Raghav Mittal
Menu
Approach →
Services
Shopify CRO Find and fix the friction between a visit and a completed checkout: product pages, collections, cart, checkout, speed, trust and tracking. Subscriptions Launch, fix or migrate Shopify subscriptions: offer design, app choice, customer portal, cancellation flows, failed-payment recovery and retention reporting. Shopify & Web Build or clean up websites that are fast, trackable, SEO-ready, and designed around how buyers actually decide. Shopify Automation Connect Shopify orders, customers, inventory, fulfilment, support, subscriptions, and reporting so your team spends less time moving the same information. SEO & UX Find the crawl, speed, mobile, UX, and conversion issues that stop good pages from ranking or turning traffic into leads. AI Workflows Use AI practically inside business workflows: Excel analysis, PPT reporting, SOPs, dashboards, content, and team documentation. Automation Replace manual follow-ups, spreadsheet relays, and status chasing with trigger-based systems that move work automatically. CRM Systems Turn your CRM from a contact database into an operating system for pipeline, ownership, follow-up, and reporting. Lead Systems Capture, qualify, assign, and follow up with leads across forms, CRM, WhatsApp, email, and sales teams. GST + Tally Connect the work between sales, invoices, GST review, Tally, reconciliation, and management reporting so finance operations have fewer blind spots. TallyPrime Integration Connect approved orders, invoices, inventory and customer records to TallyPrime with maintained mappings, confirmed results and reconciliation. Tally TDL Development Adapt the work inside TallyPrime with scoped TDL fields, validations, reports and document layouts, tested against your installed release. Shopify + Tally Connect Shopify orders, payments and refunds to Tally with clear stock ownership, approved accounting mappings and daily reconciliation. Distribution Software Connect buyer catalogues, tier pricing, purchase orders, warehouse tasks and stock movements in one distribution application. How we workScope, review and first-milestone payment View all servicesBrowse the complete delivery bench →
Blueprints → Work → Blog → Free Audit → Book a strategy call
AI Agents· Nov 20, 2026· 4 min read

AI Spreadsheet Cleanup: Preserve Product Identifiers

Review AI spreadsheet cleanup without losing leading zeros, product identities or source values. Includes a reconciliation example and exception checks.

AI Spreadsheet Cleanup: Preserve Product Identifiers

Article overview

  • An AI spreadsheet-cleanup workflow should preserve product identifiers as identifiers, even when they contain only digits. Define protected columns before changing data types, trimming values or asking an assistant to deduplicate rows.
  • This is especially relevant when a spreadsheet moves between a supplier, an ecommerce catalogue and an operations tool. Each application can apply its own formatting and import assumptions.
  • Identify product codes, external record IDs, postal codes and other reference fields. Keep their original representation where the destination requires it.
  • Microsoft's guidance on leading zeros and large numbers explains why import types matter in Excel.

Putting this workflow into practice? Explore AI Workflow Automation for implementation scope, controls, and next steps.

An AI spreadsheet-cleanup workflow should preserve product identifiers as identifiers, even when they contain only digits. Define protected columns before changing data types, trimming values or asking an assistant to deduplicate rows. A clean-looking workbook can be unusable if a code no longer identifies the same item.

This is especially relevant when a spreadsheet moves between a supplier, an ecommerce catalogue and an operations tool. Each application can apply its own formatting and import assumptions. Test the final exported values, not only what the current cell appears to display.

Mark fields that must not be calculated

Identify product codes, external record IDs, postal codes and other reference fields. Keep their original representation where the destination requires it. A numeric-looking code is not necessarily a quantity, and two similar-looking codes are not necessarily duplicates.

Microsoft's guidance on leading zeros and large numbers explains why import types matter in Excel. Choose the supported text-import controls for the actual version rather than assuming a later display format restores information already lost.

Write a transformation brief the reviewer can check

Specify which columns can be cleaned, which can be standardised and which must be preserved exactly. For example, removing accidental spaces from a descriptive category may be acceptable while changing an external item code requires a controlled mapping.

Tell the assistant how to represent uncertainty. If a value is unreadable or a code conflicts with the master, return a review item with its source row. Do not ask the model to infer a missing digit from nearby records and silently replace it.

Keep the original and transformation evidence

Work from an approved copy and retain the source reference. Store the transformation instructions, output version and review decisions together. This makes a later failed import investigable without requiring someone to remember which prompt was used.

Restrict the data supplied to the assistant to what the task requires and what the organisation permits it to process. Product-data work can still contain private commercial fields. Sharing permission should be resolved before uploading a workbook, not assumed from the ability to open it.

Worked example of two codes becoming one

Illustrative scenario: a source catalogue contains codes 00123 and 123 for different products. A cleanup converts both to the number 123 and then treats the second row as a duplicate.

Check Expected result Reason
Source-to-output identifier comparison Both original codes retained They identify different source records
Master lookup Each resolves to its approved item Formatting alone does not establish identity
Duplicate review No automatic merge from numeric equality The loss occurred during transformation

Keep a fixture containing both codes, a long identifier and a value that resembles a date. Reopen the exported file through the intended destination's import path. A screenshot of the workbook is insufficient evidence if the exported value changes.

Validate independently of the model

Compare protected fields exactly, reconcile record counts and check the keys that link to downstream systems. Validate quantities and totals with ordinary deterministic checks where possible. A valid table structure does not prove the contents remain correct.

The AI extraction review guide explains the distinction between well-formed output and verified values. Apply the same principle to catalogue cleanup: model confidence should not substitute for a source comparison.

Prepare the handover and rerun rules

Record which file is approved for import and what should happen when the source is revised. A later run must not silently overwrite a manually approved mapping. Preserve the reason for corrections and use difficult rows as regression examples.

If the cleaned data feeds management reporting, follow the Excel-to-presentation workflow only after the input is reconciled. Explore AI workflow automation, or describe the workbook and destination to scope the transformation and review checks.

Frequently asked questions

Can formatting a column later restore a damaged identifier?

Not reliably. A display format may make a value look similar without recovering the original source information. Compare against the retained source and apply a reviewed correction when the original representation matters.

Should an AI assistant decide which product codes are duplicates?

It can assist a review, but the merge rule must come from the business's identity model. Keep the source references and require evidence before combining records that downstream systems treat as different products.

AI Workflow Automation

I design workflows that use Claude, DeepSeek, OpenAI, spreadsheets, slides, docs, and business data to save real team hours.

Explore the service →
Turn the idea into a working system

One recurring report or document workflow

Describe the report or document, its source files and who approves the final output.

Prefer a conversation? Book a call →
Keep reading
View the full archive →