Automation, AI or
custom software?

1. Check what your existing software can do

Write down the task, who performs it and where it gets stuck. Review the features and licences you already have. A form, approval setting or built-in integration may solve the problem with less ongoing work than a new application.

2. Use rule-based automation for predictable steps

When a condition and the required action can be defined precisely, an integration may be enough. For example, a validated enquiry can become a record in another system. Plan for missing fields, duplicate requests, expired permissions and unavailable services.

3. Consider AI when language or documents vary

AI can assist with extracting information, finding relevant material and drafting text. It needs representative examples, a way to assess quality and clear review points. Measure correction time and errors as well as the time it appears to save. Do not assume a confident answer is a correct one.

4. Build custom software when the workflow needs its own application

A customer portal, internal work queue or specialist application may need controls and behaviour that existing tools cannot provide. Compare the cost of building and operating it with the cost of configuring or integrating an available product. Ownership and maintenance matter after launch.

What to prepare for a useful estimate

• The users, their working hours and the one task the first release must support.
• The systems involved, their licences and whether supported APIs are available.
• Data sensitivity, access requirements and any hosting-location constraints.
• Expected volumes, exceptions and recovery needs.
• Acceptance criteria and the people available to review the work.

Please describe these in general terms in an initial enquiry. Credentials and confidential sample documents should only be shared later through an agreed channel.

How to compare project cost

Compare the same scope across proposals. Separate discovery, implementation, migration, testing, training and support. Include recurring hosting, integration subscriptions and model usage. A low build estimate may omit the work needed to operate the system.

Xidents does not publish a universal project price or delivery period because the dependencies vary. A useful proposal should explain assumptions, exclusions, acceptance criteria and what happens if the scope changes.

Decide what success will mean before building

Record a baseline for the current process: completed tasks, handling time, errors, rework and time waiting for a decision. Repeat the measurement after a trial using comparable work. Include time spent reviewing automated output. Improvement should be demonstrated, not assumed.

Example: scope an enquiry work queue

This is a planning example, not a customer deployment. Start with one enquiry form, a small staff review queue and one supported destination system. A first release might validate required fields, retain the original message, detect a repeated submission and let a person approve the next step.

• Discovery: map the current handoff, check API access and agree who can view or change an enquiry.
• Acceptance: a valid enquiry is recorded once; an invalid one shows a useful error; a failed transfer remains visible for recovery; unauthorised users cannot view records.
• Optional AI trial: suggest a category and summary for review. Include ambiguous and misleading messages; measure incorrect suggestions and correction time.
• Handover: document monitoring, retry handling, access ownership and the costs of operating the queue.

Choose representative volumes and time targets with your team. A small demonstration cannot establish production accuracy, capacity or a return on investment.

A small example from our own website

Our contact-form implementation uses a rule-based backend to validate and record enquiries. It shows why a reliable integration can be the right first step even for a business that also offers AI services.