The AI culling time audit: a log, not a guess
You can tell in one week whether an AI culling tool is capable of helping you. Telling what it actually saved you takes a season, because the answer moves with job type, frame count, tiredness and the week you shot three times. This is the log that answers it, the columns worth keeping, and what to read once four weeks are in.
A trial answers one question. A log answers the one that matters.
There are two different measurements here and most photographers only ever attempt the first. A before and after trial tells you whether assisted selection can help you at all: you time your normal process on a few jobs, run one comparable job the assisted way, and compare like with like. If you have not done that yet, do it first. The full method is in our before and after workflow audit, and it will save you from buying a fix for a problem you do not have.
This is the second measurement. You already pay for the tool, or you are three weeks into a trial and about to be charged. What did it save you across a season of real work, rather than across the one clean job you tested it on? That needs a running log, because a single result is a demo. Job type moves it. Frame count moves it. A dark reception moves it. So does the week you shot Friday, Saturday and Sunday and picked all three on Tuesday night.
The arithmetic that turns logged minutes into hours and money already exists, in the honest math on how much time AI culling saves, along with the costs the marketing version leaves out. Treat that article as the formula and this one as the instrument. The log produces the numbers you put into it, from your own studio, in your own year.
One rule before you start. A log records what happened, not what you hoped would happen. If a job went badly because you were slow, that job still goes in.
One row per job, written the same evening
Keep it small enough that you will actually fill it in. One row per job, eight columns, in whatever you already open every day. A note on your phone beats a beautiful spreadsheet you never touch.
| Column | What goes in it | Why it earns its place |
|---|---|---|
| Date and job type | Wedding, portrait session, corporate event, second shooter included or not | The whole answer lives in the sort by type at week four |
| Frames in | Total frames that reached the drive | Nothing else in the row means anything without it |
| Active minutes | You clicking, zooming, comparing, correcting | This is the number that limits how many jobs you can take |
| Waiting minutes | Previews building, files copying, uploads, frames being scored | This is the number that limits how fast you can deliver |
| Sittings | How many separate times you sat down with this job | Three short evenings cost more than one, and totals hide it |
| Correction minutes | Putting back frames it passed over, clearing ones it proposed | Where a saving quietly gives itself back |
| Picks and delivered | Frames shortlisted, frames finally delivered | Keeps the quality question attached to the time question |
| Note | One line, plain words | "Second shooter files arrived unsorted" explains an outlier a year later |
Write the row the night you do the work, or the next morning at the latest. Reconstructing a week of rows on Sunday produces a tidy log full of round numbers that all flatter whichever answer you had already decided on. If you miss one, leave it blank. A blank row is honest and a guessed row is not.
Set the boundaries once, then never move them
The clock starts when the first card goes into the reader. It stops when the selects are ready for editing, meaning the picks are chosen, rated or flagged, and sitting where your editing step expects to find them. Everything in between belongs to the log, including the catalog housekeeping that only exists because of the way you pick.
Pause the clock for dinner, for a phone call about a different job, for the school run. Do not pause it for the small tasks that are genuinely part of the work, like renaming a folder or fixing an import you started wrong. Those are the minutes that decide whether a saving is real.
Keep your minutes and the computer's minutes in different columns and never merge them. Twenty minutes of analysis that runs while you make dinner is not the same twenty minutes as sitting at a desk unable to start. A tool that halves your active time while adding an hour of upload has made your capacity bigger and your turnaround worse, and only two columns make that visible.
Then leave the definitions alone for the whole season. If you move a boundary in week five, you no longer have one log with twelve rows. You have two logs with six rows each, and neither is long enough to tell you anything.
Tag the job type, because a median of everything is a median of nothing
A twelve hour wedding in the rain and a forty five minute headshot session do not measure the same thing. If your log mixes them, the number at the bottom describes a job you have never shot. Tag every row and keep the tags coarse: full wedding day, wedding with a second shooter, portrait session, event, commercial. Five tags is plenty. Twelve is a filing system you will abandon in September.
The tag matters because frame count is not a smooth dial. A full wedding day is typically 2,000 to 4,000 RAW frames, and a shoot twice the size of your usual job usually costs more than twice the time, because near identical frames arrive in clusters and clusters are judged against themselves. We took that curve apart in culling time by shoot size, and it is the reason a tool can look brilliant on your family sessions and ordinary on your weddings.
Second shooter jobs deserve their own tag. Two cameras on the same ceremony scatter near duplicates across two folders, which is a harder problem than a burst, for you and for any software.
If you want something to compare your first rows against before your own median exists, our culling time benchmarks give a range by job type. Use them as a sanity check, not as a target. A benchmark that disagrees with your log is a question about your log, not an instruction to work faster.
The column that stops the log from flattering the tool
Minutes saved are only real if the gallery still holds up, so every row carries two counts alongside the clock. First, frames you had to put back: keepers the tool passed over that you found on review. Second, frames it proposed that you would never send a client. The first number is the one that decides whether the tool stays.
Check the misses against a list you wrote before the season started, not by scrolling everything again. For a wedding: processional, vows, first kiss, family formals, first dances, speeches, exit. For an event: speakers, sponsors, group photos, the wide room. Software narrows the stack. The list is what stops speed from turning into an omission you only hear about from the client.
Watch your keep rate across the season as well. If the share of frames you deliver starts drifting upward while your delivered count stays flat, the tool has not made you faster, it has made you more agreeable, and a bigger gallery is a longer edit. Keep rate is the cheapest early warning in the whole log, and it costs you two numbers you already have.
Do not review everything anyway
If nerves send you back through every frame after the assisted pass, your workflow has not changed and your log will show a saving that does not exist in a normal week. Either trust the pass and check the list, or note plainly that you did a full review, and count those minutes.
Fence off the setup and the learning weeks
The first job with any unfamiliar app carries an hour of settings, preferences, one false start and a search for where the ratings went. That hour is real, it is worth counting, and it is not what next March looks like. Log it once, on its own line, marked setup. Do not smear it across job rows where it will make every early week look worse than it was.
Then treat the first two weeks as the learning tax and read them separately. Everyone is slower at first, almost nobody counts it, and it is the single most common reason a photographer concludes in week one that a tool did nothing.
Once both are fenced off, the break even question is easy arithmetic with your own numbers in it. If setup cost you ninety minutes and each job since has saved forty five, it paid for itself on the second job and everything after that is profit. If it saved twelve minutes, it has to survive a subscription, a fragile export step and one more login before it earns a place.
Check your control while you are at it. If you already pay for the Adobe photography plan, the thing you are comparing against is not nothing. Lightroom Classic and cloud Lightroom both ship Assisted Culling now, under Catalog Settings, Metadata, Assisted Culling, scoring subject focus, eye focus and eyes open, and Adobe describes it as optimized for portraits and headshots. We went through what it does and does not do in our Lightroom Assisted Culling review. A log that compares a paid tool against a manual baseline, when a capable one is already included in software you own, answers the wrong question.
What to read at week four
Four working weeks is the earliest the log says anything. A full season says more. When you sit down with it, read the median rather than the average: one fourteen hour disaster of a job drags an average somewhere your calendar will never go, and it is the typical week you have to schedule around.
Read it in this order, one number at a time.
| The number | What it decides |
|---|---|
| Median active minutes per job, by job type | How many jobs you can take next season |
| Median waiting minutes per job | The delivery date you can safely promise |
| Correction minutes as a share of the minutes saved | Whether you actually trust the pass |
| Misses on your always review list | Whether the tool stays at all |
| Sittings per job | Whether the work now fits in one evening |
Then look at the shape across the weeks, not just the totals. A saving that grows over the first month is you getting good at it. A saving that shrinks is usually one of two things: the novelty wore off and you started reviewing everything again, or the easy jobs came first. Both are worth knowing before you renew for a year.
Price the saved hour before you spend it
Now turn the log into a decision. Take the annual cost of the tool, divide it by the jobs you actually ran through it, and put that next to the median minutes it saved on those jobs. That gives you a cost per hour recovered, which you can compare with something real: your own hourly cost of picking, worked out in what an hour of culling actually costs you.
Decide what the hour becomes before it arrives, or it will quietly become more email. An hour is worth different things depending on where you are in the year. In February it can become a paid session or a portfolio edit. In the middle of a stacked run of weddings it should usually become sleep, which is the argument running through eight weddings in six weeks. A gallery that goes out a day earlier is worth more than an hour of active time saved, and both are worth less than not picking photos at 1am.
Where Kepla sits in this, honestly. The picking app for iPhone, iPad and Mac is still being built: the idea is a first pass on the phone on the drive home, then the same shoot open on the Mac for the frames that need a big screen, with nothing ever deleted, moved or renamed, so every frame it set aside stays one tap away. The booking page is live and free today. When the picking app ships, run this log against it exactly as you would against anyone else's, and keep the rows you already have. A tool that will not survive your own log does not deserve your season.
FAQ
How long should I keep a culling time log?
Four working weeks is the minimum, and a full season is better. Four weeks gets you past the learning period and usually covers more than one job type. A season is what tells you whether the saving holds in the busy stretch, which is the only stretch where it matters. After that, stop logging and only start again when something structural changes.
Should I use the average or the median?
The median, in almost every case. One unusually long job, a soaked camera, an unsorted second shooter card, pulls an average somewhere your calendar will never go. The median describes the typical job, and the typical job is what you schedule around. Keep the outliers in the log with a note, because the note is often more useful than the number.
Can I audit a tool I have already been using for a year?
Yes, with one addition. Log forward as normal, then pick one job manually as a control, using the process you used before. Choose a job type you shoot often and do not look at the tool's suggestions first. Without a control you are measuring your current speed, not the tool, because you no longer remember how long the old way took.
What do I do about the jobs I forgot to log?
Leave the row blank and move on. An estimate written a week later is not data, it is a memory, and memory flatters whichever answer you already prefer. A log with eight honest rows beats one with fourteen rows where six were reconstructed on a Sunday. If you keep forgetting, cut the number of columns rather than the number of jobs.
Does the subscription cost belong in the log?
Keep it as one line per month rather than a column on every job. At the end, divide the annual cost by the jobs you ran through the tool to get a cost per job, then set that beside the median minutes saved on those jobs. That single comparison is usually enough to make the renewal decision without any further arithmetic.
Measure the evenings, then take them back.
Kepla for Mac clears the obvious misses from a card, names the reason on every frame it sets aside, and leaves the choosing to you. Nothing is ever deleted, moved or renamed. Free through the private preview · the first hundred photographers keep it at $99 a year.