From Human in the Loop to Expert in the Loop: Fixing 30 Years of DAM Metadata
- Old DAM collections were tagged for filing, not for today’s questions about provenance, consent and AI.
- Classic human-in-the-loop review suffers from fatigue, automation bias and single points of failure.
- Expert in the loop spreads decisions across the right people and lets consensus decide.
Why is old content in a DAM so hard to fix?
Every DAM has a folder nobody wants to open. Call it the kitchen junk drawer: the place where ten, twenty or thirty years of digital content was uploaded, sometimes dumped, before anyone thought about provenance, consent or AI disclosure.
Moving away from creative chaos is the main reason organisations invest in a DAM. It is a good first step, because it shows how big the mess is. But a DAM investment without a plan for metadata is incomplete.
Solid metadata is what makes a DAM succeed.
How did legacy metadata get so thin?
There are plenty of reasons. The administrator was in a hurry. A technical problem wiped fields. Someone planned to come back later.
The bigger reason is history. In the early days of DAM, metadata meant a filename, a description, a photographer credit and, with luck, a few keywords. Running the DAM was not a full-time job. It went to a junior creative, or to the intern.
What are organisations being asked about their content now?
The questions have changed:
- Can you prove where this image came from?
- Was it altered?
- Was any part of it generated by AI?
- Who owns the copyright?
- If there are people in it, did they give consent?
For many teams it is hard enough to answer those questions for last week’s uploads, let alone for thirty years of files in the junk drawer. Brands, publishers and agencies are now asked for content credentials nobody imagined would matter.
Why can’t AI or one reviewer solve it?
Retrofitting metadata and provenance to legacy content is a large job. Giving it to one person is exhausting. AI processing helps, but it makes mistakes and lacks the context to judge what matters. Correcting metadata still needs people.
That is where human-in-the-loop (HITL) workflows come in. In HITL, files are routed so a person can step in when data is missing, ambiguous, wrong or incomplete.
HITL improves data quality, but it brings problems of its own. People get tired. They start trusting the machine’s suggestion by default, which is known as automation bias. And no single person knows the answer to everything.
HITL can feel like a safety blanket, yet it often fails because the work is relentless and unrecognised. As anyone who has done it knows, adding metadata is boring and thankless.
What is expert in the loop?
That got us thinking. What if we recognised domain expertise and expanded human in the loop into experts in the loop?
Expert in the loop (EITL) is a review model in which the people who know the content best (brand managers, product specialists, archivists, rights owners) confirm, correct or reject metadata, including AI-generated tags, and their decisions are checked by peer consensus before they become the permanent record.
Human in the loop vs expert in the loop
| Human in the loop (HITL) | Expert in the loop (EITL) | |
|---|---|---|
| Who reviews | Whoever is available, often one administrator | People chosen for what they know about the content |
| Who decides | A single reviewer | Peer consensus across several experts |
| Workload | Long review queues | One-asset tasks that take seconds |
| Main risks | Fatigue, automation bias, one point of failure | Needs the right experts invited |
| Recognition | Usually none | Contributions tracked and rewarded |
| Best for | Simple, generic checks | Brand, product, people, rights and archive metadata |
How TagQuest puts experts in the loop
We wanted a lightweight system that could connect to a DAM, track metadata contributions and reward expertise, with friendly competition between invited individuals and teams.
We agreed on the name in five minutes. TagQuest turns adding metadata into a manageable process where invited participants check existing metadata, add new terms, and every decision is validated by peer consensus.
No single person’s best guess becomes the permanent record. The taxonomy improves too, because the terms experts add show where the official structure needs to grow. Introducing TagQuest
The digital assets already exist. Your quest, should you choose to accept it, is to improve the metadata that makes your DAM work.
Frequently asked questions
What is expert in the loop?
Expert in the loop (EITL) is a review model in which the people who know the content best confirm, correct or reject metadata, including AI-generated tags, and peer consensus decides what becomes the permanent record.
What is the difference between human in the loop and expert in the loop?
Human in the loop adds a person to an automated process. Expert in the loop chooses that person for their knowledge of the content, spreads decisions across several experts and accepts results by consensus.
Why does human-in-the-loop review fail on large archives?
One reviewer cannot know every asset, the queue never ends, and tired reviewers start accepting machine suggestions by default.
How does TagQuest use expert in the loop?
TagQuest invites the people who know each collection to confirm, correct or add metadata in short tasks, then accepts results by peer consensus and records who contributed.