Most large organisations aren’t deciding whether to buy a digital asset management (DAM) platform any more. They’ve had one for years. It went through a rollout, a training programme and probably a migration or two, and now sits somewhere in the middle of the content supply chain doing a job. But what happens when an AI assistant starts looking for those assets too?
The appetite is real, the results less so. Bynder’s 2026 State of DAM report, which surveyed 2,000 leaders at organisations of 1,000+ employees already using a DAM, found 46% of marketers expected AI to lift productivity last year, but only 37% actually saw it. Bynder’s own explanation is that the divide isn’t about who’s adopting AI but about who has the foundations to use it effectively.
DAM vendors are moving quickly. Aprimo has introduced agents for tasks including metadata, compliance and production. Adobe is bringing AI into more of its content supply chain. The possibilities are interesting, but they raise a basic question: can your organisation tell which content is current, approved and safe to use?
This is the pattern we explored in The AI Value Gap. If people struggle to find the right content today, an agent needs more than a better search box to solve the problem.
The adoption gap most DAMs already have
For most organisations the problem isn’t the platform. It’s that the platform never quite became part of how people work. Henrik de Gyor calls this the DAM adoption gap: the DAM launches, but people keep asking colleagues where files are, save copies to shared drives, or search several systems for an asset they know exists somewhere.
The research is fairly consistent. Forrester’s 2026 data found 67% of DAM decision-makers struggle to reuse, update or retire existing content, and put it down to gaps in discovery and metadata. As Forrester frames it, search failures become trust failures, and once people can’t find the right asset they recreate it or pull it from somewhere unofficial. In a Bynder survey of 3,400 creatives and marketers, 51% said they’d wasted money producing assets that went unused because nobody knew they existed or could find them.
AI features haven’t changed this. They’ve mirrored it. When G2 surveyed ten leading DAM vendors earlier this year, some reported more than 75% usage of their AI capabilities while others put it nearer 11–25%. Six of the ten blamed trust gaps, integration limits or resistance to automation, and G2 concluded that AI adoption in DAM is increasingly a governance and operational maturity challenge, not a technology one.
Put simply, the organisations getting value from AI in their DAM are mostly the ones already getting value from the DAM itself.
Where does your DAM actually sit?
It helps to be honest about which stage you’re at, because each needs a different next step, and the wrong move at the wrong stage is where a lot of AI budget quietly disappears.
| 1. Repository | 2. Managed library | 3. Connected system | 4. Agent-ready | |
|---|---|---|---|---|
| Adoption | Central team only, most bypass it | Core teams, regions partly | Default source for teams and partners | People and systems draw from it |
| Metadata | Folders and filenames | Taxonomy, applied unevenly | Governed and mostly consistent | Consistent, enriched, machine-readable |
| Rights | Emails and spreadsheets | Some fields, rarely updated | Structured and enforced | Plus usage scope and AI provenance |
| Right next step | Fix adoption and taxonomy | Enrich and govern | Automate enrichment and variants |
Most organisations we work with sit somewhere between stages 1 and 3, often at different stages across brands or regions. Agents can help you move up a stage, particularly through enrichment, but they can’t help you skip one. Putting production agents on a stage 1 or 2 DAM mostly produces the wrong content faster.
Measure what the DAM changes, not how busy it is
A lot of DAM reporting still counts activity, such as logins, uploads and downloads. It’s easy to collect and says very little about whether the platform pays for itself. De Gyor argues for reporting outcomes instead, and it’s the right instinct. The measures that tend to matter are:
- Time to find an approved asset
- Self-serve rate: requests resolved without contacting the DAM team
- Reuse of existing assets, and fewer duplicates recreated
- Approval cycle time, and so time to market
- Rights exposure: assets still live near or after their expiry
This matters more now for two reasons. It gives you a baseline, without which any claim that agents “saved time” is guesswork. And the DAM is about to be judged in places most teams don’t yet measure at all.
AI search changes who’s doing the looking
Your DAM now has three quite different audiences, and most were set up with only the first in mind.
People searching the DAM. Natural language and visual search are now fairly standard, and a real improvement on keyword boxes and folders. But they surface whatever’s in the library, expired and duplicated assets included. Better search on poor metadata just finds the wrong asset faster.
Internal assistants. Aprimo’s MCP server, for example, connects approved content to platforms like Microsoft Copilot Studio, Glean and n8n. A colleague asks Copilot for “the latest autumn hero image” and never opens the DAM at all. That’s great, as long as “approved” really means approved.
AI search and shopping. This is where it stops being a back-office issue. Google has said Lens handles nearly 20 billion visual searches a month, and roughly one in five are shopping-related. ChatGPT Shopping, meanwhile, doesn’t crawl for products. Merchants push a structured feed, with a main product image as a required field. Those images, and the descriptions next to them, very often start life in your DAM.
Two of those three audiences can’t use judgement. A designer can usually spot an out-of-date image. A copilot or a product feed will simply trust what the metadata says, which is why maturity matters far more than it used to.
It’s not just search. Where could AI support first?
Discovery is where I’d start, but it’s not the only place AI is showing up in content operations. Following the same logic as our last piece, it’s worth judging each use case by the outcome it moves, not the feature it comes with.
Enrichment and discovery
AI can add tags, captions, alt text and other descriptions when content enters a library. Arthrex reported a 130% increase in content discoverability after applying Aprimo’s Metadata Agent to more than 100,000 images. It’s a vendor-published result, but the task is easy to recognise: help people find and reuse content they already have.
Variant production
AI can help create versions of an approved asset for different audiences or channels. Lumen reported cutting the time to produce four Meta ad variations by 65% using Adobe GenStudio. The case is interesting partly because it includes review and approval. Producing more versions only helps if teams can get the right ones out safely.
Brand and rights checks
Automation can flag an image that appears off-brand or out of rights. It still needs accurate information about what is allowed and someone to resolve cases the rules don’t cover.
Provenance
Teams need to know whether AI was used to create or change an asset, and what happened to it afterwards. The EU AI Act’s Article 50 transparency obligations apply to certain AI-generated content, depending on its type and use. Recording AI involvement when the asset is created is easier than trying to work it out months later.
Performance insight
The least mature, and potentially the most valuable. Once performance data sits with the asset, the DAM can tell you what to reuse, refresh or retire, not just what exists.
| Use case | Outcome it moves | Maturity needed | Typically |
|---|---|---|---|
| Enrichment and discovery | Reuse, findability, AI search visibility | Stage 2+ | Quick win |
| Variant production | Speed to market, personalisation | Stage 3+ | Quick win once approvals are joined up |
| Brand and compliance checks | Risk, brand consistency | Stage 3+ | Foundation-first |
| Provenance and AI labelling | Compliance, trust | Stage 3+ | Foundation-first, but time-sensitive |
| Performance insight | Content ROI, sharper briefs | Stage 3–4 | Foundation-first |
Pick one job people are tired of doing
An Orange Logic customer started with tagging group photos. It already had consistent employee metadata, which it used to help identify people in new images. Orange Logic reported that the time required to publish group photos fell by roughly 70% within six months.
That figure is vendor-published, so it isn’t a forecast for another organisation. What’s useful is how they started: with a specific job, data they could trust and a result they could measure.
If you’re assessing an agent or another AI feature, ask:
Which job will it improve?
What information will it use?
What happens when that information is missing or wrong?
Which actions need a person’s approval?
How will you spot and correct a mistake?
What an agent-ready DAM tends to look like
It’s worth checking for all of these, rather than assuming from one.
The metadata means the same thing to everyone.
One agreed taxonomy for products, campaigns, markets and asset types, not three regional versions of it. Agents amplify inconsistency; they don’t smooth it out. The Orange Logic result only worked because the underlying metadata was already governed.
Rights are structured, not remembered.
If a person has to check an email to know whether an image can be used, an agent can’t know at all. Realtor.com and its parent Move Inc. are being sued by a photographer who licensed them images through 31 December 2024 and alleges they kept using them afterwards. That’s a very ordinary failure, and exactly the kind an agent would repeat at scale.
Scope is as clear as expiry.
A production agent needs to know not just whether an asset can be used, but what it’s allowed to become. In May 2026 a New York model sued Rainbow Shops, alleging it used AI to generate new images from her shoot when her contract only allowed minor edits. The claims are unresolved, but the lesson for DAM owners is clear enough.
Assets describe themselves.
Alt text, captions and attributes written for someone who can’t see the image, because increasingly that’s who’s reading. One Acquia customer’s PIM pushes product data into the DAM, which triggers AI-generated alt text and descriptions, with custom prompts behind each field so the AI follows instructions instead of guessing.
AI use is recorded at creation.
Every AI-generated or AI-altered asset is flagged, along with who edited it. That’s what turns Article 50 from a retrospective scramble into a field you already have.
Be a bit sceptical of the label
Not everything called an agent is one. Gartner expects over 40% of agentic AI projects to be cancelled by the end of 2027, and estimates only around 130 of the thousands of vendors claiming agentic capabilities are the real thing. Auto-tagging has been in DAMs for years, and variant generation is easy to demo but hard to govern. In any demo, it’s fair to ask what the agent decides without being told, what data it relies on, and how a person reverses what it did.
Humans in the lead
At Dexata we talk about keeping humans in the lead, and the examples above show why. Arthrex and the Orange Logic client got results because someone owned the rules. The Realtor.com and Rainbow Shops disputes are what it looks like when nobody does.
That still leaves plenty for people to own: the taxonomy, which actions agents can take alone and which need sign-off, what assistants and external channels can draw from, and the audit trail when things go wrong. It’s reassuring that this is already the norm. Across brand governance, metadata, quality and channel adaptation, 40–44% of Bynder’s respondents said automation does the work while people make the final decision.
It changes the DAM role, honestly for the better: less chasing uploaders to fill in fields, and more designing the rules the whole content operation runs on. As we said in MarTech COE: Has AI Made It Irrelevant?, the function doesn’t go away. It’s just accountable for more than it used to be.
Where to start?
Before the next DAM upgrade or agent add-on, it’s worth answering three questions honestly:
- Where does our DAM sit on the maturity scale today, and does that vary by brand or region?
- If a customer searched for our products through Google Lens or ChatGPT today, would they find the right images and descriptions?
- For every asset in active use, could we say when its rights expire, what it’s allowed to become, and whether AI touched it?
The answers usually split into the same two piles as last time: a handful of quick wins that can move this quarter, and foundations that need sequencing into the roadmap first.
That’s why DAM sits inside our MarTech & AI Diagnostic: start with what you need your content to do, assess what your existing platform can already enable, and build the roadmap from there.
If you’re unsure what an assistant would find in your content library, our complimentary MarTech & AI Value Snapshot is a place to start. We’ll look at how your assets are stored, found and approved, identify two or three AI use cases worth exploring, and flag the gaps that could hold them back. Drop us a line.
About The Author
