Agencies find shots in a client's footage in four ways. Someone remembers it, someone logged it, someone searches a transcript, or the footage sits in a search tool that takes a written description of the shot and returns time ranges. Most agencies use a mix, and the mix changes as a client relationship gets older. On a new shoot, memory and the edit bins do the work. A couple of years and several editors later, only a log or an index can still answer "do we have a wide of the stage with the sponsor wall behind it?"
The people who were there
The fastest search engine at any agency is the editor who cut the project. It's also the least reliable one over time. It fails when that editor leaves, when the footage came from the client's archive and not your own shoot, and when the request is for material that never made it into a final cut. Scrubbing fills the gap for a while, and it works for a single interview. It doesn't work for a folder of event recordings. Handling a single request for an old shot walks through that case.
Logs, bins and selects
The traditional answer is structured logging at ingest: scene, shot type, who's on screen, and a note on anything worth flagging, kept in a spreadsheet or as markers in the edit project. Selects reels do the same job for the best moments. A good log is very fast to search. The catch is that it only holds what the logger thought to write down on the day, in the words the logger chose. If a client asks for "the moment the host points at the screen", the log probably didn't anticipate that question.
Transcripts
For interviews, testimonials and panel talk, a timecoded transcript finds lines well. It finds nothing in b-roll, crowd shots or product close-ups, because none of those have words. A library searched only by transcript is only half searchable.
A visual search layer
The newer route is to upload the client's footage to a search tool that indexes it once and then answers descriptions of what's on screen. You type the shot you want and get back time ranges, each with a short explanation of why it matched. The costs are an upload for each client, a subscription that meters usage, and a review step, because every result is a candidate. The limit is that these tools only find footage. Cutting it is still the editor's job. Most agencies set this up with one library per client. In Vivu, each client gets its own private project and every search runs inside a single project, so you search one client's footage at a time, and each range comes back with a one-line reason you can check against the frame. You can export the ranges you keep as original clips and hand them to the editor.
Describe the shot the way the camera saw it
Visual search matches what's visible, so descriptions work best when they stick to posture, action, framing and background. "The standing host holding a microphone and gesturing" finds the host at each point in a long recording where they stand up with the mic. "A wide shot of the panelists seated on stage with the backdrop behind them" finds the wide by its composition. Names don't help. A search tool can describe what someone looks like, but it doesn't know who they are. If you need shots of a particular person, describe how they look in that footage and what they're wearing.
Adding detail to a query tends to make the reasons more specific, because each detail gets checked against the frame. It also makes the query stricter, and that matters when nothing comes back.
When a search comes back empty
An empty result can mean two things. Either the shot doesn't exist, or the description was stricter than the footage. A recording shot mostly from the back of a room may simply have no close-up of a single speaker, and no query will produce one. Before you tell a client the shot doesn't exist, drop one condition at a time ("single", the hands, the framing) and see what comes back. If the looser queries return only wide shots, that's a strong hint the shot was never filmed, and it's worth a quick scrub to confirm before you give an answer.
When you don't need a search tool
If the client is new, the footage is from your own shoot, and the editor who logged it is still on the team, the bins are enough. The same goes for clients whose footage is mostly interviews, where a transcript covers most requests. A search layer becomes worth it once requests reach into footage that nobody on the current team has watched.
Where your gap is
Look at how your last few shot requests were resolved. If they were answered quickly from memory or from the edit project, your current system works. If they ended with someone scrubbing an old event recording, or with telling the client "we don't think we have that" without being sure, your gap is visual search, and that's the part to fix first.
FAQ
Can an agency search all of its clients' footage at once?
Not in a project-per-client setup. Agencies commonly keep footage separated by client, and search tools organized around projects run each search inside one project. To look for a shot across several clients, run the search in each client's project in turn.
Can visual search find shots of a specific person by name?
No. Visual search describes what it sees, such as appearance, posture and position in the frame, and it doesn't identify who someone is. Describe the person as they appear in that footage.
Do we still need to log footage if we have visual search?
Yes, but less of it. Record the basics at ingest (client, shoot, date, camera, usage rights), because a search tool doesn't know any of that. Leave shot-by-shot description to search, and keep human notes for judgments like which take the client approved.
Can we search competitors' ads the same way?
You can only upload footage you already have, so first check that you have the right to use it. For client work, stick to footage the client owns or has licensed to you.