition and a signed-off visual example.
- Forbid adding unvalidated terms, otherwise the lexicon drifts again within weeks.
- Handle exceptions with tags, instead of creating one more layout for every isolated case.
The designer was not there, and the project had to cope
One organisational detail weighed heavily. Throughout the homogenisation phase, the layout role was in transition and the person who actually executes the layout was unavailable. The lexicon was therefore built with the business team alone, the people who know the rules for presenting lots.
Once the role was filled, a second pass with the designer became necessary. She brings a different eye, that of the person who opens the file and corrects it. Her feedback covered typography, logos, print margins and the quality of photo exports, all subjects a business team does not see.
80% generated, 20% reworked by hand
That split explains everything else. A system aiming at 100% would have to code every exception, at a cost that quickly exceeds the gain. A system stopping at 80% leaves control over what gives a catalogue its personality, and it costs a fraction of the first.
Why the file has to land in InDesign
Since part of the catalogue will be reworked, the output format becomes the most structuring decision of the project. The system writes an IDML document, the InDesign interchange format, which imports and edits like any other layout file.
- Paragraph styles are named, so a brand correction propagates everywhere in a single move.
- French typography is respected, non-breaking spaces included, which a generic export almost always gets wrong.
- Images relink automatically from the Drive, and every version produced is archived.
- Photos are never downloaded. The system reads their metadata to place them, which avoids moving hundreds of high-resolution files.
What the system produces today
| Item | Value |
|---|---|
| Lots processed in the catalogue | 245 |
| Pages generated in one run | Around 180 |
| Share of the catalogue generated untouched | Around 80% |
| Page layouts available | 26, in six families |
| Photos downloaded to compose | None |
| Composition time for a 210-lot catalogue | A few seconds |
| Output format | Editable InDesign document, dropped on the Drive |
The export button lives in the spreadsheet. It produces the cover, the opening and closing pages, the captions in the right styles, then drops the document on the Drive next to the photos. Data entry feeds from existing working documents, which avoids retyping the lots.
Under the hood, orchestration runs on n8n, with a spreadsheet for entry and checks. It is the same foundation described in our tutorial on building an AI agent in n8n, applied here to a document production chain.
Automating a layout, a request that comes up everywhere
This case looks like a niche. Yet the same request appears in three forms across almost every company Tandem works with. The product catalogue, the recurring report, and the sales deck.
Slide generation is the most visible version of the subject, with tools like Gamma promising a finished document from one prompt. Our comparison of AI presentation generators covers what those tools can do. They shine when the brand guidelines are loose and the content is short.
A sale catalogue sits at the other end of the spectrum. The guidelines are strict, typography matters, the volume runs past a hundred pages and the file has to stay editable. No off-the-shelf generator covers that, hence a bespoke chain. The right reflex is to place your own case between those two extremes before picking a tool.
What Tandem would do differently
- Bring the executor into the first session. The business team knows the rules, the person who opens the file knows the constraints. Both views are needed.
- Write the dictionary before the code. Every layout name deserves a definition and a visual example signed off before development.
- Announce the target percentage upfront. Saying a fifth of the catalogue stays manual avoids a disappointment and steers every technical decision.
- Build the error check into the first version. Flagging the faulty row costs little and avoids blind generations.
- Test on a whole catalogue. Edge cases show up at lot 180. A twelve-lot extract lets every one of them through.
This kind of project is what Tandem calls autopilot, a process that runs end to end with a human picking it up at the finish. Where it stops and an autonomous agent starts is covered in our article on AI agents versus AI workflows.
Should you automate your catalogue layout
Yes, as soon as three conditions hold. The volume is repetitive, the composition rules can be written down, and a person picks the file up at the end. If any one is missing, the project will produce a demo and nothing more.
The most underestimated condition is the second. A sale catalogue, a product sheet and an annual report all rest on implicit rules, and that formalisation work decides the outcome. If this is your situation, our AI audit exists precisely to map those rules before anything gets built, and our n8n agency offer builds the chain behind it.
