Practical research guide · desk research and BespokeData methodology · 18 Nov 2026
Analytics platform selection
Compare options against the decisions, controls and ongoing costs your business actually needs.
Intended reader: Business owners and finance, operations or technology leads choosing a first analytics tool or reviewing an existing one.
Set decision-led requirements first. Reject options that fail hard gates, test the remaining choices with representative data, and compare total cost and exit requirements—not just licence price.
Executive answer: requirements before rankings
Choose the smallest workable option that meets your essential requirements and can be tested with representative data. Do not start with a vendor leaderboard, a licence discount or a demonstration built on perfect data. Write down the decisions the system must support, the people who will use it and the controls it must enforce. Eliminate options that fail a hard requirement before comparing weighted preferences. A high total cannot compensate for a missing permission boundary or an unusable export.
Microsoft's planning guidance connects BI strategy with business alignment and solution planning with requirements and proof-of-concept work [M1, M2]. That is useful context, not independent evidence that a Microsoft product is best for your organisation. This guide's option categories, scores and calculations are original illustrative planning tools. There are no live vendor prices, product endorsements or universal recommendations here.
1. Write a decision-led requirements sheet
For each recurring decision, specify the owner, frequency, input systems, required measure and action. For example: a branch manager reviews yesterday's unfulfilled orders each morning and assigns exceptions before the daily dispatch cut-off. Specify how fresh the data must be and what delay is tolerable. Distinguish an attractive visual from a necessary outcome. A dashboard that refreshes every minute is not automatically useful when the responsible person can only act once per day.
List access roles, data sensitivity, expected users, support responsibility and permitted hosting arrangements. Define a hard requirement as something that makes the option unacceptable if absent, not merely inconvenient. Examples might include approved regional storage, row-level access separation or a complete export in an agreed format. Confirm these with the relevant owner. Mark unverified claims as unknown and ask for evidence; a sales assurance is not the same as a tested capability in your configuration.
2. Compare three kinds of option
A spreadsheet-based workflow may be appropriate for a small, stable dataset and a few authorised users when ownership, version control and checking are clear. Its apparent simplicity can hide manual refresh work and fragile joins. A managed BI option may provide repeatable reporting and shared views, but still needs someone to define measures, administer access and maintain connections. Neither category solves inconsistent source definitions simply by importing them.
A governed data platform may suit several systems, complex permissions or reusable shared data products. It can also introduce engineering, administration and operating costs that a small team cannot sustain. These are categories, not vendor capability guarantees. Particular products vary. Consider whether a narrower process change or an improved extract could meet the immediate decision need before committing to a larger architecture. Document the future requirement that would justify moving beyond the simpler option.
3. Use an explicit weighted rubric after hard gates
For the worked example, choose five preferences: decision fit 30%, governance fit 25%, usability 20%, maintainability 15% and cost fit 10%. The weights total 100%. They reflect this hypothetical organisation's priorities, not a BespokeData standard. Rate each option from 1 to 5 using evidence: 1 means materially weak for the agreed requirement, 3 means adequate with identified trade-offs and 5 means strongly demonstrated. An untested feature should remain unknown rather than receiving a flattering middle score.
Involve the people who will operate the system in scoring. Record a short rationale and a test reference beside each value. Keep hard requirements in a separate pass/fail/unknown register. Only calculate a comparable total after all essential gates pass. For uncertain preferences, show a range or defer ranking until the pilot provides evidence. A numerical rubric is a structured conversation, not a substitute for judgement or a guarantee that the winning option will be adopted.
4. Worked hypothetical scores and arithmetic
Assume all three illustrative options have passed the same hard gates. In the preference order above, the spreadsheet scores 4, 2, 4, 2, 5; managed BI scores 4, 4, 4, 3, 3; and the governed platform scores 3, 5, 3, 5, 2. These values describe no actual vendor. They are chosen to make the trade-offs visible. The CSV includes the criteria, weights, scores, weighted contributions and space for evidence.
For the spreadsheet, the total is (30×4 + 25×2 + 20×4 + 15×2 + 10×5) ÷ 100 = 3.30 out of 5. Managed BI is (120 + 100 + 80 + 45 + 30) ÷ 100 = 3.75. The governed platform is (90 + 125 + 60 + 75 + 20) ÷ 100 = 3.70. Managed BI ranks first under these particular inputs, but its margin over the governed platform is only 0.05 points. That is a reason to examine assumptions, not declare an overwhelming winner.
For a sensitivity check, move ten percentage points from usability to governance: weights become 30, 35, 10, 15, 10. The totals become 3.10, 3.75 and 3.90 respectively. The governed platform now leads. This does not prove the second weighting is correct. It shows that the decision depends on the relative importance of governance and usability. Agree that priority explicitly, and test disputed scores before making a commitment.
5. Pilot representative cases, including failure
Prepare a small authorised dataset that represents ordinary work and difficult exceptions: duplicate IDs, amended orders, missing fields, timezone boundaries and restricted records. Ask each candidate to produce the same defined output. Reconcile totals with a checked baseline, test refresh failures and observe whether a non-specialist can find and resolve an exception. Measure task completion and support burden, not just whether the chart looks polished. Keep any necessary test-data masking documented.
Include permission tests with different roles and an export test that another tool can read. Confirm whether promised features require a different plan, additional service or custom work. Record the exact version and configuration tested. Agree pilot acceptance criteria in advance and name who signs them off. A supplier demonstration may help explore possibilities, but it does not replace your own test cases or establish that the eventual production arrangement meets your requirements.
6. Evaluate total cost and the exit path
Build a cost register for licences, setup, data cleaning, migration, connectors, compute, storage, administration, support, training and ongoing change. Identify the billing unit and who controls it. Separate one-off costs from recurring costs, and note what happens as users, refresh frequency or data volumes grow. Use supplier quotations for actual prices and record their date and assumptions. This guide deliberately supplies no invented pricing or payback claim.
Estimate internal time as well as invoices. A hypothetical two hours of weekly maintenance for 48 working weeks is 96 hours per year; it is not free because no subscription appears on an invoice. If comparing money totals, apply an explicitly agreed internal hourly cost and consistent currency. Include the transition period when old and new workflows coexist. Do not compare a fully supported option with an unsupported one as if their scope were identical.
Before signing, establish how to export data, measure definitions, transformation logic and access documentation. Test a usable export rather than assuming a download button preserves everything. Clarify termination, retention and deletion obligations with the appropriate advisers. Plan ownership if the original builder leaves. An exit requirement is valuable even when you expect to stay: it reduces dependence on undocumented logic and makes routine recovery more credible.
7. Use the scoring worksheet and record the decision
The CSV contains the illustrative rubric and contribution calculations as values, not executable spreadsheet formulas. Add your own hard-gate evidence separately or use the final gate row as a starting point. Change weights only after agreeing priorities; then recalculate every option consistently. Retain the previous version so the reason for a changed ranking remains visible. Add confidence and pilot evidence to the notes column rather than treating guessed scores as observed facts.
Write a final decision memo that states the selected scope, rejected alternatives, unresolved risks, operating owner and review trigger. The right answer may be a time-limited pilot, a better spreadsheet or a deferred purchase while source definitions are repaired. Implementation success also requires time for training and ownership of measures. A tool selection is a beginning, not proof that the business will use the information to make better decisions.
| criterion | illustrative_weight_percent | spreadsheet_score | managed_bi_score | governed_platform_score | spreadsheet_contribution | managed_bi_contribution | governed_platform_contribution | evidence_and_notes |
|---|---|---|---|---|---|---|---|---|
| Decision fit | 30 | 4 | 4 | 3 | 1.2 | 1.2 | 0.9 | Hypothetical scores; replace with pilot evidence |
| Governance fit | 25 | 2 | 4 | 5 | 0.5 | 1 | 1.25 | Hard requirements must pass separately |
| Usability | 20 | 4 | 4 | 3 | 0.8 | 0.8 | 0.6 | Observe representative users |
| Maintainability | 15 | 2 | 3 | 5 | 0.3 | 0.45 | 0.75 | Name operating owner |
| Cost fit | 10 | 5 | 3 | 2 | 0.5 | 0.3 | 0.2 | Use dated quotes; no vendor prices here |
| Total | 100 | — | — | — | 3.3 | 3.75 | 3.7 | Scores out of 5; not vendor rankings |
| Hard gate: replace with requirement | — | — | — | — | — | — | — | Pass/fail/unknown; failure excludes option |
Limitations and next steps
This desk-research guide is not a procurement audit, live comparison or independent expert review. Vendor documentation is contextual and can change. The hypothetical weights cannot settle your security, legal or commercial requirements. Start with three decision statements and the hard-gate register, then obtain current evidence and quotations. A commissioned assessment can help define requirements and evaluate authorised pilot evidence within a scope agreed in a proposal.
Dated method & source register
Read and reviewed 18 Nov 2026. Desk research and original BespokeData recommendations and hypothetical examples; no primary study, independent expert review or Atora generation.
- [M1] Microsoft · BI strategic planning
Business alignment guidance from a platform vendor, not an independent product comparison.
- [M2] Microsoft · BI solution planning
Requirements and proof-of-concept planning; vendor-specific context.
Apply this to your business
Agree a scope around your decision and authorised evidence.
Discuss this goal →Read the fictional sample report →