Key points
- Across 1,471 IT projects, the average cost overrun was 27%, and one project in six overran its cost by 200% on average.1
- A later study of 5,392 IT projects found overruns follow a power law, with a fat tail of extreme overruns that averages hide.2
- A peer-reviewed NeurIPS paper found real-world machine learning systems commonly carry massive ongoing maintenance costs.3
- Building the group's own pathways on a live platform, with the client owning what is built, puts the first year into pathway work. wai builds custom clinical AI this way for hospitals, groups and governments.5
Summary
Build the group's own pathways on a live clinical platform, with ownership of the data, configuration and workflows written into the contract. The group gets a system of its own in months, and the overrun and maintenance risk of building from nothing stays off its books.
The group runs clinical AI of its own on its first pathway within the year, and owns what was built. The deciding factors are overrun risk and who maintains the system in year three.
See it in your clinicThirty minutes on your own patients and doctors.
Request a demoSee a care pathway running end to end on wai, from booking to signed letter, and the parts a group would make its own.
| Factor | Build from nothing | Build on a live platform |
|---|---|---|
| Time to the first patient | Longest, foundations first | Short, pathway work only |
| Cost overrun exposure | Full exposure to the IT project tail | Bounded by a fixed first build fee |
| Maintenance | Permanent in-house team | Platform maintained, group owns its pathways |
| Governance documents | Group writes and keeps all of them | Platform documents plus pathway-specific ones |
| One patient record | If the group builds it | Yes, every surface on the same record |
| Ownership | Full | Agreed in the contract, data and configuration with the client |
| Exit | Not needed | Portability and exit terms agreed before the build |
Own the pathways and the data
Speech recognition and drafting models improve every month, so the value for a health system lies in owning its pathways and its data, with the right to leave written in. The table below sets building from nothing beside building on a live platform, factor by factor.
What building costs
In a sample of 1,471 IT projects analysed by Bent Flyvbjerg and Alexander Budzier, the average cost overrun was 27%, while one project in six overran by 200% on average and ran almost 70% late.1
A study of 5,392 IT projects in the Journal of Management Information Systems found that overruns follow a power-law distribution, many small overruns and a fat tail of extreme ones, traced to interdependencies between technical components.2 A clinical AI system has many such parts, from audio and transcription to the record, message channels, payments and the clinician's signature.
Budget for the tail. Ask the engineering lead how the budget holds if this is the one project in six.
Maintenance is the larger bill
In a peer-reviewed paper presented at NeurIPS in 2015, D. Sculley and colleagues argued that it is common to incur massive ongoing maintenance costs in real-world machine learning systems, from data dependencies and configuration to changes in the outside world.3
Guidelines and regulator rules change every year, so a build from nothing carries a permanent engineering team to keep pace.
Governance comes with the code
The organisation that builds clinical software takes on the developer's duties. It writes the intended use, keeps the clinical safety case, assesses data protection risk, logs every draft and approval, and monitors performance after go-live.
On a platform the group adds the documents specific to its pathway. Ask any partner for the documents behind each duty, using the evidence pack to request from a clinic AI vendor.
Time to the first patient
In Deloitte's 2026 outlook, based on a survey of 180 health system executives in six countries, about 30% ran generative AI at scale in select areas and 2% had deployed AI across the whole enterprise.4
A build on a live platform spends its first year on the group's own pathways, where the clinical and commercial value sits, with the patient record and the audit log already in place.
Your own system on a platform already running
wai builds custom clinical AI for hospitals, clinic groups and governments on its live platform, around the client's workflows, and the client owns what is built.5 Data, workflow, configuration and model rights, portability and exit terms are agreed in the contract before any build starts.5 The group starts from one pathway with measurable outcomes, agrees safety and acceptance gates before live use, and adds the next pathway once the first is proven.5
The commercial shape is a fixed fee for the first build and then a monthly fee per site, easy to set against a build budget.5 More detail is in custom healthcare AI for hospitals and governments.
Where to start
Pick one pathway with a named clinical lead and two or three measures that matter, for example time from referral to first appointment and time from consultation to signed letter. Cost each option against it over five years, with an allowance for the overrun tail and a line for maintenance staff.
Write the ownership terms first. Whichever option wins, the group holds its data, its configuration and a tested export, as set out in the clinic software exit plan. The group then has a system of its own, running for patients on its first pathway, with the rights to it in writing.
Bring one pathway and its numbers, and see the first build and what the group would own.
Frequently asked questions
Is it cheaper to build clinical AI in house?
When does building from nothing make sense?
When the group already has a large engineering and clinical safety team, a need no platform meets, and a board that accepts the overrun tail.
Who owns a system wai builds for us?
The client. Data, workflow, configuration and model rights, portability and exit terms are agreed in the contract before the build.5
How is a custom build with wai priced?
The first build carries a fixed fee and each site a monthly fee after that, with clinician seats and payments covered in one agreement.5 Product prices are published on the pricing page.
Where are the records hosted?
wai hosts each clinic's records in its own country, under that country's rules. We set this up with your clinic.
Who approves what the AI drafts in a custom build?
The clinic. Clinical documents are drafted for the clinician to review and sign, and patient messages are drafted for the team to approve, with the approval rules for each action agreed before live use.5
Sources
- 1Harvard Business Review, Flyvbjerg and Budzier, 2011Why Your IT Project Might Be Riskier Than You Think (arXiv copy of the Harvard Business Review article)Read 14 October 2026Read source
- 2Journal of Management Information Systems, Flyvbjerg and colleagues, 2022The Empirical Reality of IT Project Cost Overruns, Discovering a Power-Law DistributionRead 14 October 2026Read source
- 3Advances in Neural Information Processing Systems 28, Sculley and colleagues, 2015Hidden Technical Debt in Machine Learning SystemsRead 14 October 2026Read source
- 4Deloitte, 20252026 Global Health Care OutlookRead 14 October 2026Read source
- 5wai, 2026wai for clinic groups and chainsRead 14 October 2026Read source
Clinical AI of your own, live for patients and owned by the group.
Fifteen minutes on one of your pathways with the people who build on wai, and a reply to the group within one working day.
Partner with us
See it on your own patients, then build it with us
Thirty minutes on your own patients, doctors and rules, in your market. We reply within one working day, and the clinics that partner with us early shape what we build next.
See what wai does between the visits