Pilardin Adds Google Drive and Network Excel Connectors for Enterprise RAG

Google Drive and network Excel files become queryable in Pilardin

Pilardin has added Google Drive and network-hosted Excel workbooks to its connector lineup, extending the platform's enterprise RAG layer to the two places where a large share of day-to-day business knowledge actually lives. Administrators can scope specific shared drives, folders, or file-share directories as sources, and users can then ask questions in natural language against those files from the same workspace where they already query databases, ERP, CRM, and document libraries. The change is narrower than a data-platform migration, but it closes a gap that has frustrated internal AI programs for years: the answer usually exists, and it usually exists in a spreadsheet that nobody indexed.

The practical effect is that spreadsheets stop being only a reporting endpoint and become a queryable layer. A finance analyst can ask about a variance in a budget workbook; a procurement lead can ask which suppliers in a tracking file breached agreed lead times; an operations manager can ask which cost centres in a regional model exceeded plan. Because the new sources sit alongside live system connectors, one question can pull a figure from a governed database and another from a workbook that was never modelled, without the user needing to know which source answered which part.

Why shared drives and Excel are the hardest sources to make trustworthy

Excel is technically structured but practically wild. A typical departmental workbook carries merged header cells, several tabs representing different periods or scenarios, hidden helper columns, comments that change the meaning of a number, and at least three circulating versions with names like Budget_v6_final_copy. Indexing that file as a wall of text produces answers that look precise and are wrong. Making it useful requires table detection, header inference, sheet-level scoping, and enough lineage to tell a user which tab and which range an answer came from.

Google Drive adds a second problem: permissions that move. Sharing in Drive is per-file and per-folder, and it changes constantly as people rename, move, and re-share files. A network share built on directory permissions has the same characteristic in a different form. If retrieval honours permission only at indexing time, a user can receive a summary of a file they would not be allowed to open today, which is a compliance failure rather than a feature. Permission-aware retrieval has to evaluate access against the requesting user at query time, the same standard Pilardin applies to its SharePoint and database connectors.

• Multi-tab workbooks where each sheet represents a different period, entity, or scenario
• Merged cells and multi-row headers that break naive table parsing
• Duplicate and stale copies competing to be treated as the source of truth
• Folder-level sharing that changes after the index was first built

What changes for finance, operations, and shared-services teams

The teams that benefit first are the ones that never received a clean data model. Finance and FP&A groups run planning in workbooks because the model changes faster than a warehouse can absorb it. Operations teams track exceptions in sheets because the process is local and the volume does not justify a system. Shared-services and project teams maintain registers that are effectively hand-built databases. Natural-language questioning of those files does not replace the workbook; it removes the manual scanning that turns a simple question into twenty minutes of filtering and cross-checking.

There is a real ceiling worth stating honestly. A spreadsheet is only as reliable as the person maintaining it, and connecting one to an AI workspace does not validate its contents. The useful mental model is that Pilardin makes the workbook answerable, not authoritative. Where a number drives a decision, the response should be traceable to a named file, sheet, and range so a human can verify it. Where the workbook duplicates a system of record, the system of record should win, and the answer should make that hierarchy visible rather than blending the two into an average.

• Queries reach the long tail of files that never made it into a BI model
• Answers carry source references so a reviewer can open the exact file and range
• Spreadsheet assumptions and live system figures can appear in one response
• The workbook stays the working surface; governance moves to scope access and lineage

Next steps: scoping, residency, and rollout discipline

Rollouts tend to fail on scope rather than technology. A workable sequence is to pick one or two high-value folders, such as a planning directory or a project register, connect them, verify that permissions behave the way the business expects, and only then expand. Connecting an entire corporate Drive on day one mostly guarantees noisy retrieval and a painful compliance review. Residency also matters here: organisations with data-location requirements can run on the instance that matches their policy, with pilardin.com serving international deployments and pilardin.ir serving Iran, which is often the deciding constraint for regulated teams.

Two operational questions deserve an answer before go-live. First, freshness: is the workbook synced on a schedule, and does the answer surface the last refresh so a user can judge currency? A stale figure presented confidently is worse than no answer at all. Second, permission drift: who owns the review when a folder's sharing rules change, and does the platform re-evaluate access continuously rather than once at index time? Teams evaluating connector scope and packaging can review the options on the pricing page (/pricing), and a pilot can be started from /signup.

What to watch

The most important thing to watch is not whether the connector works but whether people trust it. Users who know a spreadsheet is messy will discount answers drawn from it unless the answer shows its work: file name, sheet, cell range, and last refresh. Lineage will shape adoption more than retrieval quality will, and it remains the area where most platforms are still thin.

Second, watch how organisations handle the overlap between workbooks and systems of record. As more files become queryable, the temptation is to treat every sheet as authoritative. The mature pattern is a hierarchy: systems of record for regulated numbers, workbooks for context and planning assumptions, and a visible note when the two disagree. Teams that establish that hierarchy early will get real value from Google Drive and network Excel integration; teams that skip it will simply automate the distribution of stale figures with more confidence than before.

FAQ

Does connecting Google Drive give Pilardin access to every file in the drive?

No. The connector is scoped: an administrator selects the shared drives, folders, or directories that should be indexed, and retrieval evaluates access against the signed-in user at query time. In practice, a user asking a question receives information from files they could already open in Drive, and not from restricted files they could not. Scope should still be reviewed periodically, because folder structures and sharing rules change over time and yesterday's sensible scope can become today's over-exposure.

How current are answers drawn from network Excel files?

Freshness depends on how the sync is configured for that source. Scheduled refreshes suit planning directories that change on a known cadence, while on-demand syncs suit files edited ad hoc. Regardless of the setting, an answer should state when the source was last indexed so a user can judge whether the figure is current enough for the decision in front of them. If a workbook underpins a live decision, refreshing before the query is the safer default.

Can one question combine a spreadsheet with ERP or CRM data from a live system?

Yes. Because Google Drive and network Excel are connectors inside the same workspace as ERP, CRM, and database sources, a single question can draw on both. A question about regional spending, for example, can combine a figure from a financial system with a planning assumption held in a workbook, with the response citing each source separately. That combination is usually where the value appears: it answers questions that previously required someone to export, paste, and reconcile by hand before anyone could act.