The four costs after launch
Once software is live, four kinds of cost arrive every month. Some are small, some grow with use, and one is easy to forget until something breaks.
| Cost | What it covers | What makes it grow |
|---|---|---|
| Hosting | Servers, database, storage, backups, monitoring, and our management if we run it | Users, data, uptime requirements, the number of environments |
| Third-party services | Email sending, SMS, maps, payment processing fees, search, analytics, licences | Usage: messages sent, transactions, seats |
| Support and maintenance | Security updates, upgrades, small fixes, monitoring, someone to call | The response time you need, how much changes around the software |
| AI models | Tokens on a frontier API, or GPU servers for an open-source or custom model | Traffic, the length of prompts and answers, the size of model |
We never type running costs into a guide, because provider prices change. The worked examples below show each cost from our rate card today, as monthly ranges.
Hosting: what you are paying for
Hosting is more than a server. A production system needs a database with automated backups you have tested restoring, somewhere to store files, a way to deploy changes safely, monitoring that wakes someone when it fails, and usually a second environment where changes are tried before they reach customers.
- Size follows use. An internal tool for one team runs on very little; a public platform with mobile apps and payments needs more, and more redundancy.
- Managed services cost more per unit than a bare server and much less in people's time. For most clients, a managed database is the right call.
- The provider matters less than people think. DigitalOcean, AWS, Google Cloud, Azure, Hetzner and Vultr can all run a typical app well; pick on data location, the services you need and who will operate it.
- If you host it yourself, the hosting bill is yours and so is the operating work. We can set it up on your account and hand it over.
Third-party services: small lines that add up
Most products lean on a handful of outside services, each billed on its own terms. One by one they look trivial. Together, and as usage grows, they become a real line in the budget.
- Transactional email and SMS: billed per message, so password resets, receipts and alerts all count.
- Payment processing: a fee on every transaction, set by your payment provider, plus fees for disputes and refunds.
- Maps, address lookup and geocoding: usually free up to a limit, then billed per request.
- Error tracking, analytics and uptime monitoring: billed by volume or by seat.
- Licences for fonts, icons, components or data feeds, some yearly, some per user.
We list every third-party service in the handover, with who holds the account and how it is billed, and we bill any we buy on your behalf at cost.
Maintenance is not optional
Software that nobody updates does not stay the same; it decays. The frameworks, libraries, databases and operating systems underneath it keep moving, and security fixes only reach you if someone applies them.
- Databases have support windows. PostgreSQL, for example, supports each major version for 5 years after its first release, and recommends always running the current minor release [1].
- Runtimes do too. Node.js long-term support releases typically receive critical fixes for 30 months, and the project says production applications should only use long-term support releases [2].
- Mobile apps face store deadlines. Google Play requires new apps and updates to target a recent Android version, from August 31, 2026 Android 16 for most apps [3], so an app that is not upgraded cannot ship updates.
- Dependencies publish security advisories every week. Most need nothing; some need a same-day patch.
None of this is a sign of poor software. It is the normal life of any system built on modern tools. The choice is between small, regular upgrades and a large, risky one every few years when something forces it.
A useful habit is a short maintenance review each quarter: which versions are you on, which reach end of life in the next year, and what upgrades are scheduled. It turns surprises into calendar entries.
Support plans: what you are buying
A support plan is a reserved amount of an experienced team's time each month, a promised response time when something goes wrong, and someone who already knows your system. Our plans come in tiers.
| Plan | Suits | What it includes |
|---|---|---|
| Essential | Internal tools and quiet products | Security updates, dependency upgrades, monitoring, and fixes during business hours |
| Standard | Customer-facing products in daily use | All of Essential, more hours for small improvements, and a faster response |
| Priority | Products the business depends on, and AI systems in production | All of Standard, the fastest response, and attention outside business hours |
If your own team will run the software, you may need no plan at all, or only a small one for upgrades and the occasional question. We would rather hand over well than sell you hours you will not use. Support plans and retainers are billed monthly in advance.
AI model costs
AI features add a cost that behaves differently from hosting. On a frontier model through an API, you pay per token, so the bill follows use: quiet months are cheap and a successful launch is not. On an open-source or custom model, you pay for GPU servers, which cost about the same every month whether they are busy or idle.
- Shorter prompts cost less. Send the passages that answer the question, not the whole document library.
- Caching helps. Repeated instructions and documents can often be cached and billed at a lower rate.
- Smaller models for simple steps. Classification and routing rarely need the largest model.
- Limits per user and per day stop one runaway script from becoming a month's bill.
- At steady, high volume, compare the token bill with a served model. There is a point where a GPU server is the cheaper option.
Four products and what they cost to run
The worked examples below show the range from our rate card today, with the build and each monthly cost: hosting, support and, for the AI examples, the model. They are priced when you open the page.
Worked example, priced now
An internal tool
Eight screens for one team, with sign-in, an admin side and file uploads, on small DigitalOcean hosting with the Essential support plan.
- Build
- ≈ US$33,700 to US$51,600, delivered within 13 weeksCAD 48,000 to 73,500
Prices in your currency are estimates from today's Bank of Canada rate. All invoicing is in CAD or USD.
Running it
- Hosting
- ≈ US$176 a monthCAD 250 a month
- Support
- ≈ US$1,000 a monthCAD 1,425 a month
Worked example, priced now
A web and mobile product
A customer-facing web app with iOS and Android apps, payments, integrations, notifications and analytics, on medium AWS hosting with the Standard support plan.
- Build
- ≈ US$108,000 to US$165,000, delivered within 27 weeksCAD 153,600 to 234,900
Prices in your currency are estimates from today's Bank of Canada rate. All invoicing is in CAD or USD.
Running it
- Hosting
- ≈ US$492 a monthCAD 700 a month
- Support
- ≈ US$2,500 a monthCAD 3,565 a month
Worked example, priced now
A busy AI assistant on a frontier model
An assistant with tools and document search, heavily used, on a frontier model through its API, with Priority support. The model line follows traffic.
- Build
- ≈ US$86,100 to US$132,000, delivered within 21 weeksCAD 122,600 to 187,400
Prices in your currency are estimates from today's Bank of Canada rate. All invoicing is in CAD or USD.
Running it
- Hosting
- ≈ US$3,970 a monthCAD 5,655 a month
- Support
- ≈ US$5,010 a monthCAD 7,135 a month
- Model running cost
- ≈ US$3,480 a monthCAD 4,955 a month
Worked example, priced now
The same assistant on an open-source model
The same assistant on an open-weight model served on rented GPUs at AWS, with Priority support. The model line is the GPU server, whatever the traffic.
- Build
- ≈ US$93,100 to US$142,000, delivered within 22 weeksCAD 132,600 to 202,800
Prices in your currency are estimates from today's Bank of Canada rate. All invoicing is in CAD or USD.
Running it
- Hosting
- ≈ US$7,020 a monthCAD 10,000 a month
- Support
- ≈ US$5,010 a monthCAD 7,135 a month
- Model running cost
- ≈ US$6,530 a monthCAD 9,300 a month
Keeping running costs down
- 01
Right-size after launch
Start with headroom, then look at real usage after a month or two and scale down what is idle.
- 02
Turn off what you do not use
Old environments, forgotten storage buckets and test servers are the most common waste.
- 03
Watch third-party usage
Set alerts on email, SMS and API spend. A bug that sends the same message in a loop is expensive.
- 04
Keep up with upgrades
Small, regular upgrades cost less than an emergency migration when a version reaches end of life.
- 05
Measure the AI bill per feature
Know which features use the most tokens or GPU time, and whether they earn it.
Who holds the accounts
Running costs are easier to control when the bills come to you. We recommend that the cloud account, the domain, the app store accounts, the payment provider and every third-party service are opened in your company's name from the start, with us added as users. You see every invoice as it arrives, nothing is held hostage if you change suppliers, and a provider's cost alerts reach someone who can act on them.
Where we do host and manage a system for you, the hosting line in our estimate includes the provider's price plus our management, and the handover documents say exactly how to move it to your own account if you ever want to.
How to budget
Budget the first year in two parts: the build, and twelve months of running costs, including support. Revisit the running costs after three months of real use, when you know your traffic. And keep a small reserve for the unplanned: a security patch that needs testing, a provider that changes its terms, a feature customers ask for in the first month.
When you compare quotes, compare the first two or three years, not the build alone. A cheaper build on an unusual stack, or with no tests, often costs more to run and more to change.