Insight / Software development
Published
Build, integrate, script, replace or buy: choosing the right software work
Audit the work and the systems you already have, then compare five routes to a useful software solution.
When someone asks for a new application, the useful first question is what they need to accomplish. They may be copying information between two tools, waiting for an approval, working around a missing feature or paying for a system they barely use. Each problem suggests a different investment. The choice begins with an audit of the work, the people doing it and the software already in place.
Audit the work before choosing a solution
Follow one important task from start to finish. Who starts it? Where does the information come from? Which decisions need a person, and which steps repeat the same rules every time? Note where delays, errors and duplicate entry occur. Talk to the people who use and maintain the current process; the written procedure may miss the workaround that keeps it running.
Inventory existing subscriptions and internal tools alongside that workflow. Record what each one does, what it costs over a consistent period, who can administer it, how data can be exported and which integrations are already available. Check access controls, contractual commitments and the skills the team has to support a change. A system can look replaceable on a feature list while carrying a critical reporting or compliance task in practice.
Then identify the strategic part of the work. If a workflow is how the business delivers a distinctive service, it may justify more ownership and control. If it is a common supporting task, a well-supported existing product may be the better foundation. Neither label decides the answer by itself; it tells you where to investigate first.
Five routes from the same audit
Buy when a product already handles the task well enough and the team can adopt its way of working. Compare the full subscription, setup, training, support and exit costs. Test the limits that affect your actual workflow, including permissions and data export, before committing.
Integrate when the right systems exist but people must carry information between them. Decide which system owns each record, then connect the handoff through supported interfaces. Design for missing fields, duplicate events and failed transfers so staff can see what needs attention.
Script when a narrow, repeatable task follows stable rules. A small automation might prepare a report or check that a file arrived. Give it an owner, a log and a way to recover from errors. A script that silently changes important records can create more work than it saves.
Build when a specific workflow is central to the business or existing products cannot meet a verified requirement. Start with the smallest end-to-end version that proves the value of owning it. Include the less visible work in the estimate: authentication, permissions, accessibility, monitoring, support and maintenance.
Replace when the current system’s constraints or operating burden justify moving its responsibilities elsewhere. Map every dependency before a cutover, migrate a representative set of data, and plan how to check the results and recover if something goes wrong. Replacement may combine a purchased service, an integration and a small amount of custom software.
Keeping the current arrangement for now is also a decision. Record the reason and a trigger for revisiting it, such as a change in volume, cost or reliability.
Compare costs against a real first version
Use the same time horizon for each route. Start with what the business already pays in subscriptions and staff time. Add the proposed setup or development effort, recurring fees, hosting, maintenance, training and migration. Include the cost of continuing a manual step if the first version leaves it in place.
Estimate time savings from the minimum viable version you would actually release, rather than from every feature on a wish list. Observe how long the current task takes, how often it occurs and who performs it. Run a small trial where possible, then adjust the estimate for exceptions and review work. This makes the comparison useful without pretending that an early estimate is a guaranteed return.
For example, if staff copy website enquiries into a customer system, first test whether an existing integration can create a draft record for review. Measure whether that removes retyping while preserving the human decision about how to respond. A new customer platform would solve a much larger problem; the audit should show whether that larger problem exists.
Choose an architecture you can live with
Whichever route wins, document where the source of truth sits, who can change data, what happens when a connected service fails and how to leave the solution later. These choices shape security, ownership and maintenance long after launch. They also make it easier to add another workflow without rebuilding the first one.
Move in stages: agree on one outcome, test one representative workflow, review it with the people who use it, and expand only when the evidence supports the next step. Our software development process explains how we turn that work into reviewable releases; the parts of an application helps identify the responsibilities a custom build must carry.