What SmartResearch Can and Cannot Do

Who this is for. Everyone, before their first session. See Roles and Access Reference.

This article sets expectations before you start, so the first session does not disappoint. Most early frustration with SmartResearch comes from one of two beliefs: that it can reach data nobody has connected, or that it can change something. Neither is true, and knowing that in advance makes the tool far more useful.

Read the two tables below together. The first is what to bring it; the second is what to take elsewhere.

Capabilities

Each row is a capability you can count on, with a note on how it behaves in practice.

Table 1. Capabilities

It can

Notes

Answer questions about your finance data in plain language

No query syntax, no report builder. Ordinary business phrasing works.

Compute governed metrics

Where a metric is defined in Metrics Studio, the answer uses that definition and cites it, so two people asking the same question get the same basis.

Show its working

A step-by-step reasoning trace with timings, so you can see which sources were consulted and in what order.

Return the underlying rows

With paging and CSV export, so you can check the detail behind a total or take it into a spreadsheet.

Chart a result

Where a visualisation helps (for example, a value plotted against a typical range).

Suggest follow-up questions

Based on what you just asked, which is often the quickest way to narrow an investigation.

Remember context within a chat

Follow-ups are answered against the conversation so far, so you can say “and for last quarter?” without restating the question.

Apply your personal defaults

Entities, time range, fiscal year, formats and response guidance, applied without being restated.

Be taught a repeatable workflow

As a Skill, with steps, defaults and output format, so a recurring analysis runs the same way each time.

Limitations

Knowing these limits before you start prevents most early frustration; each row explains why the limit matters.

Table 2. Limits

It cannot

Why it matters

Change anything

Questions read data. SmartResearch does not post, approve, pay or edit records. Nothing you ask can alter your ERP.

See beyond the selected instance

The SmartFlow instance decides the system of record and mailbox in scope. Data behind a different instance is invisible.

See data a role is not entitled to

Answers respect access, so two people can legitimately get different totals for the same question.

Guarantee correctness

The product states it can make mistakes and asks you to validate responses. Treat a figure that will move money as a lead, not a conclusion.

Fix upstream data

Where a figure is implausible it will usually say so, but the cause sits in the source system and has to be fixed there.

Answer without a connected source

With no instance selected there is nothing in scope, and no phrasing will change that.

Two people getting different totals for the same question is normal, not a bug. The usual causes, in order of frequency: different entity scope in User Preferences, different default time range, a different SmartFlow instance, and role-based row filtering. Check those four before raising a ticket.

What this means in practice

  • Ask freely. Because nothing can be changed, there is no risk in exploring. The worst outcome of a badly phrased question is an unhelpful answer.

  • Check before you commit. For a figure that drives a payment, a write-off or an external report, confirm it in the system of record.

  • Name the metric. Using a governed metric’s name in your question gets you a cited definition instead of an ad hoc calculation.

  • Read the trace when surprised. It usually shows why an answer differs from what you expected.

Related information