How to speed up your bulk tasks

When you run Claira across a large set of documents, each document is handled one at a time behind the scenes. A few small choices on your end can make the whole run noticeably faster — without changing the quality of what you get back. This guide covers five simple principles and the steps to apply them.

1. Keep your prompt short and focused

Your prompt runs against every document in the set. The shorter and clearer it is, the less work happens on each document — and that small saving multiplies across hundreds or thousands of documents.

What to do:

• Ask one clear question or give one clear instruction per run.
• Cut background, examples, and "nice to have" requests you do not strictly need.
• Avoid asking for a long write-up when a short answer or a simple label is enough.

For help shaping a prompt, see Prompting.

2. Match the model to the task

Claira's model picker offers a faster default and a stronger reasoning option. Sorting documents into obvious categories, flagging a clear feature, or answering a simple yes-or-no rarely needs the stronger model. Asking Claira to think harder than the task requires slows every document down.

What to do:

• Use Gemini Flash (the default) for straightforward sorting, labelling, and yes/no questions.
• Switch to Gemini Pro 2.5 only when the task involves genuine judgment, ambiguity, or borderline calls.
• If your results already look good on Gemini Flash, leave it on Gemini Flash.

See Model selection for what each model is tuned for.

3. Trim your document set before you start

The more documents in a run — and the larger each one is — the longer it takes to finish. The fastest way to shave time off a bulk task is to remove documents that do not need to be scanned in the first place.

What to do:

• Narrow your Nuix Discover selection to the documents that actually matter before launching the bulk task.
• Remove obvious clutter such as duplicates and irrelevant attachments from your selection.
• For early prompt testing, run on a small sample first with single review (free on every plan) before launching a large bulk task.

4. Run large jobs at off-peak times

Claira shares model-processing capacity across everyone using it. Single-document scans are prioritised so individual reviewers stay responsive, which means a large bulk task running during busy hours can take longer than the same task running on a quiet evening or weekend. Running when fewer people are active means more capacity is free for your job.

What to do:

• Start or schedule very large runs outside normal business hours where possible.
• Early mornings, evenings, and weekends are usually the quietest.
• If you need results during the day, split a very large job into a couple of smaller runs so you can react to results as they land.

5. Use the scan speed slider deliberately

Live bulk scans — the runs that stay in your browser tab — show a Scan speed slider in the progress area, with five positions from Slowest to Fastest. Every run starts at Balanced, which holds a little speed in reserve to keep resource-availability errors rare. When conditions are good, that reserve is yours to spend.

What to do:

• If the run is going cleanly — the failed counter staying at or near zero — and you want results sooner, push the slider toward Fastest.
• If documents start failing with resource-availability errors, pull the slider down a notch or two; the errors fade quickly at slower speeds. (Those failed documents are refunded automatically and can be scanned again from the Bulk Tasks list.)
• You can adjust at any time mid-run; changes take effect right away and the estimated duration updates to match.

See Adjusting the scan speed for the full details.

Putting it all together

Before you launch a bulk run, take a moment to check that your prompt is tight, your model choice matches the task, your document set is trimmed to what matters, and your timing avoids the busiest hours. Then, once the run is under way and looking clean, feel free to turn the scan speed up. A few seconds of setup can save a lot of waiting on a large job.