One quote says $80, the next says $8,000: pricing an automation fix

Two quotes for the same broken Zap

On one screen, a business owner posts the job: find where the automation is failing, fix it, send a short note on what caused it, all inside 24 hours. Budget: $15, fixed price. On another, a Zapier Platinum partner sells a 75-minute implementation session for $297, and the certified partners at the top of the same directory won't take a project for less than $10,000.

Same broken thing. Two prices that don't share a decimal point.

The instinct is to assume one side is a rip-off and the other is a bargain, then to split the difference. That instinct is what gets people burned, because the gap isn't markup. The two prices are quoting two different pieces of work, and only one of them includes finding out what's actually wrong.

Why is the gap so wide?

Because there is no single market. There are two, they use the same words, and almost nobody sells between them. Counting live fixed-price jobs on Upwork on 15 August 2026 gives a shape that looks nothing like a price range:

Source: Upwork job search, counted 15 August 2026, logged out, US results. Job boards move daily and Upwork personalises by location, so treat this as a snapshot of a shape rather than fixed numbers.

On the n8n query the $500-$1,000 band is empty. Not thin: empty. Zapier's middle is nearly as thin, at 8 posts across both middle bands against 27 below $100. Make is the exception, with a real $100-$500 tier. Where the middle is missing, someone collecting three quotes isn't sampling a range, they're sampling two economies that advertise under the same heading.

The certified channel shows the same split from the other side. Zapier publishes what each partner will take as a minimum, and those minimums run from $100 to $10,000 inside the vendor's own approved directory - a hundredfold spread, with the budget brackets running from $100-$250 at one end to $10,001-$200,000 at the other. Make's partner directory publishes tiers and regions but no prices at all, so on that side a buyer can't see the floor before making contact.

What does $15 actually buy?

An attempt. At that price nobody can afford to look for the cause, so what gets sold is a change that makes the symptom stop, and the risk stays with you.

Read what the $15 job actually asked for: diagnosis of where it fails, a repair, and a written note on the cause, delivered within a day. That is the full shape of a professional fix. What fits that price is checking the two or three most common causes and stopping when the error clears.

Sometimes that works. When it doesn't, the second bill is the real one:

"Our simple 14 days bot took over 70 days to complete and still doesnt work. Theres daily bugs and issues, doesnt pull info we agreed on and crashes or stops working all the time."

— one buyer's review of a freelance-marketplace order, January 2026

The 2026 twist is that the cheap end now advertises what it is. One live posting offers $5-$10 an hour and says plainly: "We are not looking for a high-cost senior software engineer", asking instead for skill at prompting AI tools to debug errors quickly. That is a coherent product. It is simply not the same product as a diagnosis, and it is worth knowing which one you're buying.

What is the four-figure quote paying for?

Mostly for finding the cause, which is the part nobody sees and everybody underestimates. The hard part of a broken integration is rarely the repair. It's reproducing the failure well enough to be sure you've found it, and that is where most of the paid time goes. Anyone who has debugged an integration will recognise the shape.

Four things move that number, and a buyer can recognise all of them:

1. Whether the failure reproduces. An automation that broke once at 3am, in a step that leaves no trace, costs more than one failing every run. Zapier guarantees no more than 60 days of run history, and at most 10,000 runs — a ceiling, not a promise — so a problem reported late may not be diagnosable at any price.
2. Access. Every hour spent waiting for credentials is an hour someone is scheduled to work and can't. Shops price that risk in, or they discover it mid-job and raise the invoice.
3. Whether anyone documented it. A workflow whose builder has left, with no notes and no naming convention, has to be read before it can be fixed.
4. What it wrote while it was broken. This is the one people forget. A Zap stuck in a loop spends the month in successful runs, and every one of them wrote something. Cleaning up duplicated records can be larger than the repair.

The costs nobody puts in the quote

Three of them, and none appear on the invoice you're comparing. The platform keeps charging while the thing is broken, the meter runs differently on every platform, and urgency is priced separately.

Start with the meter, because the same outage produces three different bills. Make charges a credit for every trigger check: "Trigger modules use 1 credit per run, regardless of whether they return data" (docs). Zapier doesn't bill failed steps, but re-running a fix does: "Any successful steps will count towards your task usage, even if they were already counted in a previous run" (docs). n8n is the most forgiving of the three: a polling trigger "only counts when it finds new data", and error-workflow runs don't count at all (docs).

Then urgency, which can be the whole difference between two quotes. One WordPress repair shop publishes the identical task at $99 within 24 hours, $149 within 12, and $299 within 4 - a threefold spread from response time alone, from one vendor, on one page.

Then the shape of the engagement. Some shops bill a discovery audit before quoting the work, typically crediting it back if you proceed: £1,500 at one agency, $500 and up at another which refunds it outright if the diagnosis shows automation isn't your bottleneck. That is not a trick, and in the Zapier channel it's the exception rather than the rule — the directory has a field for it, and most listed partners tick it. But a "free" quote from a shop with a $2,500 minimum and a paid audit from a shop with no minimum can easily end up the same price.

The downtime number that matters is your own

If a vendor justifies a price with an industry-wide cost-of-downtime figure, ask what size of company it describes. Those surveys are about enterprises, and a number about companies with a thousand employees says nothing about a six-person agency.

The arithmetic that works for a small business is your own. Take your revenue per working hour, multiply by the hours the thing was broken, add the hours someone spent re-entering data by hand. For most small operations that lands in the hundreds, occasionally in the thousands - and it's the only figure that should decide whether a fixed-price fix is worth it today or next week.

The one place that arithmetic understates the damage is anything that touches leads. An enquiry answered within the hour is a conversation; the same enquiry answered the next day is usually a competitor's customer. A form that silently stopped delivering isn't losing you leads next quarter. It's losing them today.

How do you tell a real scope from a guess?

By what you're asked before a number appears. Anyone can produce a price; only someone who intends to be bound by it needs to know what they're walking into.

n8n's own bug report template requires ten fields before a maintainer will look at a problem: version, the steps to reproduce it, what you expected, what happened instead, the environment. A shop that quotes your outage without asking the same kind of question is pricing a guess, and a guess is what gets revised once the work starts.

The second signal is where the boundary is drawn. Look at how the fixed-price fixes on the n8n jobs board define themselves: one covers "one existing broken path, covering up to three relevant nodes"; another is bounded to "up to 3 connected workflows or 40 steps, on 7 days of run history". The unit is the system, not the clock. Hours are the seller's problem; nodes, steps and days of history are things you can check.

The third is what happens when the job turns out to be bigger. A change policy worth the name says two things in writing: what counts as a change - scope, assumptions, access, data quality, a third-party dependency moving - and what follows when one of them does: a variation agreed with you, an updated timeline and fee, or a pause until you decide. Written triggers and a named response are the version you can check. The alternative is silence, followed by an invoice.

Where our line is

We sell the middle band the market keeps skipping, at a fixed price agreed before anything starts, with an acceptance checklist written down first. The size is confirmed by free scoping, and if scoping shows the job is smaller, the smaller price applies.

It isn't always worth paying. A single Zap writing to the record that triggers it, a filter you can add yourself, an integration that needs reconnecting: those are an evening with the vendor's documentation, and the vendor's docs are good. Pay someone when the failure won't reproduce, when it has been writing bad data for weeks, when nobody left notes, or when the hours you'd spend are worth more than the fix. You can size the job yourself before talking to anyone, and what a scoped repair covers end to end is written out separately. The same failure in Zapier and in Make prices differently, because the meters differ.

Sources

Prices and counts below were checked on 15 August 2026. Marketplace listings, partner directories and vendor plans change without notice, and several vary by country.

• Upwork job search - live fixed-price job distribution, counted 15 August 2026
• Zapier Solution Partner Directory - published partner minimums and discovery-call policy
• Make Partner Directory - tiers published, prices not
• How features use credits, Make - a trigger costs a credit even when it returns nothing
• Replay Zap runs, Zapier - replays re-bill successful steps
• Understand executions, n8n - what does not count against quota
• View and manage your Zap history, Zapier - the run-history ceiling
• n8n bug report template - the fields a maintainer needs before looking at a problem
• Agency audit fees, emergency-repair tiers and the marketplace review quoted above were read on their publishers' own pages on 15 August 2026 and are given without names