A well-researched need does not require building the first solution you imagine. Designing also means deciding what not to build.
When an organization discovers a problem, its response can seem obvious: “We need an app,” “We should buy a machine,” or “We must hire more people.” Yet these are hypotheses about solutions, not automatic consequences of the diagnosis. A team could reduce reservation mistakes by changing its confirmation procedure, buying software, adapting a tool, or developing a custom product. Those alternatives have different costs, risks, user experiences, and dependencies. The UK Government Digital Service recommends exploring and testing multiple solutions during its alpha phase before committing to full construction [R01]. That is a useful methodological reference, not a requirement that every business adopt UK government phases.
1. Separate the problem, outcome, and alternatives
Start from the problem statement and prioritized need, not a list of tools. Distinguish:
| Element | Question | Hypothetical example |
|---|---|---|
| Observed problem | What fails today, and whom does it affect? | Reservation changes are communicated too late |
| Desired outcome | What should improve for that person? | Staff can see and confirm changes without duplicates |
| Constraint | What condition must be respected? | Small team and unreliable connectivity |
| Solution alternative | What approach could produce the outcome? | Calling procedure, purchased calendar, or custom system |
| Critical hypothesis | Which assumption could make the option fail? | Employees will follow the new procedure |
A value proposition describes the benefit we promise; selecting an alternative means explaining how we might deliver it. If the need is to reduce missed changes, “a calendar with twenty modules” is not yet a specification. The competitor research in Pillar 1 and value proposition in Pillar 2 provide inputs, but neither makes this design decision for you.
If the outcome that each option should improve is still unclear, revisit the value proposition. Comparing solutions requires a concrete promise as its reference, not a technology preference.
2. Expand the option space before narrowing it
Generate possible approaches before picking a winner. You can ideate individually for ten minutes, compare ideas with colleagues, and then combine or simplify proposals. Questions such as “What if we could not use an app?” or “What could be handled manually for one month?” counter an attachment to preferred technology.
To generate plausible alternatives:
- Write the outcome, target user, and context.
- Describe the current workaround, including invisible human work.
- Produce at least three materially different paths—change a process, adapt an existing offering, and build something new—when applicable.
- Imagine specific failures for each: staff absence, unavailable materials, integration issues, rejection by users, or recurring costs.
- Remove options that violate non-negotiable constraints; do not give them compensating points for low price.
- Retain at least one lower-investment path to check whether value truly requires a complex solution.
Ideation does not guarantee originality or quality. It must connect back to evidence and the capacity to deliver. Every project does not need a huge workshop or a dozen options; it needs a comparison of genuinely distinct approaches.
3. Consider building, buying, adapting, outsourcing, or partnering
These decisions can be combined: buy a base system and build an integration, outsource manufacturing while retaining design ownership, or deliver manually while investigating automation.
| Path | Potential benefit | Dependency or cost to investigate |
|---|---|---|
| Build | Control over features and evolution | Engineering, security, support, maintenance |
| Buy | Potentially faster start | Licenses, functional limits, vendor continuity |
| Adapt | Reuse a foundation and customize it | Compatibility, licenses, customization limits |
| Outsource | Access skills or capacity | Contracted quality, knowledge transfer, changes |
| Partner | Share assets, channels, or resources | Coordination, accountability, dependency, exit |
Compare total cost, not just the purchase price: staffing, testing, upgrades, data handling, training, support, vendor switching, and exit. A low-cost license can be expensive if it doubles the work; seemingly free custom software creates maintenance commitments that someone must fund.
Before placing customer data in an external tool or automating sensitive decisions, investigate the relevant privacy, security, permissions, and responsibilities. Commercial availability does not prove that a product meets the obligations of your context.
4. Build a decision matrix without inventing precision
Choose criteria grounded in the desired outcome and constraints: fit to the problem, user experience, adoption effort, estimated total cost, timeline, reliability, upgrade path, vendor dependence, and compliance risk.
First check mandatory conditions. If a provider cannot meet an essential obligation, a “5/5 price score” cannot make it acceptable. Then use verifiable data, ranges, or explicit labels (“known,” “estimated,” “unsupported”) and explain differences. Weighted matrices help only if weights and scores are justified; arithmetic does not turn guesses into objectivity.
An illustration, not a real evaluation:
| Option | Core value | Up-front investment | Dependency | Critical uncertainty |
|---|---|---|---|---|
| Revised manual procedure | Might handle basic changes | Low | Staff availability | Will it be followed at peak times? |
| Existing software | Might centralize changes | Medium, unconfirmed | Vendor and connectivity | Does it fit the actual flow without double work? |
| Custom development | Might closely match the workflow | High, to be estimated | Technical team | Does the customization justify lifetime cost? |
The table does not announce a winner. It shows what needs to be learned first. When evidence is missing, write “unknown” rather than assigning arbitrary scores.
To prevent the matrix from hiding decisive dependencies, see how to assess design constraints. A cheap alternative that fails a mandatory condition cannot win through an average score.
5. Identify hypotheses that could change the selection
Each option rests on different assumptions. For manual operations, workload may be the concern. For an off-the-shelf tool, it could be integration or data privacy. For custom software, it could be development and maintenance capacity.
Create one record per uncertainty: option, claim, consequence if wrong, current evidence, least burdensome suitable test, decision condition, and owner. If no one can say what observation would change their preference, the team may be choosing out of attachment rather than evidence.
Reuse the Pillar 3 guides for selecting experiments and designing protocols. The distinctive task here is comparing design alternatives and identifying evidence that might rule out each one, not reteaching every experimental technique.
6. Make a provisional decision and retain rejected options
A useful decision record contains: target outcome, alternatives considered, criteria, evidence, constraints, provisional selection, conditions that would invalidate it, and review date or trigger. Keep the reasons for rejecting attractive options. This prevents debates from restarting without new facts and helps explain future changes.
Do not confuse a provisional selection with an irreversible commitment. You can use a manual process in a pilot while preserving a digital option for later growth. The transition should have a condition: “We will reconsider automation if observed volume exceeds demonstrated capacity without harming service.” Teams must define that threshold based on their context; it is not a universal standard.
Worked example: last-minute reservation changes
A fictional restaurant logs twelve late changes across two weeks. This is a description of its log, not an industry-wide rate. Requests arrive through calls and messages but reach the shared calendar at different times.
Option A—procedure: assign one person per shift, keep one shared change log, and make confirmation visible. It is cheap to prepare but relies on consistent behavior. Option B—existing tool: trial a shared calendar with alerts; it might reduce duplication, but setup, connectivity, and privacy need examination. Option C—custom development: implement special interfaces and rules; it offers flexibility but increases delivery time and maintenance.
The team does not select C merely because it looks more professional. It proposes trying A for five working days, checking timestamps, ownership, and confirmation evidence. In parallel, it investigates whether B can handle exceptions without duplicating work. If A meets the need at limited scale, C is deferred; if A fails because of operational complexity, B is reconsidered. This is not a prediction of results—it defines what must be learned before spending more.
A provisional selection is not a specification: the next step is to turn the intended outcome into requirements and acceptance criteria that support building and checking the chosen option.
Common mistakes and limitations
Picking technology before defining the outcome; calling two visual variations two solution alternatives; comparing only initial investment; treating a quote as guaranteed total cost; inventing scores; overlooking security, accessibility, or applicable obligations; dismissing manual approaches out of bias; mistaking an impressive demo for sustained capacity; switching options without recording why.
Comparison reduces uncertainty but does not guarantee that the selected option delivers value. Prototypes, pilots, or later implementations still have to test the assumptions that remain.
What you should have when finished
A solution alternatives matrix containing a clear problem and outcome, three comparable paths, mandatory conditions, grounded costs and risks, critical uncertainties, planned tests, and a provisional decision. Someone else should understand both why one option was selected and why the others were not.
References
- [R01] UK Government Digital Service. How the alpha phase works. https://www.gov.uk/service-manual/agile-delivery/how-the-alpha-phase-works