Relativity’s Custom Analyses capability inside aiR for Review shifts e-discovery from a rigid, pre-configured environment to an extensible, user-defined engine. By allowing legal teams to define matter-specific data point extraction, classification, and privilege criteria in plain English, it eliminates the traditional trade-off between manual workflows and expensive custom database development, fundamentally changing how bespoke review protocols scale.
Every litigation matter is different. The data is different, the legal issues and risk are different, and the questions to answer do not rarely fit neatly into a pre-packaged coding layout. Historically, this created a compromise: use purpose-built AI tools for the limited tasks they were designed for and fall back on manual review for everything else. Relativity’s release of Custom Analyses is a direct attempt to close that gap.
Custom Analysis is one of five analysis types available in aiR for Review. For a full walkthrough of relevancy, key document, and issue tagging setup, see our step-by-step aiR for Review guide.
What is Relativity aiR’s Custom Analyses?
Relativity aiR is the platform’s suite of generative AI tools built into the review workflow, spanning several purpose-built project types: aiR for Review (responsive and issue coding), aiR for Privilege, aiR Assist (general insight into documents), and Custom Analyses. Most of these tools are built around a defined ediscovery task with a set output, like flagging responsive documents or drafting privilege log descriptions.
Custom Analyses is the outlier. It’s the aiR suite tool that allows you to think outside the box. It is built for the questions your matter raises that don’t map to a standard ediscovery category. Instead of working within a cookie-cutter project type like responsive or hot, you write the criteria yourself in plain language and apply it to your document population. That flexibility is what expands the potential use cases well beyond traditional ediscovery analyses.
How Does Custom Analysis Work in Practice?
At its core, Custom Analyses lets legal teams define their own extraction, classification, and assessment criteria using plain-language instructions. There’s no prompt engineering, no query building, and no separate model training required.
If your document review protocol requires identification of documents hitting a specific indemnification threshold, or classifying internal communications by whether they discuss a named corporate policy, you can now create the logic to tag the documents directly in the aiR for Review interface and execute it across an entire document population.
In practice, this opens up use use cases that lived outside prior automated review capability, such as:
- Contract term extraction in commercial disputes pulling specific clauses, dates, and obligations across thousands of unorganized agreements
- Highly specific privilege identification applying complex, multi-variable privilege criteria consistently across sensitive internal investigations
- Regulatory compliance mapping matching unstructured documents against precise statutory or internal policy language
- Mixed-file-type triage analyzing images and non-standard file formats to surface information traditional text-only indexers miss
These additional uses are possible because Relativity’s Custom Analyses is not limited to only text files. It is able to read image based files alongside your text based files within the same process. The current file types that are supported are:
| Text File Types | Image File Types | Restricted File Types |
|---|---|---|
| .doc | .jpeg | .exe |
| .docx | .png | .com |
| .txt | .gif | .pif |
| .bat | ||
| .msg | .scr | |
| .eml | ||
| .xls | ||
| .xlsv | ||
| .csv | ||
| .html |
If your team doesn’t have bandwidth for setup and validation, Percipient runs Custom Analysis projects end to end.
Real World Use Cases for Relativity aiR Custom Analyses
Custom Analyses shows its value most clearly in matters that call for something a standard coding set ups were not built to do. For example:
- Contract provision extraction.
- Pulling a specific provision, like governing law, out of a large batch of contracts. This has clear application in M&A due diligence, where counsel needs a fast, consistent read across hundreds of agreements.
- Job site safety review.
- In a construction matter, running Custom Analyses across job site photos to flag safety issues, such as missing fall protection.
- Handwritten notes.
- Analyzing images of handwritten notes to build a searchable set out of material that would otherwise sit outside a standard review.
- Metadata reconstruction.
- Extracting email headers, to, from, date, subject, from an opposing production of PDFed emails that came over without metadata.
Real World Custom Analyses Use Cases Utilized by Percipient
Making the unsearchable searchable.
The handwritten notes example offers an even broader benefit. Custom Analyses provides an easier way to search in data that isn’t typically searchable, like images. Instead of manually reviewing every photo or scanned page, or relying only on extracted OCR text, you can target your review using aiR-generated descriptions of the documents.
We have used this in large-scale construction document reviews to identify specific equipment, work areas, or conditions at issue, without manually sorting through thousands of unlabeled project photos.
Targeted review through creative prompting.
Because you control the instruction language, your input dictates how aiR structures its reply. The trick is prompting aiR to structure its answer with a consistent, scannable marker at the front, so you can filter or search on that marker across the entire document population instead of reading every response. Proper planning during this process will save your team a lot of time and headache while also making it easy to search your data set.
In a construction dispute, this works well on image files: prompt aiR to lead each photo’s description with a flag for whether a disputed piece of equipment is present, so you can isolate the relevant images out of a large photo dump without opening each one.
The common thread is that you’re not just asking aiR a question, you’re designing the shape of its answer so the output itself becomes searchable.
Building category tags into classification output.
Classification alone gets you documents sorted into buckets, but sorted isn’t the same as searchable. The fix is the same principle as before: prompt aiR to lead its classification with a short, consistent tag rather than burying the category inside a longer descriptive answer.
In an FDA approval matter, you could prompt aiR to determine which stage of the approval process each document relates to, preclinical research, investigational new drug application, clinical trials, and open its answer with a short code for that stage. Once that tag sits at the front of every response, pulling every document tied to the IND phase becomes a search for the tag instead of a manual read through the whole bucket.
Considering how AI-assisted review capabilities like Relativity’s aiR for Review and Custom Analyses fit into your next matter?
Talk to our team about building a defensible review strategy around it.
Why Plain-Language Prompting Increases Document Review Defensibility
The core strength of aiR Custom Analyses isn’t the fact that it utilizes an underlying large language model, it’s the shift in control back to the user. Before the availability of generative AI, technology-assisted review (TAR) and predictive coding often felt like black-box exercises. Users loaded a seed set of documents, let the algorithm calculate relevance vectors, and hoped the resultant statistical analysis held up under the scrutiny of opposing counsel.
Custom Analyses is built on iterative validation instead. Because the instructions are written in plain English, review teams can run the instructions on a small sample set, review the output, adjust the instruction language, and validate results before scaling up.
This test-adjust-verify loop is exactly what makes an AI-assisted workflow defensible. You’re not defending a black-box algorithm, you’re defending your own written instructions and the validation record you built to test them.
When to and When Not To Use Custom Analyses
Custom Analyses isn’t the right fit for every question. It works best when the analysis needed does not match an existing aiR feature type. When it does, use the tool built for that task instead:
- Use aiR Assist when you need general insight into the documents rather than a structured extraction, for example, understanding what role Mr. Smith played in Mr. Y’s termination from XYZ Corp.
- Use aiR for Privilege when you need custom descriptions generated for use in a privilege log.
- Use Custom Analyses when the question you want to answer is specific to your matter and doesn’t fit either of those boxes.
Why Native Integration Inside Relativity Matters
Moving data between document review phases is often when the complexity of e-discovery projects increases, costs escalate, and security risks rise. One of the most significant aspects of Relativity aiR Custom Analysis is that it runs entirely native to the platform.
Historically, specialized AI classification meant exporting data to a third-party analytics tool and importing coded metadata back into the e-discovery platform of choice. Such a workflow can introduce formatting compatibility errors, and fragments the record of how review decisions were reached. Keeping the entire analysis inside the environment where the data already lives keeps the process clean and auditable from collection through production.
Not sure whether Custom Analysis fits your matter, or want Percipient to build and validate the prompts for you?
Where Does Custom Analyses Fit Into a Modern Document Review Strategy?
Tools like Custom Analyses are powerful, but they’re most effective in the hands of teams who know how to validate AI output, structure a defensible workflow around it, and know when a matter calls for more than an off-the-shelf approach. It is in projects like these where experienced managed review and e-discovery partners add value, not just running the tool, but building the review protocol around it and standing behind the results.
If you’re evaluating how custom, AI-assisted classification could fit into an upcoming matter, it’s worth a conversation before the review starts, not after.
The Bottom Line
The era of rigid, one-size-fits-all legal technology is ending. The ability to encode your team’s specific review methodology directly into native AI tools means your legal judgment can finally scale with the data. But technology is never a substitute for a sound process; the teams that succeed will be the ones who master the collaboration between human validation and machine efficiency, not the ones who hand over the keys.
Frequently Asked Questions
What is Relativity Custom Analyses? Custom Analyses is a project type within Relativity aiR for Review that lets legal teams write their own extraction and classification criteria in plain language, rather than using a pre-built ediscovery category like responsive or issue coding. It applies that logic across a document population, including image-based files.
How is Custom Analyses different from aiR for Review or aiR for Privilege? aiR for Review and aiR for Privilege are built around defined ediscovery tasks, like responsive coding or privilege log descriptions. Custom Analyses is open-ended: you define the extraction or classification criteria yourself, which makes it suited to matter-specific questions that don’t fit those standard categories.
Can Custom Analyses read images and handwritten documents? Yes. Custom Analyses can process image file types (such as JPEG, PNG, and GIF) alongside text files in the same run. This includes analyzing photos for visual issues, like job site safety conditions, or reading handwritten notes to create a searchable output.
When should you use aiR Assist instead of Custom Analyses? Use aiR Assist when you need general insight into a document set rather than a structured, repeatable extraction, for example, understanding a specific individual’s role in an event described across the documents.
Is Custom Analyses output defensible? Yes, when it’s validated properly. Because the criteria are written in plain English, teams can test instructions on a sample set, review the output, refine the language, and document that validation process before scaling across the full population. That test-adjust-verify record is what supports defensibility, not the underlying model alone.
The post Relativity Custom Analyses: AI Document Review, Your Rules appeared first on Percipient – Legal Services Powered by Technology.