International Digital Health ROI Model for Buyers
A useful international digital health ROI model is a decision record, not a sales calculator. It compares a named local workflow with a credible alternative, keeps actual cash separate from released capacity and health outcomes, and leaves missing evidence visible. Build one country record at a time before considering a regional total.
This guide is for technology evaluation and procurement planning. It is not medical, legal, tax, accounting, reimbursement, compliance, or investment advice. Local finance and qualified clinical, privacy, security, legal, and operational owners must validate the inputs and decide whether use is permitted. A positive ROI cannot override a failed clinical, safety, legal, privacy, or security gate.
Gate 1: fix the decision and country boundary
State the decision before opening a spreadsheet: pilot, continue, expand, renegotiate, hold, or exit. Name the country, contracting entity, budget holder, care setting, workflow, population, language, product version, deployment model, and comparison alternative. Declare whose perspective the model represents, which currency and price date apply, and the start and end of the evaluation horizon. If patients, caregivers, providers, payers, or ministries bear different costs, show those perspectives separately instead of hiding a transfer inside a net figure.
Evidence required: a signed scope record, accountable decision owner, approved comparator, current workflow map, budget boundary, currency basis, measurement horizon, and explicit exclusions.
Open rule: if the decision, country, entity, perspective, comparator, currency, or horizon is missing, mark the gate open — local owner/evidence required. Do not reuse one country's approval or business case as a regional answer.
Gate 2: make the baseline and comparator reproducible
Define the same metric for the current and proposed pathways. Record numerator, denominator, unit, data source, extraction version, eligible cohort, exclusions, missingness, and measurement windows. Include review, exception handling, rework, support, downtime, and abandoned cases in both paths. Note seasonal changes, staffing changes, policy changes, concurrent projects, and selection differences that could explain a before-and-after result.
Evidence required: versioned extracts or reports that another reviewer can reproduce, plus a data-quality note and a comparison design appropriate to the decision.
Stop rule: if the baseline cannot be reproduced or the comparison changes the population, work definition, or measurement period without a defensible adjustment, stop the ROI claim. Keep the finding as an unresolved observation rather than a saving.
Gate 3: classify each effect before valuing it
Classify every claimed effect before assigning a value: cash outflow or receipt, workforce capacity, quality, access, experience, equity, clinical or safety outcome, or unresolved harm. Record who receives the effect and who must act for it to be realized. A shorter task may release capacity without changing payroll, overtime, contractor spend, revenue, or another cash line. A health or experience outcome can remain decision-relevant in its native measure without being converted into money.
Evidence required: a benefit ledger with a unique benefit ID, measurement definition, recipient, evidence status, realization action, owner, negative effects, and overlap check.
Stop rule: do not move capacity or a native outcome into cash without evidence of an actual budget or receipt change. Do not omit an adverse result because it weakens the headline case.
Gate 4: reconcile lifecycle costs and benefit overlap
Bring current quoted cost inputs into the model, then add local integration, localization, training, governance, clinical review, privacy and security review, implementation, monitoring, support, rework, parallel running, renewal, migration, and exit obligations. Identify central costs and country allocations. Assign a unique ID to every cost and benefit so the same invoice, labor item, saving, or released hour cannot appear twice.
Evidence required: versioned cost lines with country, entity, payer, period, currency, source, due date, allocation method, overlap key, and a list of mandatory but unpriced obligations.
Open rule: when a mandatory cost is unknown, show the known subtotal and mark the model open — local owner/evidence required. Do not call a partial subtotal total cost or complete ROI.
Gate 5: test downside and stopping conditions
Test the conditions that would weaken or reverse the case: lower adoption, delayed go-live, higher review and rework, additional support, benefit decay, currency movement, local cost changes, incomplete country rollout, incident response, contract termination, data export, and transition back to the prior workflow. Separate a one-driver sensitivity from a combined downside case.
Evidence required: a versioned scenario sheet in which every changed input names its source, owner, affected period, and relationship to other inputs, plus explicit hold and stop thresholds.
Stop rule: if the decision only works when unknown inputs are silently treated as zero or when a plausible exit cost is omitted, do not approve it as a complete case. A modeled downside is a planning test, not a statistical confidence interval.
Gate 6: require country-level realization and revalidation
After a pilot or launch, reconcile modeled inputs with observed evidence and with actual budget actions. Record whether released capacity was used, whether a planned cash action occurred, and whether support, safety, equity, or access outcomes changed. Revalidate when the country, population, language, care pathway, hosting region, integration, product version, commercial terms, or local unit costs change.
Evidence required: a country decision record with modeled and observed values kept distinct, budget confirmation, negative findings, limitations, approval status, expiration date, and next review trigger.
Stop rule: do not expand from one country merely because its model is positive. Hold expansion when required controls did not operate, a high-impact issue remains unresolved, or the target country lacks local evidence.
Questions by accountable owner
Finance and procurement owners
- Which entity, budget, currency, and period owns each payment or supported receipt?
- Which investment IDs are already inside outflows_with, and has any one been deducted twice?
- Which required costs remain unquoted, centrally allocated, conditional, or due at exit?
Local operations and workforce owners
- Do the current and proposed workflows include the same review, exception, rework, and support tasks?
- Where will released hours actually go, and will that action change cash, capacity, or neither?
- Which costs continue when adoption is low, go-live is delayed, or the workflow is stopped?
Evaluation, clinical, and data owners
- Can another reviewer reproduce the cohort, comparator, windows, exclusions, and missing cases?
- Which effects are observed, modeled, or vendor assertion, and what limits attribution?
- Are negative findings and country, language, pathway, and subgroup differences preserved?
Regional governance and decision owners
- Are costs shifted to patients, caregivers, providers, or another public or private budget?
- Which inputs and controls must be revalidated before a new country or workflow is added?
- Who can hold or stop use when an economic or non-economic threshold is crossed?
Five buyer-owned ROI records
Every record should carry an evidence link, source version, observed date, limitation, accountable owner, and next action. Use open — local owner/evidence required when one of those fields is unresolved.
A. Country baseline and comparator record
Capture the entity, country, workflow, cohort, product version, perspective, comparator, baseline and follow-up windows, metric definition, unit, extraction method, missingness, concurrent changes, and reproducibility result. This record decides whether the two pathways are comparable, need adjustment, or cannot support an ROI claim.
B. Benefit evidence and realization ledger
Give each effect a benefit ID and classify it as cash, capacity, quality, access, experience, equity, clinical or safety outcome, or unresolved harm. Record the recipient, evidence status, measurement, negative effects, causal limitation, realization action, verification date, and whether another benefit overlaps.
C. Cash and capacity reconciliation
List country, entity, perspective, currency, price date, period, with-and-without outflows and receipts, investment IDs, central allocations, internal transfers, capacity hours, and unpriced obligations. Show which entries are modeled and which are observed. A finance owner should be able to reproduce every total without consulting a vendor slide.
D. Scenario and sensitivity sheet
Keep planned-use, downside, and stop or exit inputs in separate columns. For each changed driver, record the source, range or state, coupled drivers, period cash result, capacity result, native outcomes, hard stop, and break-even boundary if it can be computed without inventing inputs.
E. Transferability and realization decision
Compare source and target countries across population, language, pathway, regulation, support, local unit costs, infrastructure, and control operation. Record validation still needed, budget action, expand, hold, or stop status, decision expiry, negative findings, and the next review. This record prevents a regional average from erasing a country-level loss or harm.
Keep cash, capacity, and native outcomes separate
- Cash: actual outflows and supported receipts within the declared entity, perspective, currency, and period.
- Capacity: comparable staff hours released or added; it does not automatically become cash.
- Quality, access, experience, and equity: report these in native units with numerator, denominator, cohort, and limitations.
- Clinical and safety outcomes: keep the clinically meaningful native units and qualified review separate from the financial calculation.
- Unresolved harm and risk: record the event, exposure, severity, affected group, control, owner, and open decision rather than assigning a favorable monetary value.
- Lifecycle cost: include implementation, local operation, control, support, rework, renewal, transition, and exit cash lines without double counting them.
The categories can appear together in a decision pack, but they should not be summed into one attractive score. A decision may be unaffordable despite valuable non-cash outcomes, or financially positive while failing a safety or rights gate.
Use transparent planning identities
Use these identities only after every term has a declared unit and evidence status:
cash_delta = outflows_without - outflows_with + receipts_with - receipts_without
All four terms must use the same country, entity, perspective, currency, and period. Receipts are supported cash receipts, not a monetary value assigned to health, time, experience, or equity. If an avoided cost is already expressed by the difference between outflows_without and outflows_with, Do not add the saving again.
net_cash = sum(cash_delta across non-overlapping periods)
State the horizon and keep periods non-overlapping. This is not net present value. This guide does not define a default discount rate; if discounting is needed, local finance must specify the price basis, inflation treatment, discount method, and consistent timing. Payback depends on the cumulative cash series, and later exit costs can reverse an earlier crossing.
cash_roi = net_cash / investment_cash × 100
investment_cash is a positive subset of outflows_with for the same decision scope and horizon. Do not subtract it again from net_cash. This percentage is not a benefit-cost ratio, not a full cost-benefit analysis, and not a cost-effectiveness result. When the denominator is missing or less than or equal to zero, or any mandatory cost remains unknown, do not calculate a complete ROI. Show known cash lines and open items instead.
capacity_hours = (minutes_without - minutes_with) × matched_volume / 60
Use the same completed-work definition and period. Both minute terms must include review, exceptions, rework, and support where applicable, so those minutes are not deducted a second time. matched_volume is comparable completed volume, not forecast demand. Keep capacity_hours outside cash_delta unless an evidenced action changes an actual cash line.
Reconcile overlap before aggregation. Do not count both the central line and its allocation. Eliminate transfers between included entities. A saving, receipt, capacity release, avoided overtime line, or extra output with the same underlying resource must have one benefit ID and one treatment. To form a group view, align a common currency and period, eliminate internal transfers, sum country net_cash and investment_cash separately, and calculate group ROI as the sum of net_cash divided by the sum of investment_cash. Do not average country ROI percentages, and continue to show each country's result.
Build three blank scenarios
Start every input blank; unknown is not zero. A verified zero needs a source, and not applicable needs an explanation. These scenarios are planning cases, not confidence intervals.
Planned-use scenario
Enter only owner-approved assumptions for adoption, implementation delay, review and rework, support, benefit decay, FX, local costs, country expansion, realization action, and renewal or exit. Keep forecast entries labeled as modeled even when they come from a contract or implementation plan.
Downside scenario
Change one driver at a time to show sensitivity, then test a combined downside where related changes occur together. Include lower adoption, delay, higher rework and support, benefit decay, adverse FX or local costs, slower country expansion, and any control or safety cost that remains within the permitted use case.
Stop and exit scenario
Model the date on which new use stops, which obligations continue, and the cash and capacity needed for notice, parallel running, data export, record reconciliation, migration, support, remediation, and contract exit. Preserve native outcomes and unresolved harm outside the cash total.
Apply attribution and sensitivity without inventing certainty
A simple before-and-after difference does not establish causality. Record changes in population, staffing, demand, policy, season, measurement, and other projects. Where feasible, a credible comparator can improve interpretation, but it does not remove every bias. Label each result observed, modeled, or vendor assertion and keep its limitation beside it.
Vary one driver at a time to identify influential assumptions, then test a combined downside for related drivers. Keep negative findings and abandoned assumptions; do not select only favorable outcomes or favorable countries. If an effect cannot be reproduced, leave it open instead of converting it to a cash benefit.
For regional reporting, preserve country records first. Convert cash only with an approved common currency and period, eliminate internal transfers and central allocation overlap, then calculate the group ratio as sum net_cash divided by the sum of investment_cash. Show country exceptions, costs borne outside the chosen perspective, and native outcomes alongside the group cash view.
FAQs
Can ROI be calculated before a pilot?
A pre-pilot model can compare transparent scenarios, but it remains modeled. The pilot should replace assumptions with reproducible local evidence and record what still cannot be attributed.
Is released capacity a cash saving?
No. Capacity becomes cash only when an evidenced action changes a real outflow or supported receipt. Otherwise report the hours and their actual operational use separately.
Is positive ROI permission to expand?
No. Economic evidence is one input. Required clinical, safety, legal, privacy, security, equity, and operational gates still need their accountable owners and evidence.
Can one country's ROI be reused elsewhere?
Not without local validation. Currency, unit costs, population, workflow, language, infrastructure, regulation, support, adoption, and benefit realization can all change the result.
What if required inputs are missing?
Show the known lines, label the missing item and owner, and stop the complete calculation. Do not replace an unknown with zero or a vendor benchmark.
Handoff to adjacent decision pages
Use pricing questions for versioned cost inputs and commercial exposure. Use the vendor evaluation guide for product and supplier fit, evidence, and procurement risk. Use pilot planning for pilot execution, measurement operations, and exit readiness.
Use the buyer checklist for procurement completeness, compliance questions for local compliance review, and the security review for security review and control evidence. This ROI record supplies an economic view to those decisions; it does not complete their reviews.
Evidence boundary
The evaluation structure above is an editorial buyer framework, not an agency-endorsed template. The UK Office for Health Improvement and Disparities economic evaluation guide supports comparing alternatives, declaring perspective and horizon, and separating resource use from unit cost. Its budget impact guide distinguishes affordability for a budget holder from value for money. The cost consequence guide supports showing different positive and negative effects separately, while the cost benefit guide explains why monetary valuation has limits. The before-and-after guide supports the attribution caution above. These UK methods are not global product or accounting rules.
The WHO pages for monitoring and evaluating digital health interventions, the Digital Implementation Investment Guide, and the Global strategy on digital health 2020–2027 provide implementation, readiness, monitoring, and resource context. This review covered the overview pages only; the linked full publications were not treated as reviewed line by line.
These sources do not establish product performance, local workflow fit, legal applicability, or vendor approval. No product tests, vendor quotes, customer interviews, local performance measurements, or completed financial review are claimed. Buyers must obtain current local evidence and qualified review before relying on the model.