Buy Software, Use No-Code or Build a Custom AI Workflow? A Small-Business Decision Guide
Buy the simplest solution that can deliver your accepted business outcome reliably. An existing software feature, a no-code workflow and custom development solve different coordination problems. The right choice is not the one with the most agents or the fewest lines of code. It is the option whose constraints, total cost and operating responsibilities fit your process.
Start with a workflow specification, not a platform shortlist
Describe the trigger, authorised input, required output and person who accepts it. Name the exceptions, volume, response window and consequences of a wrong action. For example, drafting a customer follow-up is different from sending it; proposing a record update is different from changing a live CRM. Those boundaries determine the necessary controls before anyone selects a tool.
Ask whether your current application already supports the task. A feature that staff can configure and own may be preferable to another service, even if its demonstration looks less impressive. Check the actual account's plan, permissions and limitations. Do not assume a product supports a particular connector, volume or privacy arrangement because another customer uses it.
Compare approaches on the same outcome
| Option | Good fit | What must be checked |
|---|---|---|
| DECISION | Choose the simplest adequate route | Acceptance, full cost and named operator |
| Existing software | A standard task already supported | Plan limits, data handling and export |
| No-code orchestration | Stable steps between supported systems | Permissions, retries, duplicate handling and change ownership |
| Custom workflow | Specific rules or interfaces that other options cannot meet | Testing, deployment, maintenance and handover |
No-code does not mean no engineering responsibility. Someone still owns access permissions, retries, duplicate prevention, failed jobs and changes to connected services. Custom code does not remove those duties either. Its benefit is control over requirements that justify the extra implementation and maintenance work, not a magical escape from ongoing operations.
Compare total cost rather than the subscription line
Include setup, data cleanup, integration, workflow usage, model charges, monitoring, review and support. Add the cost of a change when the process or a connected service evolves. Compare equivalent volumes and acceptance criteria. A cheap monthly plan can still be an expensive solution if staff repeatedly reconcile missing records or repair failed actions by hand.
Write down the exit route before buying: which accounts belong to the customer, which files and configurations are delivered, how records can be exported and what another supplier needs to continue the work. Access to a hosted product is not automatically ownership of its source code. An understandable handover and customer-controlled credentials reduce avoidable dependency without promising that every component is transferable.
Test one vertical slice before making the larger commitment
Select a representative trigger, one source system and one accepted output. Run ordinary, incomplete, duplicate and failed inputs through the same slice. Begin with read-only or draft behaviour where possible. Permit writes only after the approval rule, error path and recovery method have been tested. A polished dashboard cannot compensate for unreliable data underneath it.
- Keep: the solution meets acceptance criteria and an identified person can operate it at an acceptable total cost.
- Change: the benefit is plausible, but a narrow integration or process issue prevents reliable delivery.
- Stop: a simpler alternative performs better, important errors remain uncontrolled or nobody owns operation.
Include the operating model in the purchase
For ongoing support, define what is monitored, when alerts are reviewed, who handles incidents and which changes are included. Availability commitments and response windows must be explicitly agreed. Maintenance is not an unlimited promise to redesign the workflow whenever business policy changes; separate routine upkeep from new functionality and usage expenses.
The free vendor-brief template makes these choices comparable before sales calls. LEMAIT offers a one-process diagnostic at EUR 350–600 and a bounded pilot from EUR 960; broader implementation starts at EUR 3,000. Each agreed milestone is fully prepaid. A diagnostic may recommend using your existing software instead of commissioning development. Ongoing support is quoted only after scope, access and responsibilities are known.
Source and disclosure: France Num's adoption guide provides useful context for progressing from existing tools to integrations and dedicated projects: https://www.francenum.gouv.fr/guides-et-conseils/intelligence-artificielle/comprendre-et-adopter-lia/lintelligence-artificielle-6 . The decision matrix and operating questions are LEMAIT's own analysis, prepared with AI assistance and editorial review. This is not an independent benchmark of vendors, a live tariff comparison or a claim of prior deployments on every named platform.
Prepared with AI assistance and checked against the linked public sources by LEMAIT. Scenarios are illustrations, not customer results. No ranking, ROI or legal-compliance guarantee.