Real World Evidence Platform Development: Custom Build vs Off the Shelf
Licence first. TriNetX for feasibility and Aetion when a study has to withstand challenge will answer most evidence questions faster and more cheaply than anything you build, and that is the honest call for most outcomes research teams.
On this page
Licence first. TriNetX for feasibility and Aetion when a study has to withstand challenge will answer most evidence questions faster and more cheaply than anything you build, and that is the honest call for most outcomes research teams. Build only once you licence data from three or more vendors, the joining has become your problem, and a regulator or payer is going to interrogate your methods.
What the licensed platforms actually do well
You have a renewal quote on the desk and someone on the evidence team has asked whether the analysis layer should be bought or built. Before anything else, give the incumbents their due, because several of them are genuinely strong and a fair comparison is the only one worth running.
TriNetX is a federated network sitting across provider records, and for feasibility it is hard to beat. You can size a cohort in an afternoon instead of commissioning an extract and waiting six weeks for it. Aetion is the closest thing to a purpose built real world evidence (RWE) analysis environment, with reproducibility and regulatory transparency as stated design goals and a curated method set a reviewer will recognise. Flatiron Health is an oncology data asset built on electronic health record derived and abstracted records, and the abstraction is the expensive part they do properly. Komodo Health is a claims derived asset with analytics layered on. Datavant is the tokenisation layer that joins otherwise separate assets without anyone exchanging identifiers.
Buy, and buy without embarrassment, if you run a handful of feasibility questions a quarter and licence a rigorous platform for the two or three studies a year that have to withstand challenge. That operating model is right for most health economics and outcomes research (HEOR) teams. It costs a fraction of a build, and we say so to people who arrive asking us to quote a platform.
Where they stop: the cohort nobody can reproduce
One failure moves a team across the line, and it is always the same failure.
You publish an analysis showing a treatment effect in a claims derived cohort. Nine months later a payer's own analysts run their version, get a different number, and ask how the cohort was defined. The analyst who built it has moved teams. The code set for the indication lives in a spreadsheet with three tabs, one of them called final_v2_use. The claims extract has refreshed twice since, so the same query on today's data returns a different denominator. The exclusion for prior therapy was applied in one script and the washout period in another.
Nothing dishonest happened. The work was never built to be re-executed, and re-execution is the whole product here.
A cohort is not a query. It is an index date rule, a lookback window, an inclusion phenotype expressed as code sets across ICD-10-CM, CPT, HCPCS, NDC, RxNorm, LOINC and SNOMED CT, then exclusions, a washout, a censoring rule and covariates with their own lookbacks. Change any one and the effect estimate moves. A vendor platform holds your study inside its method set and its release notes, which is workable until the challenge arrives and the answer to what changed is somebody else's changelog.
The second thing products model badly is your data licence. Contracts restrict permitted purposes, name which affiliates may touch the data, forbid re-identification attempts, constrain publication below a minimum cell size, require deletion at term end and reserve audit rights. Most teams enforce all of that with an email from legal and good intentions. Enforcing it in software, so an output below the minimum cell size is suppressed automatically and expiry raises a deletion task with evidence attached, is not something you configure into a platform you do not own.
The arithmetic: per seat fees against the cost to build
Run this with your own numbers, because the quote in your inbox is the only price that matters.
Split your renewal into named analyst seats and per study charges. Suppose the quote is $18,000 per named seat a year and 12 people need one. That is $216,000 before a single study fee, and it recurs whether or not the science was good that year. Add the fixed price studies you commission, then add the extract fees your data vendors charge on top of both.
Now put a build on the same line. A first release costs once, then roughly 15 to 20 percent of that figure a year to keep alive. Against a $216,000 subscription, the crossover lands between year two and year three.
Watch seats rather than studies. Below roughly eight to ten people who need daily access, a subscription wins comfortably, because you are buying a small amount of software and a large amount of somebody else's method work. Past roughly 25 named seats, or past roughly 40 formal studies a year drawing on three or more separately licensed datasets, the subscription buys you less every year while the joining problem stays entirely yours.
There is a quieter number in the same comparison. Count the hours your epidemiologists spend rebuilding code sets that already exist somewhere in the organisation. In teams we have worked with that is the largest line of all, and nobody has it on a slide.
What a custom build actually costs
From Digital Heroes delivery experience, a first release covering ingestion of two or three licensed datasets, mapping to a common data model with the source faithful layer retained, versioned phenotype and cohort objects and a reproducible execution record runs $110,000 to $230,000 and ships in 14 to 20 weeks. A full platform adding tokenised linkage, licence term enforcement, note extraction with validation reporting, analysis packaging for regulatory and payer submission and compute cost governance runs $300,000 to $750,000 phased over 9 to 16 months.
Two lines nobody quotes. Data migration runs 10 to 25 percent of the build cost, and it sits at the top of that range when your phenotypes exist only as analyst scripts, because a person has to read each one and decide what it meant before it can be versioned. Year two and every year after runs 15 to 20 percent of build cost annually, covering vendor file format changes, vocabulary releases, cloud spend and the enhancements a live evidence function always generates.
What pushes the number up is specific to this category. Every additional source vendor is its own onboarding project, because delivery formats and refresh cadences never match. Tokenised linkage across assets adds real design. International data means a separate hosting and privacy design per country. Note extraction is not the model, it is validating the model against a manually abstracted gold standard, which is a study in itself. What keeps the number down is onboarding two datasets properly and running three real studies on them before you add a third source.
The four situations where building wins
Regulatory fit comes first. If your evidence supports label discussions, a submission under the United States Food and Drug Administration real world evidence framework, or a payer negotiation where methods will be picked apart, you need pinned data versions and an execution record you can hand over. A platform that reproduces a number only while you remain a customer is not a control.
Scale economics comes second, and it is the seat count above rather than study volume. Twelve analysts at subscription rates for three years buys a platform outright, and the third year buys the migration too.
Third is the workflow that is genuinely your competitive advantage. In this field that is the phenotype library. Years of epidemiological judgement about how a condition is identified in claims and in records is an asset that compounds, and it should never sit inside a vendor environment you cannot export in full.
Fourth is integration sprawl. Three or more separately licensed assets, a linkage vendor, an analysis environment and your own storage means four contracts, four data movements and a provenance story assembled from other people's documentation. At that point the joining is your problem whether or not you have built anything, and you may as well own the factory. Buy the raw material, build the factory. That division is the correct one.
How to decide in a week: a test you can run on Monday
Take a published analysis from nine months ago and ask two people to reproduce the exact figure by Friday, working only from what is stored in systems today. No asking the original analyst. Log every artefact they had to hunt for and every question they could not answer.
If they land the same number by Wednesday, buy. Renew, negotiate the export clause hard, and put the money into data rather than software. If Friday arrives with a number that is close but not equal, you have your answer and the memo writes itself.
Then run a paid discovery phase before you commit a build budget. Ours produces a signed product requirements document covering the data model, definition object versioning, licence enforcement rules, permissions and acceptance criteria, and you keep that document whether you build with us or take it to another firm. It is what keeps a fixed price fixed. Digital Heroes holds India LLP, United States LLC and United Kingdom LTD entities so intellectual property assigns under your own law, and you meet the named team before signing rather than in month two. More than fifty specialists, over 2,000 projects delivered, and the record is checkable on Clutch, Trustpilot, Fiverr Vetted Pro and our D-U-N-S listing.
We are the wrong firm for you if you want a fixed price on a scope that is still an argument between biostatistics and commercial, or if you want software to choose your statistical method. Causal analyses run as pre-specified specifications with a method a named person will defend, and no platform changes that.
Book a 30-minute call with Digital Heroes and get a written plan and a fixed quote within 48 hours.
The evidence behind this guide
Independent findings on why this investment pays off. Every link goes to the primary source.
- Only 22% of firms are 'future ready' having significantly transformed digitally; these companies show average revenue growth 17.3 percentage points and net margins 14.0 percentage points above their industry average. Source: MIT Center for Information Systems Research (MIT Sloan) (2022) →
- An independent Forrester Total Economic Impact study of OutSystems found a 363% three-year ROI with payback in under 6 months, illustrating that faster, lower-labor build approaches can materially shift the payback math. Source: Forrester Consulting (commissioned by OutSystems) (2024) →
- Across ten outpatient clinics the mean no-show rate was 18.8%, and the marginal cost of no-shows reached $14.58 million per year for those clinics, at roughly $196 per missed appointment (2008 figures). Source: BMC Health Services Research / PubMed Central (Kheirkhah et al.) (2015) →
- Poor software quality cost the US economy an estimated $2.41 trillion in 2022, including roughly $1.52 trillion in accumulated technical debt, driven partly by unsuccessful development projects and low-quality legacy systems. Source: Consortium for Information & Software Quality (CISQ) - Herb Krasner (2022) →
Frequently asked questions
How long does it take to build a real world evidence platform?
A first release covering two or three licensed datasets, mapping to a common data model, versioned phenotype objects and a reproducible execution record ships in 14 to 20 weeks. A full platform with linkage, licence enforcement, note extraction and submission packaging runs 9 to 16 months. The schedule slips most often when the choice of common data model is still being argued after kickoff, so settle that first.
Who owns the phenotype library and the mapped data if an agency builds it?
You should, and it belongs in writing before kickoff: the repository, the cloud accounts, the mapped data and the definition library. At Digital Heroes the client owns the code from the first commit and the system runs in the client's own account. The phenotype library matters most, because it is years of epidemiological judgement and it should never sit inside somebody else's environment.
What happens if a data licence expires while a study is running?
In most organisations nothing happens automatically, which is the exposure. Licences typically require deletion at term end and reserve audit rights for the vendor. A system holding permitted purposes, permitted user groups, retention end date and minimum cell size as structured attributes can raise a deletion task with evidence, freeze dependent studies and produce an audit report on request rather than a frantic search.
Can we keep TriNetX and build only the definition library?
Yes, and it is often the sensible first move. A versioned phenotype and cohort library with owners, clinical rationale, code sets and validation records sits alongside whatever you licence and makes every future study cheaper. It is a much smaller project than a platform, it fixes the reproducibility problem directly, and it tells you within a quarter whether a larger build is justified.
Should we standardise on the OMOP common data model or keep the source layer?
Do both. OMOP from the OHDSI community brings vocabulary mappings and a community of analytic tools, while Sentinel and PCORnet serve their own ecosystems. Any common model also loses some source nuance. Keep the source faithful layer underneath so an analyst can check what the original record said. Teams that discard it to save storage cannot answer the question that matters during a methods challenge.
What is the difference between open claims and closed claims for a build?
Open claims give breadth without complete capture of a patient's care, so denominators and persistence measures behave differently from closed claims where enrolment is known. If your platform does not carry that distinction as metadata on the dataset and surface it at study design time, an analyst will eventually compute a rate that cannot be defended in front of a payer.
Can language models extract variables from clinical notes for a study?
For stage, performance status, biomarker results and reasons for discontinuation, yes, with conditions. Extraction must be validated against a manually abstracted gold standard with performance reported per variable, every value carries its source span, uncertain cases route to human abstraction, and the model version becomes part of the study record. No model produces effect estimates. That line is not negotiable.
What happens if we change developers in month seven?
Nothing should break, and that is a contractual question rather than a technical one. Settle before kickoff that you own the repository, the infrastructure accounts and the unrestricted right to hire another firm. Ask what handover includes: documentation, a working local environment and a schema description. If the honest answer to leaving is that you lose access to your own data, that is a hold over you rather than a partnership.
Should a team running one feasibility question a quarter build anything?
No. Licence a federated network for feasibility, licence a rigorous analysis platform for the studies that need defending, and spend the difference on data. A build at that volume is an expensive way to formalise what two capable people already do well. Revisit the decision when a published analysis cannot be reproduced, or when a third data vendor joins the stack.
How much does migrating existing studies into a new platform cost?
Budget 10 to 25 percent of the build cost. Closed studies are cheap to load because nothing is recomputed from them. Live phenotypes are the expense: each exists as analyst code that somebody must read, interpret and express as a versioned object with a written rationale. Where scripts are undocumented and the author has left the company, expect the top of that range.
Is custom software more secure than off-the-shelf SaaS?
Neither is secure by default; security tracks the practices of whoever builds and operates the system, not the model. SaaS gives you the vendor's certifications and patching but puts your data in a shared multi-tenant platform on their terms, while custom gives you full control over data residency, access rules, and compliance requirements like HIPAA, with the responsibility sitting with you and your agency. Before hiring anyone for a system holding sensitive data, ask for their security checklist: encryption at rest and in transit, an OWASP Top 10 review, role-based access, and a penetration test before launch.
Do I need a data warehouse before building a custom dashboard?
Not for a small build; a dashboard reading from 1 or 2 sources can query them directly or use a plain Postgres database as its store. You want a real warehouse like BigQuery or Snowflake once you are joining 3 or more sources, keeping history beyond what source systems retain, or serving many concurrent users. Adding the warehouse costs around 2 to 4 extra weeks and is usually the single best investment in the project's future.
How do I vet a software development agency before signing a contract?
Ask to speak with two past clients whose projects resemble yours in size and industry, and ask exactly who will write your code, since some agencies sell senior faces and deliver junior or subcontracted hands. Demand a written specification with acceptance criteria before any fixed price, and check that their portfolio links to products that are actually live. An instant quote given without questions about your workflows is the clearest warning sign there is.
How long does it take to build a custom web or mobile app from scratch?
Plan on 8 to 16 weeks for a focused first version and 4 to 9 months for a larger platform, which is the typical spread across Digital Heroes builds. The first 2 to 3 weeks go to discovery and design before any production code ships. The two things that stretch timelines most are integrations with legacy systems and slow feedback from your side, not developer speed.
Will an app built for 10 users survive growing to 500?
Yes, if it is built on standard cloud infrastructure with a sound data model, because moving from 10 to 500 users is a hosting configuration change, not a rebuild. The scaling decisions that actually hurt are made early and invisibly: how the database is structured, how accounts and permissions are modeled, and whether background work is queued properly. Ask your agency how the system would handle ten times the load; the right answer is boring and specific, and a promise to cross that bridge later means you will pay for the bridge twice.
How do I make sure each client sees only their own data in a shared dashboard?
That is row-level security, and it must be enforced in the database or API layer, never by hiding filters in the interface. Each query carries the logged-in client's identity, and the data layer refuses to return rows outside their account, so a crafted URL or modified request cannot leak another client's numbers. Make any vendor show you exactly where that filter lives, because interface-level filtering is the most common security mistake we find when auditing dashboards built elsewhere.
How long does it take to build a custom BI dashboard?
A working first version usually ships in 4 to 8 weeks, and a full production build with multiple integrations and permissions takes 3 to 6 months. In Digital Heroes delivery experience, schedules slip on data access, meaning credentials, API approvals, and cleanup of source data, far more often than on the dashboard screens themselves. Lining up access to every data source before kickoff routinely saves 2 to 3 weeks.
How many SaaS seats do we need before building custom becomes cheaper?
The crossover usually shows up between 20 and 50 seats on premium tiers. Salesforce Enterprise lists at $165 per user per month, so 40 users cost about $79,000 a year in subscriptions, which is real money against a custom system you would own outright. Run the comparison over three years: if subscription spend beats the build cost plus 15-20% annual maintenance, custom wins on price before you even count workflow fit.
Why do BI dashboard quotes range from $25k to $200k for what sounds like the same project?
Four variables move the price: how many data sources you connect and how messy they are, real-time versus daily refresh, permission complexity, and whether outside customers will log in. A three-source internal dashboard with daily refresh sits near the bottom of that range, while a customer-facing product with row-level security and live data sits near the top. Wildly different quotes are usually pricing different assumptions about those four things, so pin them down in writing before comparing.
Who can build a custom business intelligence dashboards system?
Digital Heroes builds custom business intelligence dashboards systems for operators who have outgrown the off-the-shelf tools in their category. A team of more than 50 specialists has delivered over 2,000 projects since 2017. Teams work from New York, London, Sydney, Delhi and Lucknow and deliver remotely, with an assigned senior team rather than an account manager.
Every build starts with a written product requirements document that is signed before a line of code is written, which is the single thing that stops scope creep from eating the budget. Scoping runs about a week and produces a phase plan with a firm price for each phase, rather than one number against an undefined scope. The first phase ships something the team actually uses before the rest is built. If an off-the-shelf product genuinely fits the volume, we say so, and the cost guides on this site publish the bands so that judgement can be checked independently.
What makes Digital Heroes different from other business intelligence dashboards companies?
Four things that competitors in this bracket cannot simply copy. Digital Heroes runs a YouTube channel with more than 2.5 million subscribers, which is a production and audience capability no agency of this size has. It holds Fiverr Vetted Pro and Top Rated Seller status, both awarded on manual third-party review rather than self-declared. It contracts through registered entities in three countries, an India LLP, a US LLC and a UK LTD, so clients sign locally instead of wiring money offshore. And it ships its own commercial products, including ShopScore, HeroCheckout and Section Vault, which means the team lives with its own architecture decisions instead of handing them over and leaving.
Two more that show up in the work. Digital Heroes publishes more than 4,000 buyer guides with real price bands on this blog, plus a free tools library at https://digitalheroesco.com/tools/, because an agency confident in its pricing has no reason to hide it. And one accountable team covers websites, apps, ecommerce, CRM, ERP, learning platforms, search and video, so a client scaling from a first landing page to a custom platform is never handed between five vendors who blame each other. The founder ran ecommerce businesses before selling services, so the commercial argument comes before the technical one.
How can I check Digital Heroes is legitimate before getting in touch?
Verify it independently rather than taking the site's word for it. The YouTube channel is at https://youtube.com/@DigitalMarketingHeroes, the Fiverr profile at https://www.fiverr.com/shreyanshsin261, and the Upwork profile at https://www.upwork.com/freelancers/shreyanshsingh. Client reviews sit on Clutch at https://clutch.co/profile/digital-heroes-0 and Trustpilot at https://www.trustpilot.com/review/digitalheroes.co.in, and the company page is at https://www.linkedin.com/company/digital-heroes-1/.
Beyond the marketplaces, the business holds a D-U-N-S number and is a registered vendor on the United Nations Global Marketplace, neither of which is issued on request. Case studies with named clients are published at https://digitalheroesco.com/case-studies/. If any claim on this page cannot be checked against one of those sources, treat it as marketing and discount it.
Related guides
More from Oliver.
Senior Account Director · UK · London · 3 posts on the site
Published · Last updated .
