Skip to content
Interface language: English
Imperium / Signal

Build, buy or keep it manual: three ways to maintain one capability

Compare building, buying and manual work around one task: exceptions, skills, maintenance, changing requirements and exit.

Article language: English

Fine white strands cross a wooden piece with gold-coloured teeth on a workbench.
In this article

A team needs to reserve equipment for workshops reliably. Three camps quickly form: “let's write our own application”, “let's buy one” and “we have a spreadsheet”. Each defends a solution before anyone describes what must work on Monday morning. The spreadsheet does not object. It is usually the last to learn that it has been promoted to a critical system.

Building, buying and manual work are three ways to maintain the capability to perform a task. Compare them over the same scope: outcome, exception handling, maintenance and changing requirements. An application's feature count does not answer whether the company will be able to use it in six months.

For equipment reservations, first check date conflicts, a set change before a workshop and handoff to a substitute. If a manual workflow handles these at the current volume, it remains a real option. If a tool is to take some work, it must show results on the same cases. Building also needs a specific owner for later fixes and maintenance.

Compare the same work through to its end

In a fictional workshop company, a team must allocate equipment, confirm a date, prepare handover and record its return. A damaged component and a booking changed one day before the event are also possible. Comparing only date entry chooses a way to fill a calendar, not a way to manage equipment.

Record the outcome in task language: the right set is available in the agreed place and time, and the next team knows when it returns to use. List necessary conditions, such as recognising conflicts, ownership of changes and preserving history. Only then assess ways of delivering it.

Manual need not mean chaos. It can have a form, consistent register, one responsible person and agreed acceptance. A purchased option should not receive perfect configuration, training and ready data for free. A built option needs more than screens: a way to launch, repair and hand it to another person.

The UK Service Standard point on technology considers the total cost of ownership of technology and the ability to change a choice later. It is a useful reference for thinking about a decision, not a mandatory procedure for a Polish company. Link your own criteria to the actual task.

Also record what the task does not require. In our example the company might need allocation of its own equipment, not a public shop with payments and customer accounts. A broad product may offer those features, but they earn no extra points if they do not help the agreed task. Separating needed scope from available features keeps the comparison from becoming a tour through the whole tool menu.

Prepare identical cases for all three: an ordinary reservation, an unavailable set, a changed date and a return with a fault. Record who performs each step and what they need to decide. Comparison then reveals how work happens instead of rewarding the option with the longest presentation feature list.

  • Who resolves a conflict? In a manual workflow a named person follows a rule. For a purchase, check configuration and exception handling; for a build, design the rule and name its owner.
  • Who maintains a change? In a manual workflow the register and instruction owner does so; for a purchase, the company team works with the provider; for a build, the maintaining team does it.
  • How can an old decision be explained? Manual work needs a retained case record. For a purchase, check history and export; for a build, plan a decision trace and access to it.

These questions open the comparison. Your process may produce different answers. Their purpose is comparable scope, so human work does not disappear from one option merely because another has its own subscription.

The hardest day says more than the first screen

For manual work, test cover. Can another person hand out the right set without calling the spreadsheet owner? If the answer is “only Kasia knows what yellow means”, the present process has a dependency absent from a tool price. Instructions may reduce it before a larger project starts.

For a purchase, run your own difficult case in a trial. Do not evaluate only a provider-prepared demonstration. Check the data, exception and recovery from error. Assessing a provider before purchase (in Polish) helps distinguish a capability claim from evidence for a specific use, particularly when AI is part of the process.

For a build, check who will accept the scope and make later decisions. “We have a developer” does not answer cover, updates or whom to contact after failure. Owning code can be valuable, but does not itself create maintenance capability. That capability needs people, time and a comprehensible handoff.

Government Digital Service purchasing-strategy guidance connects build-or-buy choices to user needs and organisational capabilities, including product maintenance. It does not reduce the decision to feature count. In our example, check whether the company can operate the chosen reservation method after launch too.

Before a larger commitment identify missing evidence. For manual work it might be a week recording time and mistakes. For a purchase, a trial of your own exception. For a build, a small piece showing the hardest connection. Do not begin with three full deployments. Select an unknown whose resolution could actually change the decision and an end condition for that trial.

Aeneia conceptual photography

Do not omit exceptions from any column. If every unusual reservation still goes to a person, record that work. It may be small and justified, or consume more time than ordinary cases. The label “automatic” should not hide someone manually completing the rest of the process after hours.

Compare scale changes too. Two workshops a month and several events a day have different requirements. With a seasonal peak in enquiries (in Polish), a temporary pile-up may be the problem rather than a permanent need for a new system. The solution should fit that work distribution instead of assuming the entire year runs at peak load.

Maintenance has more than one line in the calculation

For each option list preparation, daily operation, checks, changes and ending use. For AI-supported work the total process cost (in Polish) helps. The price of generating an answer alone does not show the work needed for a useful result. Similarly a subscription alone does not describe an entire purchase. For an external service also check its unit of sale (in Polish). Specialist time, a finished item and access to a resource leave different work with the company. Only understood scope can be compared with building or doing it yourself.

Separate cost from the availability of skills. Two hours from someone with a full week will not appear in the calendar because they were entered in a calculation. For each option confirm who can do the work and when. If other tasks must move, show that as a separate organisational decision.

You need not start with a model of dozens of assumptions. Begin with effort you can identify on an actual task and mark unknowns explicitly. The decision-maker should see what was measured, estimated and needs a trial. Apparent precision can merely give several unconfirmed guesses an elegant format.

Prepare two or three possible future conditions. What if event volume grows, scope is simplified or the provider changes its offer? Strategic scenarios (in Polish) test a decision's resilience without pretending to know the future calendar. The point is not to claim which variant will surely occur.

Record transition effects. Who prepares data, corrects names and checks old reservations? Must two registers run during transition? This work does not belong exclusively to any destination tool, but the company must still do it. Without it, a new app may start with old ambiguities transferred into a more convenient form.

Test a change in requirements using one example. The company adds a second warehouse and needs to distinguish local availability. What work does that mean in a spreadsheet, purchased tool and custom application? Who confirms scope and accepts the result? This reveals more than the general label “flexible system”.

For AI-supported tools or processes, discuss with Aeneia Audit whether the present workflow, costs and responsibilities can be included in an agreed review. One task, its exception and what the company can independently maintain today are useful discussion materials.

A choice needs a way back

For a purchased tool, check whether you can take your material and reconstruct the task. An AI provider exit plan (in Polish) shows why downloading a file alone need not restore a usable process. Choose a small example and give the export to someone who did not prepare it.

For a custom solution, check knowledge transfer and how to stop further development. Also test whether a substitute for its owner can reconstruct a reservation and explain an exception. This is practical handoff acceptance independent of the tool selected. Custom code, a ready platform and a spreadsheet may need different instructions, but each should let work continue without the first author present.

Beware a decision driven by past effort. Existing mock-ups or several demonstrations do not describe future usefulness. Record future maintenance cost and the effect of stopping. If a process will close, connect the decision to winding down a service for existing customers (in Polish) where their route is affected.

Set a review date and the information needed to return to the decision. If manual work remains because volume is small, also record how a change in that premise will be recognised. If a purchase wins by handling a specific exception, keep that case in the trial after updates. The team need not reconstruct the whole discussion when conditions change.

Before approval, give the comparison to someone who will use the chosen workflow. Ask them to walk through one case and identify where they do not know what to do. Their comment may concern an instruction rather than the tool itself. Keep it with the decision and agreed fix so choosing a solution does not end discussion of how work happens.

The final choice may be mixed: a ready calendar, a custom allocation rule and manual handling of rare exceptions. Name boundaries and owners of the connections. Otherwise the company will not have chosen three sensible components; it will have built an unplanned fourth option, maintaining everything at once.

A good rationale fits concrete terms: this task, these conditions, this responsibility and this moment to revisit. A build-or-buy discussion can then end with a maintainable choice. The spreadsheet may stay, leave or receive a humbler role, with the written job description it apparently deserved long ago.

Aeneia / Aeneia editorial

Editorial article. Source links are included in the text.

About Aeneia

Brands we have worked with

  • L’Oréal
  • Garnier
  • Netflix
  • Carrefour
  • Warner Bros. Discovery
  • Biedronka
  • HBO Max
  • Wolt
  • Żabka
  • COSRX
  • Embassy of China in Poland
  • Mixa
Aeneia / Initium

From knowledge to a decision

Tell us what you want to put in order. We will agree on where to start.

Contact Aeneia

We partner with and use solutions from:

  • Google
  • OpenAI
  • Microsoft
  • Stripe
  • Anthropic
  • Linear