Twelve tools, one budget: what an SMB actually uses

1. The bill: twelve published prices in one table

The reference company below is deliberately unremarkable: eight people, of whom four work leads, a mailing list, five social channels, a website, and the usual pile of tools that grew rather than got chosen. Every price in the table was taken from the vendor's own pricing page on 30 August 2026. Where a vendor publishes several tiers, the table takes the cheapest one that a company of this size can actually run on, and it assumes annual prepayment everywhere the vendor offers a discount for it.

Thirty thousand dollars over three years, for eight people. That is $1,279 per person per year — before anyone has been paid, before hosting, before advertising. It is roughly the price of one decent used van, or a fortnight of a good developer, spent in twelve monthly instalments so small that none of them ever reaches a decision meeting.

That is the whole mechanism. Nothing in this list was bought carelessly. Each one was, at the moment of signing, obviously worth twenty dollars.

2. Every figure here is a floor

The table is the best case the vendors themselves advertise. Three things push a real invoice above it, and all three are ordinary rather than exceptional.

• The entry tier is priced for a company smaller than yours. Mailchimp's $20 Standard plan is the price at 500 contacts. The plan is sold in tiers by contact count, and the tier moves as the list grows — which is, after all, the point of having the list. Brevo works the same way: Starter from $9 and Standard from $18, priced by monthly send volume, with three seats included at Standard. Supermetrics Starter allows three data sources; a fourth channel means the $177 Growth plan. Nothing here is a trap. It is simply that the advertised number is the number for the customer you were two years ago.
• Annual prepayment is assumed throughout. On the six lines where the vendor publishes both prices, paying monthly instead costs $155.40 more a month — $5,594 over three years. Trello is $10 annually and $12.50 monthly, ChatGPT Business $20 against $25, Google Workspace $14 against $16.80, HubSpot Starter $7 against $20. Cash-flow reasons for paying monthly are perfectly good reasons. They are also an eighteen per cent surcharge on the whole stack.
• The upgrade step is where the pricing actually lives. HubSpot Sales Hub costs $7 per seat at Starter and $90 at Professional — nearly a thirteenfold step — with a mandatory one-off onboarding fee of $1,500 at the Professional tier and $3,500 at Enterprise. Take that one step with four seats and the same stack runs to $44,148 over three years instead of $30,696. The entry price is not the offer. It is the doorway.

And one line cannot be filled in at all. Bynder — the market's reference name for digital asset management — publishes no price. Its own guide to DAM costs contains no figures and directs the reader to request a quote, on the reasoning that cost depends on the size and needs of the team. That is category-normal rather than unusual; it simply means a budget comparison in this line is impossible before a sales call has happened. In the table above, the most expensive item is very probably the one shown as a dash.

3. Four things you are paying for twice

Read the table as a list of jobs rather than a list of brands, and the same job keeps appearing. This is not carelessness — it is what happens when tools arrive one at a time, each solving a problem that was real on the day it was bought.

Those four rows come to $598.66 a month of the $852.66 — seventy per cent of the bill sits on jobs that at least two tools in the house believe they are doing. Not all of it is waste: Notion and Trello genuinely do different things once a team has grown past a certain size, and a CRM's transactional mail is not a newsletter tool. But the honest test is not whether a difference exists. It is whether anyone in the building could name it.

The image row is the one worth staring at. In a company of this size a photograph typically exists in three places: in the shared drive, inside the design tool where somebody made a post out of it, and in the media library that was bought precisely because the first two were a mess. Three copies, three subscriptions, and still nobody can answer whether the licence on that photo has expired.

4. The seat is the unit — the use is not

Seven of the twelve lines are priced per seat, per user, per member or per channel — and those seven carry $756.67 of the $852.66. Not one of them is priced per use. That is the single most consequential fact about the whole table, and it is the reason the bill grows even in a quarter when nobody changes anything.

Work through the reference company honestly. Eight Notion seats, in a house where the plan is written by two people and read by the rest — $160 a month for two authors and six readers. Eight Canva seats, where three people make anything and the others opened it once to resize a logo. Eight ChatGPT seats, of which a couple are genuinely daily and a couple were provisioned during onboarding and never signed in again.

The exercise that follows from this takes twenty minutes and is uncomfortable in a specific way: it is not a vendor problem. Every one of these products has an admin screen that reports last activity. Open each of them, write down two numbers — seats paid, seats used in the last thirty days — and the gap is your answer. In most companies of this size it is somewhere between a quarter and a third of the per-seat lines.

The second-order effect matters more than the money. A tool that five people use is a tool; a tool that eight people have and three use is a place where things get lost. The plan is in Notion, except the bits in Trello. The image is in the drive, except the version that went out, which is in Canva. Paying for reach you do not have is not just expensive — it is how a company ends up with four answers to the question of where something is.

5. The billing unit moves underneath you

A subscription price is a promise about a number. It is not a promise about what that number counts, and 2026 has been a good year for watching the difference.

Make states on its pricing page, as at 30 August 2026, that as of 27 August credits are the billing unit. Its help centre documents the accompanying arithmetic: extra credits, whether bought manually or drawn automatically, cost twenty-five per cent more than the credits included in a plan. The Core plan is capped at 300,000 credits a month, the Pro plan reaches eight million. Zapier moved in the same direction on 15 June 2026, replacing a flat rate for AI steps with model tiers: a run costs 1x on Standard, 3x on Advanced — the default — and 5x on Premium, with tool calls charged at the same multiplier on top. A per-step ceiling of 75 tasks in a single run pauses the automation for approval.

Neither of these is a headline price rise, and neither will show up in a spreadsheet that tracks monthly plan cost. Both change what an unchanged month of unchanged work costs. That is the property that makes automation platforms difficult to budget: the invoice is a function of your volume multiplied by a rate the vendor sets, and only one of those two numbers is yours.

The same shape appears wherever pricing is metered rather than flat. HubSpot's Starter tier is priced per seat but its contact tiers are priced per contact. Mailchimp's tier moves with the list. Supermetrics and Windsor.ai both price by connected data source — $44 against $19 a month at the entry tier for three sources, which is a real and checkable difference, and the reason that particular comparison is the most-read page we have. None of this is hidden. It is simply that nobody re-reads a pricing page for a tool they already have.

6. What is genuinely not worth rebuilding

An article that ends “build everything yourself” would be a bad article, and it would be wrong. Some of the twelve lines are excellent value and would be absurd to replicate.

• Office, mail and identity. $14 a head for mail delivery, calendars, storage, and — more to the point — an identity system every other tool can log in against. Running your own mail server in 2026 means running your own deliverability reputation. Nobody should want that job.
• The frontier model. $20 a seat buys a share of a training run that cost more than the company will earn in its lifetime. This is the clearest example in the whole table of a subscription that is dramatically underpriced relative to what it does.
• Anything holding regulated money. Payment processing, payroll, accounting, tax filing. The subscription is not what you are buying; the certification and the liability are.
• Genuine third-party reach. A scheduling tool's value is the platform APIs it maintains against, and those change under it constantly. That maintenance is the product.

Add the design tool to that list — nobody sane rebuilds Canva — and $488.67 of the $852.66 buys something you could not make yourself. On those lines the lever is the seat count, not the vendor. The interesting part is the remaining $364.

7. What building it yourself actually costs

Full disclosure, because it belongs at the top of this section rather than buried at the bottom: we build these systems, and we run our own on them. This piece is the opening of a series that compares each of the middle-layer categories in turn — mail, CRM, image library, editorial plan, automation — and in each of them we are one of the options being compared. Treat the arithmetic accordingly, and check it.

The middle layer is the part between the platform and the frontier model: a place to keep contacts and deals, a way to send a mailing with proper consent handling, somewhere to keep images with their licences attached, a plan that records what actually went out. In the reference stack that layer is $363.99 a month — $13,104 over three years, and that is before the unpriced media-library line is filled in. These are not hard pieces of software. They are small, extremely well-understood, and their price is set by how many people touch them rather than by how difficult they were to write.

What a build costs instead is specific and worth stating plainly. Hosting for the sort of small PHP-and-SQLite application these jobs actually need runs at single-digit euros a month on ordinary shared hosting. The build itself is real work — weeks, not hours — but it is one-off, and with current AI assistance it is a fraction of what the same work cost three years ago. Maintenance is the honest liability: something has to be patched, backed up and looked at when it breaks, and a system nobody maintains is worse than a subscription.

So the rule is not “build everything”. It is narrower and it holds up: build the pieces whose price is driven by seat count rather than by difficulty, and only where you can name who maintains them. A CRM for four people is a table of contacts, a table of deals and a form. Paying per seat for that, forever, is paying rent on arithmetic. A mailing tool is a list, a consent record and a send queue — and the consent record, which is the only genuinely tricky part, is exactly the part you most want to own outright under the GDPR.

The wider argument sits in our piece on the AI stack for a small business, and the CRM case specifically in the total cost of ownership against HubSpot and in do you really need Salesforce.

8. Verdict: the afternoon that settles it

The number at the top of this article is not an argument. It is a measurement, and the only version of it that matters is yours. Producing it takes an afternoon and needs no tools at all.

• Start from the bank statement, not the tool list. The tool list contains what you remember. The statement contains what you pay. Take one ordinary month and write down every recurring charge.
• Put the unit next to each price. Per seat, per contact, per channel, per credit, per source. This is the column that tells you which lines will grow on their own.
• Multiply by 36. Not by 12. Three years is the horizon on which a build competes with a subscription, and it is the horizon on which the total stops looking small.
• Add a column for seats used last month. From each tool's own admin screen. This column, not the price column, is where the first saving is.
• Group by job, not by vendor. Where two rows describe the same job, one of them is a decision waiting to be made.

Most companies that do this find the same three things: they are paying for more seats than they use, they are paying twice for two or three jobs, and the total over three years is between two and four times what anyone would have guessed. None of that is a scandal. It is the predictable result of buying software the way it is sold — one small, obviously reasonable decision at a time.

The comparisons that follow this piece take the categories one at a time, with the same arithmetic and the same disclosure: email marketing, CRM, the image library, the editorial plan and automation. Each one is a comparison rather than an advertisement, because a recommendation without the counter-argument is worth nothing to a reader deciding how to spend thirty thousand dollars.

9. FAQ: subscription costs in the SMB

What does a normal software stack cost a small business over three years?

For a company of eight people running twelve common tools at the vendors' own published entry prices, the figure is $852.66 a month, $10,231.88 a year and $30,695.64 over three years — about $1,279 per person per year. That is the floor, not the bill: it assumes annual prepayment everywhere, the smallest published tier of every plan, and it carries one of the twelve tools at zero because its vendor publishes no price at all.

Why is the real invoice higher than the list price?

Three reasons. The entry tier is priced for a company smaller than yours: Mailchimp's $20 covers 500 contacts, and the price rises with the contact count. Annual prepayment is assumed throughout; paying monthly on the six lines where both prices are published adds $155.40 a month, $5,594 over three years. And the step most companies eventually take is priced steeply: HubSpot Sales Hub goes from $7 a seat to $90, plus a mandatory one-off onboarding fee of $1,500 — which turns the same stack into $44,148 over three years.

Which tools in a typical SMB stack duplicate each other?

Four pairs, reliably. A wiki and a task board both hold the plan. Two automation tools both move data between the same systems. A CRM with email built in and a separate newsletter tool both own the mailing list. And images end up in three places at once: the office storage, the design tool's own library and a dedicated media library. In the reference stack those four jobs account for $598.66 of the $852.66 monthly bill.

Does Bynder publish a price for its digital asset management software?

No. Bynder's own guide to DAM pricing carries no figures and asks the reader to request a quote, stating that costs depend on the size and needs of the team. That is normal in this category rather than unusual. For a budget comparison it means one line of the stack cannot be filled in at all before a sales conversation has taken place.

What changed in Make and Zapier pricing in 2026?

Both moved the unit they bill you in. Make's pricing page states that as of 27 August credits are the billing unit, and its help centre documents extra credits costing 25 per cent more than the credits included in a plan, with Core capped at 300,000 credits a month and Pro reaching eight million. Zapier introduced model-tier pricing for AI steps on 15 June 2026: a run costs 1x on Standard, 3x on Advanced and 5x on Premium, with tool calls added at the same rate. Neither is a headline price rise. Both change what a given month of the same work costs.

Is building your own software cheaper than subscribing?

Not everywhere. Office, mail and identity are not worth rebuilding, and neither is a frontier AI model — $20 a seat for the latter is dramatically underpriced relative to what it does. The middle layer is different: a CRM that holds contacts and deals, a mailing tool that manages consent and sends, an image library, an editorial plan. These are small, well-understood pieces of software whose subscription price is driven by seat count rather than by difficulty. In the reference stack that middle layer is $363.99 a month, $13,104 over three years, before the unpriced media-library line is filled in. Hosting for the equivalent self-built system runs at single-digit euros a month; the build is one-off work and the maintenance is the real liability.

Which single change saves the most, fastest?

Removing seats nobody opened last month. Every one of these products reports last activity in its own admin screen. On the reference stack, cutting Notion, Canva and ChatGPT from eight seats to five saves $182.50 a month and $6,570 over three years — with no migration, no data export and no vendor conversation. It beats any discount you are likely to negotiate.

How do I work out my own three-year figure?

Take the card and bank statements for one ordinary month rather than the tool list — the tool list omits whatever nobody remembers. Write down every recurring charge with its unit: per seat, per contact, per channel, per credit. Multiply by 36, because three years is the horizon on which a build competes with a subscription. Then add a column for how many people opened each tool last month, and group the rows by job rather than by vendor. The gap between seats paid and seats used is usually the largest single number in the exercise.