# How to choose a system for an organisation that already works Existing workflows, unknowns and a trial before change. Compare staying, integrating and migrating before choosing a product. L00P.AI · 1.0 · 2026-09-26 https://www.l00p.ai/en/resources/series/jak-wybierac-system-do-dzialajacej-organizacji/ ## 01 / Do not start with a list of products A working organisation resembles a workshop where people know where the tools are. A new cabinet may be better, but buying it does not move its contents or people's habits. When choosing a system, ask: what problem must we solve, and what must keep working during the change? Our starting point is an internal September 2026 study of radio systems. New information about the existing environment changed its recommendation. We adapt that method without disclosing a newsroom's workflows, configuration or costs. The example below is neutral; it does not describe a completed deployment. The same sequence can guide the choice of an archive, CRM or production tool. ## 02 / Separate facts, requirements and unknowns Write down one real workflow: where material arrives from, who approves it, where the output goes and what happens on failure. Include dependencies on equipment, other programs and people. [The owner's knowledge](/en/resources/series/wiedza-wlasciciela-zmienia-kod/) often determines which constraint is mandatory and which is merely inherited. Label each important statement. “We use Windows” describes the current situation. “The new workstation must use Windows” is a requirement somebody should justify. “We do not know whether we can dedicate a workstation” remains an unknown. These statements produce different candidate lists. Give each unknown a verification method, an owner and an explanation of its effect on the decision. Software version, licence edition, device model and firmware may determine compatibility. A hardware family name alone does not settle it. Missing information about a feature is not proof that it is absent. ## 03 / Include keeping the existing solution Consider an invented problem: a team wants to reduce manual reconciliation of plans and execution. It compares three options: improve the current system's report, add a small integration or migrate the whole process. Each has the same acceptance goal but a different scope of change. First check mandatory conditions, such as preserving data, supporting the required device and being able to roll back. Do not offset a missing critical function with points for an attractive interface. Only compare convenience, handling time and full cost among options that meet those conditions. Set the criteria before a demonstration, so an impressive presentation does not redefine the problem. ![Need and options lead to mandatory conditions. Failed: revise or defer. Passed: trial. Trial not accepted: correct and retest. Accepted: owner decides on deployment.](https://www.l00p.ai/wydawnictwo/jak-wybierac-system-do-dzialajacej-organizacji/wybor-en-v1.svg) Original method diagram, Codex / L00P.AI. Rectangle: activity; diamond: decision. This is a procedure, not a test result. Read the diagram as follows: describe the need and options, check mandatory conditions, then revise or defer options that fail them. Prepare a trial for the remaining ones. Failed trial criteria mean correction and another acceptance review; passing them only enables a deployment decision. No arrow replaces the owner's decision. ## 04 / Documentation narrows the choice; a trial resolves it The documentation review of 26 September 2026 covers three products, but is not a comparative test. [RadioDJ](https://www.radiodj.ro/) describes exporting now-playing information through a plugin. [mAirList documentation](https://wiki.mairlist.com/config:remotecontrol:dhd) describes DHD control; the appropriate protocol, edition and configuration must be established. [Rivendell's RDLogManager](https://opsguide.rivendellaudio.org/html/chapter.rdlogmanager.html) describes schedule import and reconciliation reports. These are different starting points, not equivalent evidence of a working integration. [Comparison appendix 1.0 — EN](/wydawnictwo/jak-wybierac-system-do-dzialajacej-organizacji/porownanie-en-v1.md) records scope, references, gaps and proposed trials. A [PL/EN JSON edition](/wydawnictwo/jak-wybierac-system-do-dzialajacej-organizacji/porownanie-v1.json) is also available. We do not reproduce historical prices or “best product” claims. A commercial quotation and availability of required functions in a particular licence need separate confirmation. ## 05 / Check what a report entry actually means A schedule says what was supposed to happen. A program log says which events the program recorded. A recording from a specified point in the signal path lets you check the audio present there. These are three different kinds of evidence. A file-start entry does not independently establish that the signal reached a listener. [AdMaster](https://www.admaster.info/en/reconciliation.html) describes reconciling a schedule against a log using a selected identifier: spot name, filename or cart number. It requires exact matching and a timestamp with seconds precision. Do not assume that a filename always appears in a report, or that a title uniquely identifies an item. In our example, we design a trial covering correct playback, a missing event, a duplicate and a mismatched identifier. We compare the schedule, log and monitoring recording, documenting the observation point and clock alignment. An empty result must be distinguished from a failed read. This is a proposed procedure, not the results of a completed test. ## 06 / Write acceptance criteria before installation Run the trial in a separate environment using material you have rights to use. Record versions, settings and the expected result of each case. The process owner should define acceptable latency and required recovery time; there is no single suitable value for every organisation. - Control: a command produces the expected action, and connection loss is recognisable. - Data: identifiers and necessary metadata transfer without silent truncation. - Reporting: missing events, duplicates and errors have distinct, understandable results. - Failure: the rollback procedure restores the required work within the agreed time. - Operation: the designated person completes the normal workflow using the prepared instructions. Retain evidence and record “not tested” for any case that was not performed. [Integration acceptance](/en/resources/series/http-200-to-dopiero-poczatek/) requires checking the outcome, not just a success message. A supplier demonstration may help design the trial; it is not acceptance of your own environment. ## 07 / Cost the transition and the return route Add discovery, migration, training, parallel operation, support and maintenance to the licence price. A free licence does not mean a zero-cost change. [Project budgeting](/en/resources/series/za-co-placi-sie-w-projekcie-agentowym/) helps separate effort, expenditure and contingency. Keeping the old computer is not yet a tested rollback plan. You need current data, access, compatible formats and a rehearsed switching sequence. The decision may be “deploy one stage”, “improve the trial” or “stay”. Record the reason and a reconsideration trigger, such as changed requirements or a measurement result. To start a conversation, bring one process description and a list of unknowns. We can help prepare trial criteria without publishing internal configurations, contracts or customer data. ## 08 / Provenance and limits We read the complete internal study and the cited public sources. Documentation was reviewed on 26 September 2026; these products were neither installed nor tested. The comparison version denotes the appendix edition, not a software version. Changes to documentation, licensing or the environment, or an error in our account, trigger a new review. Codex prepared the PL/EN adaptation, appendix and original SVG. The diagram uses rectangles for activities and diamonds for decisions; it is neither a certified model nor a measurement. No third-party illustrations were used. The author's review is not an independent second-model review. A human can withdraw the material after publication.