Skip to content

Excel, Notion, or a custom app: which should you choose?

Excel and Notion can both be very useful. The problem starts when they become a database, a permissions system, and a coordination tool without having been designed for those roles.

05 chapters

Recognise the real switch signals

Row count is not the best indicator. Useful signals are competing copies, formulas nobody dares to change, double entry, the absence of history, and access shared too widely. In Notion, the equivalent is pages that contradict one another, fragile linked databases, or no clear rule about which information is authoritative.

The need also appears when a business rule depends on one person. If their explanation is required to understand a cell colour, a tab name, or a page status, the process is no longer transferable. Custom software becomes relevant when it secures that knowledge and removes a measurable daily cost.

Business diagnosis

Three symptoms show that the file has become a system

Spreadsheet size matters less than its real role. A switch becomes relevant when it carries data, decisions, and coordination at once.

  1. Symptom 01Data

    Competing truths

    Copies, implicit formats, and re-entry make it unclear which information is authoritative.

  2. Symptom 02Decision

    Hidden rules

    A formula, colour, or convention understood by one person triggers the work.

  3. Symptom 03Flow

    Invisible approvals

    Roles, status changes, and responsibilities live in messages rather than the process.

France Num recommends choosing a tool by coverage, integration, and daily use, not only by its feature list.

Official sourceFrance Num — Business management softwareAccessed 30 August 2026
  • Several files claim to be the current version.
  • A formula error can change an operational decision.
  • The same data is re-entered in several tools.
  • Roles and validations cannot be applied clearly.

Build or buy: start from the business gap

An existing product remains the right choice when it covers the central process, integrates with tools already in place, and keeps an acceptable cost at the scale of the team. Building what already exists without a useful difference is wasted spend.

Custom development is justified when the rule that distinguishes the activity cannot be expressed, when licences grow with no relation to value, or when double entry persists despite the tool. A web application is often the simplest format for a shared desk; SaaS adds isolation across organisations and billing.

Treat data takeover as a project

Migrating does not mean importing every cell. Duplicates, implicit formats, invalid values, and the mapping between the old vocabulary and the new model have to be identified.

Takeover is repeated dry before cutover. Totals, relations, and sensitive samples are checked, then a rollback procedure is written. The new product can be correct and still fail if its starting data is not.

Data takeover

Import comes after data quality work

Useful data passes through several states before feeding the new tool. Every transformation must remain explainable and verifiable.

  1. 01

    Collect

    Gather the files, exports, and reference sets that are actually used.

  2. 02

    Structure

    Name fields, identifiers, relations, and expected formats.

  3. 03

    Transform

    Handle duplicates, invalid values, and old conventions.

  4. 04

    Validate

    Compare volumes, relations, and a business sample before cutover.

The new interface cannot compensate for a contradictory source or an undocumented transformation.

Official sourceFrance Num — Run a small business with dataAccessed 30 August 2026
  • Keep a frozen, readable copy of the source.
  • Document transformations and rejects.
  • Compare volumes and relations after import.
  • Have a sample validated by the people who use the data.

Put rights, history, and backups in the core

A shared file rarely offers the fine-grained rights a business process needs. The new tool must distinguish read, edit, validation, and administration, then log the actions that commit.

Backups only count if a restore has been tested. Infrastructure and security therefore do not come after the screens: they are part of the first usable perimeter.

Ship by use and organise the handoff

The first lot must remove a complete friction, not reproduce half of the spreadsheet. A team adopts a journey that actually solves a problem more readily than a large product still incomplete.

The people concerned test early on their real cases. Gaps are recorded, useful old habits are kept, and a clear date ends dual systems. Stellary illustrates a product where roles, data, agents, and supervision share the same state. An estimate then lets the perimeter be set against a budget.

Common questions

The remaining decisions

From how many users should Excel be left?

There is no universal threshold. Two people can already stall on competing versions, while a larger team can keep a simple spreadsheet. Risk and the cost of workarounds matter more than the number of accounts.

Must every spreadsheet function be reproduced?

No. Functions that are actually used are kept; conventions, duplicates, and calculations that no longer help are dropped. Migration is the moment to simplify the process, not to fossilise every column.

Can Excel be kept for exports?

Yes. Custom software can produce tabular exports for analysis or exchange. The spreadsheet stops being the source of truth without disappearing as an occasional working format.

Put the guide against a real project.

A few lines are enough: context, the main constraint, and the expected result.

Write to the studio