Data protection impact assessment: template

For a production using Shotlore with Google Gemini · 23 October 2026 · Not legal advice; a starting point for your data protection lead

A DPIA is advisable (and under UK and EU GDPR often required) where new technology, geolocation and footage of identifiable people meet. This template is filled in with what Shotlore does; the parts in [brackets] are the production's to complete. Print it or paste it into your own form.

1. The processing

ItemAnswer
Controller[Production company, address, contact]
ProcessorGoogle (Gemini API Paid Services) under Google's Data Processing Addendum, version [10, 7 May 2026], accepted on the production's own Google Cloud project [project name]. Nova Insight Ltd (the maker of Shotlore) holds no data from the job and is neither controller nor processor.
PurposeCataloguing the rushes of [production]: suggested keywords, captions and headlines per still or clip, and optionally transcripts of interviews, reviewed by a person before use, so the material is searchable and deliverable with metadata.
Data subjectsPeople who appear in frames (contributors, crew, members of the public), people who speak in transcribed clips, and the crew whose positions are recorded by Shotlore Field.
Personal dataImages of people at preview size; voices (transcripts) where Speech in clips is used; names typed by the production (People shown, credits); positions and times of the crew's day (Field). Special-category data only if the content itself reveals it (e.g. a documentary about health); [say whether it does].
SourceThe production's own footage and the crew's own phones.
Where it goesPreviews with instructions to Google's Gemini API under the production's key (see Security › What leaves). Positions to OpenStreetMap as cell centres. Field days to the production's own cloud folder, if shoot sync is used. Nothing to Nova Insight.
RetentionGoogle: 55 days for abuse monitoring, then deleted (Google's abuse-monitoring page, 9 June 2026). The production: the job, approved metadata and the send log for [retention period]; the shoot folder archived or deleted at wrap [say which].
International transfersGoogle may process in any country where it has facilities; Google is certified under the EU-US Data Privacy Framework and its UK extension, with SCCs in the DPA as fallback. For data of people in Kenya: [the safeguard relied on under the Data Protection Act 2019 and the ODPC's 2026 transfer guidance].

2. Necessity and proportionality

3. Risks and measures

RiskLikelihood / severityMeasuresResidual
Previews of identifiable people processed by a third party abroad[ ]Paid tier only (no training; DPA; DPF); confidential mode where agreed; on-device route for the most sensitive; send log records every call.[ ]
Google's 55-day abuse log holds content longer than the commissioner allows[ ]Disclose it; use the on-device route for that commissioner's unreleased material; Google Cloud route when available.[ ]
A free-tier key used by mistake[ ]Shotlore blocks confidential jobs and EEA/UK/CH computers without a confirmed paid key, warns elsewhere, and marks a key as free on a free-tier quota error.[ ]
Exact positions of sensitive species or private sites leave[ ]Sensitive locations on by default: rounded to about 10 km or withheld in exports, lists, reports, manifests, the files' own GPS if chosen, and in what goes to Gemini and OpenStreetMap; names withheld.[ ]
Face grouping amounts to biometric processing[ ]On the computer only; per job; only where the people have agreed; Delete face data at wrap. Not used for identification across jobs.[ ]
Crew positions (Field) reveal where people were[ ]Days stay on the phone and in the production's own cloud folder; Need-to-know keeps each member's days to themselves; Locked encrypts them and can delete them from phones after the production holds them.[ ]
Loss or theft of a computer or phone[ ]Disk encryption (FileVault/BitLocker); phone lock required on Need-to-know and Locked shoots; a lost phone is removed from the shoot folder and signed out of its account.[ ]
Inaccurate AI descriptions of people (wrong name, wrong attribute) reaching an archive[ ]Nothing is written without a person's approval; names are only those the production types; the model card lists the failure modes to check for.[ ]

4. Sign-off