Per-Seat SaaS Fees vs One Custom Tool: Five-Year Costs Compared

Usually not for a team under about a dozen seats. Add hosting, maintenance and model updates (budget about a fifth of the build fee a year) and a small custom tool costs two to two and a half times its quote over five years. Per-seat software stays cheaper until seat count, per-seat price or yearly rises push past that.

Run the sum below with your own figures and you'll have your break-even seat count within an hour. The year-by-year example, three scenarios and the hidden costs on each side show which assumptions matter most, and the middle routes near the end often beat both extremes.

Follow me on Instagram@sagnikteaches

The two sums, side by side

Both sides are simple once you write them down. The mistake is comparing a monthly subscription with a one-off build fee, which makes custom look cheap after about two years. Compare five-year totals instead.

Connect on LinkedInSagnik Bhattacharya

FIVE-YEAR SaaS COST
  seats x price per seat per month x 12      = year-one cost
  year-one cost x rise factor                = five-year cost
     rise factor: 5.00 at 0% a year, 5.53 at 5%, 5.75 at 7%, 6.11 at 10%
  + setup and migration time (hours x your hourly value)

FIVE-YEAR CUSTOM COST
  build fee (the quote)
  + 5 x yearly maintenance        (assume 20% of build fee unless quoted)
  + 5 x yearly hosting, database and backups
  + 5 x yearly AI usage           (API tokens, if the tool calls a model)
  + model-change re-tests         (assume one every two to three years)
  + your time to specify and test (hours x your hourly value)

The 20% maintenance figure is an assumption, not a law. Ask any developer quoting you what they'd charge per year to keep the tool patched, backed up and working, and use their number. If they won't give one, use 20% and treat the silence as a warning.

Subscribe on YouTube@codingliquids

Worked example: an independent bookshop, year by year

Say an independent bookshop with nine staff pays for an order-management and customer-messaging tool with AI features built in: $39 a seat a month, so $351 a month or $4,212 a year. The owner has a quote of $14,000 to build a custom tool that does the same core jobs (customer orders, "your book has arrived" messages, AI-drafted replies to stock enquiries) and nothing else. All figures here are illustrative.

Assumptions for the custom tool: maintenance at 20% of the build ($2,800 a year), hosting, database and backups at $45 a month ($540 a year), AI usage at about $25 a month ($300 a year), a $600 re-test each time the AI model it uses is retired, and 30 hours of the owner's time during the build at $50 an hour ($1,500). For the SaaS, a 7% price rise each year from year two.

The custom tool is cheaper to run from year two onwards, which is why it can look like the obvious choice. It never catches up inside five years at nine seats, though. The break-even point is where the custom total equals the per-seat total: one seat costs $39 x 12 x 5.75 = about $2,691 over five years, and $34,900 divided by $2,691 is about 13. On these assumptions, custom starts to win at around 13 seats, before you've priced in any of the risks below.

A 7% yearly rise isn't pessimistic. Microsoft raised Microsoft 365 Business Basic from $6 to $7 and Business Standard from $12.50 to $14 per user per month from 1 July 2026, at each customer's next renewal: rises of about 17% and 12% in one step. Your five-year sum is sensitive to this figure, so run it at 5% and 10% too.

Three scenarios, small to large

The same sums for three illustrative businesses, with a 7% yearly SaaS rise and 20% custom maintenance. Hosting and AI usage scale with the size of the tool.

Most businesses reading this sit in the first two rows. For them, the more useful question is usually how to pay for fewer seats, not how to replace the software.

Getting honest numbers for your own sum

An afternoon with your invoices and one email to a developer is enough. Work through these in order.

1. Pull 12 months of invoices for the software. Note the seat count each month, not just today's. Seasonal businesses often pay for peak seats all year.
2. Find the vendor's price history. Search your inbox for its price-change emails over the last three years and work out the average yearly rise. Use that as your rise factor instead of guessing.
3. List what you actually use. If the team uses four of the tool's twenty features, the custom tool only has to do four things, which keeps its quote small. If you use fifteen, the build gets expensive fast.
4. Ask for maintenance in writing. Any developer quoting a build should also quote a yearly figure for patching, backups, monitoring and small fixes, plus an hourly rate for anything beyond that.
5. Estimate AI usage from volume. Count how many times a month the tool would call an AI model and roughly how much text goes in and out each time. One million tokens is roughly 750,000 English words.

Steps 1 and 2 often change the answer before you reach a developer. The bookshop's inbox, for instance, might show the seat price going from $32 to $34 in 2024, to $36 in 2025 and to $39 this year: rises of 6.3%, 5.9% and 8.3%, which average out at about 7%. That's the figure the example uses, and it came from the vendor's own emails rather than a guess. The invoices can be just as revealing. Here is an illustrative outdoor activity centre that pays for 14 seats every month because nobody removes the summer staff:

At $30 a seat, that's $3,600 of seats needed against $5,040 paid: $1,440 a year, or over $7,000 across five years, before anyone has asked a developer for a quote. Whether you can drop seats mid-year depends on the billing: monthly plans usually allow it, while annual commitments often only let you reduce at renewal, so check your vendor's terms before you plan on the saving.

The AI line is usually smaller than owners expect. In the bookshop example, say the tool drafts 1,200 replies a month, each reading about 3,000 tokens (the enquiry, the stock record and its instructions) and writing about 400. That's 3.6 million input and 0.48 million output tokens. On Claude Sonnet 5 at $2 input and $10 output per million, it comes to about $12 a month; on the larger Claude Opus 5.5 at $4 and $20, about $24. The $25 in the example leaves headroom. Build and maintenance, not the AI, decide which side wins.

What the custom quote leaves out

• AI model retirement. A custom tool that calls an AI model through an API depends on that model staying available. Model providers retire older versions: Anthropic's model deprecations page promises at least 60 days' notice, and Claude Sonnet 4, released in May 2025, was retired on 15 June 2026. Each switch means re-testing the tool on your real cases, and sometimes rewriting its instructions. Here's how that goes wrong in practice, as an illustration: a small printing firm's quoting tool, built in the summer of 2025, called Claude Sonnet 4 by its exact model name. The API account was registered to the developer's email, so the retirement notices went to someone who had since moved on. From 15 June 2026 every request failed, staff saw an empty draft box, and quotes went back to being written by hand for nine working days while a new developer found the cause, switched the model and re-tested 40 past quotes. The cost was a few hundred dollars of developer time; the lesson was to put the API account and its notification email in the business's name on day one.
• One developer, one point of failure. If the person who built it moves on, the next developer has to learn it before fixing anything. Documentation and a code repository in your name reduce this; they don't remove it.
• Security and updates. Custom software is built from other software components that need patching. Somebody has to do it, and somebody has to notice when it hasn't been done.
• Every new feature is a new invoice. A SaaS vendor adds features across all customers. With a custom tool, the card-payment link you want next year is a fresh quote.
• Data responsibility moves to you. Customer records on your own hosting are your responsibility to secure and back up. Ask your data-protection adviser what that changes for you.
• Your own time. Budget 20 to 60 hours of the owner's or a senior person's time to specify and test even a small build; the example above assumes 30. That is time off the shop floor.

What the per-seat price leaves out

• Price rises, as above, usually announced with a few weeks' notice and applied at renewal. What to do when software raises its price for AI covers keep, downgrade or switch.
• Seat creep. Seasonal staff and ex-employees whose seats nobody removed. A quarterly seat review often pays for itself in one sitting.
• Tier gating. The one feature you need lives on the next plan up, priced per seat for everyone.
• AI add-ons and credits. Many tools now sell AI separately, either per seat or as a pool of credits that runs out. An AI add-on such as Microsoft 365 Copilot Business at $21 per user per month on annual billing can cost more than the base plan it sits on. Per-seat versus usage-based AI pricing explains which model suits which pattern of use.
• Lock-in. If exporting your records is awkward, the switching cost rises every year you stay.

Cheaper middle routes worth pricing first

Pay for fewer seats. In many teams, three or four people use the software heavily and the rest open it once a week. Some tools offer cheaper light or view-only seats for occasional users; others let several people work from one shared inbox without a seat each. Deciding who actually needs a paid licence walks through the seat-by-seat review.

Keep the SaaS and build a small layer on top. If the software has an API or a connector on Zapier or Make, a small automation can add the missing step (AI-drafted replies, a report, a sync with another system) for a fraction of a full custom build. This keeps the vendor responsible for the core and you responsible only for the layer. No-code AI tools versus a custom build shows where templates stop coping.

To see the difference in numbers, take an illustrative six-person property maintenance firm with a $12,000 quote for a custom job tracker. The only thing its current job-management software can't do is send customers a plain-English update when a job's status changes. A Make scenario can watch that status field, have an AI model draft the update and put it in the office inbox for approval. Make starts from about $9 a month. The AI part is tiny: 300 updates a month at about 1,500 tokens in and 200 out comes to roughly $0.75 a month on Claude Haiku 4.5 at $1 input and $5 output per million tokens. Add, say, $800 for a freelancer to set it up and $300 a year for a check-up, and the five-year total is under $3,000. The custom tracker, at two to two and a half times its quote, would be $24,000 to $30,000.

Drop to a cheaper tier and fill the gap. If you're on a premium plan for one AI feature, check whether a general assistant you already pay for can do that job instead.

When custom is the right call

Custom is worth serious consideration when most of these are true:

• You have 13 or more seats (or whatever your own break-even says) and expect to grow.
• Your process is genuinely unusual, and staff spend real time on workarounds because the software assumes a different way of working.
• The tool's scope is small and stable: a few screens, a few jobs, not a whole business system.
• You have a developer relationship you trust, with maintenance priced in writing.
• The code, hosting and AI accounts will all be in your business's name.

If two or more of those are missing, stay with per-seat software and work on the seat count. For the wider question of building versus buying across the business, whether to build or buy a custom AI tool goes further than cost alone.

Questions owners ask before commissioning a custom tool

Can I build the custom tool myself with an AI coding assistant?

For a small internal tool that only you use, sometimes. The build is the easy part; the hard part is hosting it securely, backing up its data, fixing it when an update breaks something and handing it to someone else if you're away. If customers or staff depend on it daily, budget for a developer to review what you've built and to be on call, and put that cost in the five-year sum.

Who owns the code if a freelancer builds it?

Only what your contract says. Ask for a written assignment of the code and documents to your business, the code stored in a repository your business owns, and all hosting and AI accounts registered to your company email. Without those, you may be paying per seat again in a different form: a developer you can't replace.

What happens if the SaaS vendor is bought or shuts down?

You may face a forced migration at a time you didn't choose. Before committing to either path, check that the software lets you export all your records in a standard format such as CSV, and test the export once a year. That risk is real but usually smaller than the key-person risk of a custom tool with one developer behind it.

Further reads

• AI Vendor Lock-In: How to Keep Your Data and Prompts Portable — How to keep your data and prompts movable on either path.
• How Much Does a Custom AI Assistant Cost a Small Business? — Build costs for a custom AI assistant in more detail.
• Annual or Monthly Billing for AI Tools: Which Saves You More? — Whether committing annually cuts the SaaS side of the sum.
• How to Negotiate AI Software Prices and Seat Numbers — Get the per-seat price down before you consider replacing it.
• How to Brief a Developer on a Custom AI Workflow You Need — If custom wins, how to brief the developer properly.
• Power Apps for Small Businesses: When It Replaces Custom Software — A low-code middle route between SaaS and custom code.
• What Is an API? Why It Matters When You Buy Software — A plain-English explanation of APIs for people buying software, with the vendor questions and pricing traps that decide whether tools can connect.
• How Much Does AI Cost a Small Architecture Practice per Month? — Priced line by line: assistants, model-aware rendering, meeting notes and automation, with monthly budgets for one, five and twelve-person practices.
• Use the AI Already in Your Practice Software Before Buying More — Five stages to find and test the AI in your practice management, accounting and office software, then buy only for the gaps it leaves.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Microsoft 365 business plan pricing and 2026 price change notices; Microsoft 365 Copilot pricing page; Anthropic API pricing and model deprecations pages (checked September 2026).