Preparing Your Data for Upload - and When to Add Expert Touch

What DataScope actually accepts

Before anything else: DataScope takes CSV, Excel (.xlsx and .xls), and JSON files. That's it - there's no database connector in the upload flow itself, so if your data lives in a warehouse or a production database, the fastest path is exporting it to one of those three formats first. Knowing this up front saves a wasted upload attempt with the wrong file type.

Once a file is uploaded, the project screen reads it and gives you a credit estimate based on the analysis type you've chosen and the file's size and complexity. That estimate is only as good as the file behind it - a messy file doesn't just make the analysis less useful, it can also change what it costs, since size and shape feed directly into the complexity multiplier. A little cleanup before you upload tends to pay for itself.

A short checklist before you upload

Headers on the first row, one row per record. DataScope's analysis assumes a standard tabular shape: column headers in row one, one observation per row after that. Merged cells, multi-row headers, or summary rows mixed into the data (subtotals, grand totals) confuse both the row count and the analysis itself. If your export tool adds a title row above the real headers, strip it before uploading.

Consistent column types. A column meant to hold numbers should hold numbers all the way down - not "1,204" in some rows and "N/A" or "TBD" in others. Mixed types are one of the most common reasons a Descriptive summary comes back less useful than expected: the platform can't average a column that's half text.

Dates in a single, recognizable format. Pick one date format and use it consistently through the whole column. A column that mixes 03/04/2026 and 2026-04-03 is genuinely ambiguous - even to a human reader, let alone an automated pass - and it's the kind of thing worth fixing in five minutes before upload rather than discovering afterward.

Decide what to do with missing values before you upload, not after. Blank cells, zeros used as placeholders, and explicit "N/A" strings all get treated differently. If a blank cell means "not recorded" for you, leaving it blank is usually more accurate than filling it with a zero that will get averaged into your numbers.

Trim the file to what's relevant. You don't need to strip every unrelated column, but a file with forty columns when the analysis only needs eight adds size - and therefore complexity, and therefore cost - without adding value. If you know which columns matter for the question you're asking, a quick trim before upload is worth the two minutes it takes.

Keep file size proportional to the question. Complexity is driven by file size, roughly tracking row and column counts. A smaller, well-scoped extract of the data you actually need for this analysis will usually answer the question just as well as the full export, at a lower complexity tier.

None of this requires special tooling - most of it is a few minutes in whatever spreadsheet software exported the file in the first place. The return is a more accurate analysis and, often, a lower complexity multiplier on the estimate.

What AI analysis gets you - and where it stops

Once a clean file goes in, the AI analysis produces statistical summaries, visualizations, and the pattern or trend detection appropriate to whichever analysis type you picked - descriptive, diagnostic, predictive, or prescriptive. It's fast: minutes, not a scheduling conversation with an analyst. For a lot of questions, especially the descriptive and diagnostic kind, that's the whole job.

Where it's worth pausing is when the output is going to inform a decision with real consequences attached - a forecast that sizes next quarter's staffing, a recommendation that reallocates a meaningful budget, a segmentation that changes who gets targeted for something. AI analysis is a strong first pass on all of these, but it's still a first pass: it doesn't know your industry's specific quirks, and it can't sanity-check its own output against context that isn't in the file.

That's what Expert Touch is for.

What Expert Touch actually is

Expert Touch is a human review of one project, done by a data scientist on DataScope matched to your project by field and track record. It costs 500 credits per project, charged when you request it. The expert checks the AI's analysis rather than starting over from scratch - validating the approach, catching anything that doesn't hold up, and adding judgment the automated pass can't.

A few things worth knowing about how it actually works:

• You're only charged if an expert is available. If no one's available for your project's field, nothing is charged.
• You can cancel before the expert starts and get everything back. Until the review has actually begun, cancelling refunds the full 500 credits - so requesting it and then deciding against it costs nothing.
• It's per project, not automatic. Expert Touch is something you request on a specific analysis, not a tier you upgrade into. Most projects don't need it; the ones that inform a real decision often do.

When it's worth the 500 credits

A useful way to decide: ask what happens if the AI's analysis is subtly wrong and nobody catches it before you act on it. If the answer is "not much" - you're exploring a dataset out of curiosity, or the output confirms something you already suspected - Expert Touch probably isn't necessary. If the answer is "we'd commit a budget, change a staffing plan, or make a claim to someone outside the team based on this," a second set of trained eyes for 500 credits is cheap relative to the size of the decision it's informing.

It's also worth reaching for on projects where the data itself is unusual - an industry-specific dataset, a metric that doesn't behave the way a general-purpose model expects, a result that surprises you. Expert Touch is matched by field specifically because domain-specific quirks are exactly what a generic first pass is most likely to miss.

Putting it together

The cheapest way to get a reliable answer out of DataScope is the same in both halves of this guide: do the small amount of prep that makes the input clean, pick the analysis type that actually matches your question, and reserve the extra 500 credits for Expert Touch on the projects where a mistake would actually cost you something. For most exploratory work, a clean file and the right analysis type is enough on its own.

See the analysis type guide for picking between Descriptive, Diagnostic, Predictive, and Prescriptive, or head to how it works to see the full upload-to-insight flow.