Photo Memory Discovery EngineWhy people can’t find photos they know they have
Step 1

Which problems are worth solving?

What this shows
Each thing people use to find a photo, scored on how often people rely on it and how badly Google Photos serves it today. The score is adapted from Ulwick’s opportunity method: how often people rely on a clue stands in for importance, and how often the feature fails stands in for satisfaction.
How to read it
Higher up means more people rely on it. Further left means the app serves it worse. The shaded top-left corner is where needs are common and badly served. Switch the buttons to see the same chart using answers from our own survey.
WORTH SOLVINGWELL SERVEDLOW PRIORITY00551010HOW WELL GOOGLE PHOTOS HANDLES IT →HOW OFTEN PEOPLE RELY ON IT →Roughly whenRoughly when: importance 10.0, served 5.8, opportunity 14.2Who was in itWho was in it: importance 9.4, served 3.8, opportunity 15.0Text in itText in it: importance 4.8, served 0.8, opportunity 8.8Where it wasWhere it was: importance 3.2, served 5.0, opportunity 3.2What was happeningWhat was happening: importance 2.2, served 0.0, opportunity 4.4Why they took itWhy they took it: importance 1.0, served 0.0, opportunity 2.0What it looked likeWhat it looked like: importance 1.8, served 4.4, opportunity 1.8

Ranked by opportunity score

Above 10 is worth pursuing, above 15 strongly.

  1. Who was in it15.0the matching feature failed for 29 of 47
  2. Roughly when14.2the matching feature failed for 21 of 50
  3. Text in it8.8the matching feature failed for 22 of 24
  4. What was happening4.4Google Photos has nothing to match it against
  5. Where it was3.2the matching feature failed for 8 of 16
  6. Why they took it2.0Google Photos has nothing to match it against
  7. What it looked like1.8the matching feature failed for 5 of 9
Pursue Support Leave (Google is already fixing it) No matching feature
Step 2

Checked against the brief

What this shows
The brief rules out improving search in general. Each problem area is marked Pursue, Support, Leave (Google is already fixing it) or Out (outside the brief).
Pursue

Dates people rely on are often wrong

In scope. Roughly when is what people remember most, and they scroll by it when search fails.

Already being solved? Partly. Dates can be fixed one photo at a time; posts say date search got worse after the AI switch.

Pursue

No way to search by life events

In scope. People place a photo by what was happening in their life, but the app has no way to use that.

Already being solved? No. Trips and Memories group some photos, but you can’t search “around when I was ill”.

Support

Text inside photos isn’t found

Partly. Matters for prescriptions, bills and screenshots, the hardest photos in our form.

Already being solved? Partly. Text search exists but posts report it missing words.

Leave

Faces grouped wrongly

Partly. Who was there is a common clue, but fixing face recognition is Google’s core AI work.

Already being solved? Yes, actively. Face grouping is an existing, improving system.

Out

Captions people typed aren’t searchable

Out of scope. The person knows the exact word they typed, so their memory isn’t the problem.

Already being solved? Partly.

Out

Search quality in general, and the new AI search

Out of scope. The brief rules out improving search in general. We keep it in mind as a risk: people distrust AI search they can’t see into.

Already being solved? Google’s own roadmap.

Step 3

From the problem to what we test first

What this shows
Teresa Torres’s opportunity solution tree: the business goal, the problems behind it, solution ideas for each, and the assumptions we must test before building.
How to read it
Crossed-out ideas already exist or are being built by others, so they fail our test for a new idea.
Business goal

Raise the share of people who find a photo they remember but can’t precisely describe.

Today: 17 of 30 survey respondents didn’t even try the search box.

ProblemSolution ideasWhat we must test
Pursue

“My sense of when is off, or the app’s date is wrong”

122 complaints about time · the date feature failed for 21 of 50 people who remembered roughly when · survey: 7 of 21 found photos were from a different year than remembered

Life-moment anchors: find by “around when I was ill”, not by year
Date repair: read the real date from WhatsApp and backup file names
Fuzzy year window: search a year either side of what you remember
Testing People place photos by life events more often than by years
Supported File names carry the real date for WhatsApp images
Pursue

“I remember what was happening, not the words”

Google Photos has no feature for it · only 11 posts mention it · survey: 16 of 30 people ticked it

Life-moment anchors: events built from the photo library itself
Follow-up question chat: Ask Photos already asks follow-ups, and classmates are building this
Untested Life events can be detected from photo clusters, places and saved documents
Untested A stranger can pick the right moment in under 10 seconds
Support

“Things I saved for later have no story”

text inside the photo wasn’t found for 22 of 24 people · prescriptions were the hardest photos in our survey

Kept-for-later shelf: prescriptions, bills and IDs grouped by what they are
Better text search: already on Google’s roadmap
Untested Grouping by document type finds a prescription faster than search
Leave

“It mixed up the people in my photos”

face grouping failed for 29 of 47, but fixing face recognition is Google’s core AI work

Outside what a product change can fix