Agency margin check: cost per client site before and after AI
Short answer: Agency margin per client site is what is left of the price you charge after you pay every cost of building and running that site: people's time, freelancers, software and fixes. Gross margin means the price minus those direct delivery costs, shown as a percent of the price. To check what AI changes, list all direct costs for one recent site, rebuild the same list for the AI way including review, rework and tool fees, and compare. AI helps margin only if total cost drops after those hidden costs, or if you can serve more clients with the same team.
You want to know if AI is really making your client websites more profitable, not just faster.
I'm Ranjit, and I run Blur Studio: 15+ years building software, 120 apps taken live, and clients who have gone on to raise $10M+. I build AI systems and processes that help businesses do a lot more with a lot less time. I also work differently from most of what else is out there, more on that further down this page. Say hello.
The fastest way to fool yourself is to count only the hours AI saved. Margin is about total cost against price, and AI adds new costs: review time, rework, tool fees and training. This page gives you a one-page sheet you can copy, a worked example with made-up numbers, the hidden costs most agencies miss, and rules for deciding what to do with the result. Use your own numbers, because mine here are invented to show the method.
What "margin" means here
There are three common measures, and agencies mix them up.
- Gross margin: price minus direct delivery cost, divided by price. Direct cost means what it took to deliver this site: staff time, freelancers, tools used for it.
- Operating margin: what is left after overhead too (rent, sales, management, admin).
- Net margin: what is left after everything, including tax and interest.
This page works at gross margin per site, because that is the level where AI changes things. For the whole-agency view, see which clients and services actually make us money.
For a sense of scale, Productive, a software company that makes agency project tools, reports in its BenchPress 2025 survey that average gross profit margin for UK agencies with revenue of GBP 1 million or more fell to 39 percent in 2024, which it calls the lowest on record, and it recommends at least 50 percent. It says the survey draws on thousands of agency owners but gives no exact sample size (Productive BenchPress 2025, published 30 April 2025). AgencyAnalytics, a reporting software company, surveyed 494 agencies from February to April 2026, and 68 percent said their operating margin was above 20 percent. That sample is people who answered a vendor survey, so it likely skews to healthier shops (AgencyAnalytics 2026 report). The two figures measure different things in different samples, so do not average them. Use them only as a rough frame, and compare yourself to your own history first.
The simple sheet
Copy this into a spreadsheet. Fill one column for a recent site built the old way and one for a site built the AI way (or for your estimate of it).
| Line | What to include |
|---|---|
| A. Price charged | Fixed fee for the build, excluding tax |
| B. Staff hours x loaded hourly cost | Strategy, design, build, QA, project management, client calls. Loaded cost means pay plus payroll tax, benefits and a share of equipment |
| C. Freelancer cost | Invoices for this site |
| D. Review hours x cost | Time to check AI output and approve it, listed separately so you can see it |
| E. Rework hours x cost | Fixes after review, after the client, and after launch (the first 30 days) |
| F. Tool fees for this site | Share of software subscriptions, AI credits, stock images, fonts, plugins, hosting you pay for the client |
| G. Revision rounds beyond scope | Hours given away because the contract allowed it |
| H. Total direct cost | B + C + D + E + F + G |
| I. Gross profit | A minus H |
| J. Gross margin | I divided by A, as a percent |
| K. Profit per staff hour | I divided by total staff hours (B + D + E + G) |
Line K matters as much as J. If AI cuts hours, your margin percent may rise a little but your profit per hour can rise a lot, which is what lets one team handle more clients.
Worked example (all numbers made up)
An agency charges a fixed $6,000 for a 6-page site. Staff loaded cost is $45 an hour. Here is the same site two ways.
| Line | Old way | AI way (assumed) |
|---|---|---|
| A. Price | $6,000 | $6,000 |
| B. Staff build hours | 50 h = $2,250 | 22 h = $990 |
| C. Freelancers | $900 | $300 |
| D. Review | 5 h = $225 | 12 h = $540 |
| E. Rework | 4 h = $180 | 6 h = $270 |
| F. Tool fees | $80 | $150 |
| G. Extra revisions | 3 h = $135 | 3 h = $135 |
| H. Total direct cost | $3,770 | $2,385 |
| I. Gross profit | $2,230 | $3,615 |
| J. Gross margin | 37 percent | 60 percent |
| K. Profit per staff hour | $2,230 / 62 h = about $36 | $3,615 / 43 h = about $84 |
In this made-up case, margin rises because hours fall far more than review and tool costs rise. Now stress test it. If review takes 20 hours instead of 12 and rework takes 12 instead of 6, total direct cost climbs by about $630 and gross margin falls to roughly 50 percent. If you also pass the saving on to the client with a lower price, the gain shrinks again. At a $5,000 price with the same costs, gross profit is $2,615 and margin is about 52 percent. Run those scenarios in your own sheet before you set prices.
Hidden costs most agencies miss
Review time
Google's guidance on AI-generated content says to manually fact-check and review it before publishing, including titles, meta descriptions, structured data and image alt text (Google Search Central). That work is real labour. Log it on its own line so it does not hide inside "build".
Fixes after launch
Wrong phone number, a form that sends to nobody, a page with a stray claim. Track post-launch fixes for 30 days per site. If AI sites need more fixes, your savings are partly borrowed from the next month.
Tool fees
- Software subscriptions: a site platform, an AI assistant seat, an image tool, a reporting tool. Divide each by the number of sites it serves.
- Usage credits: many AI tools sell points that run out and do not roll over. A heavy month can cost more than a light one, so use an average of three months, not your best month.
- Per-client hosting or plan fees: if you pay them and bill the client, include both sides. If you absorb them, they are direct cost.
Learning and setup
The first three AI sites cost more because you are building standards, fact sheets and checklists. Count that as a one-time investment, spread across the next 10 sites, not zero.
Scope creep and speed expectations
When clients learn you work faster, they ask for more. Without clear change orders, the hours you saved go back to the client. See agency scope creep and the change order template.
Switching and exit costs
If a client site is locked inside one platform, moving it later costs you or the client. Check export and ownership terms before you build. See who owns your website and content.
Idle capacity
Saved hours only become profit if you sell them or cut cost. If staff finish a site in two days and sit idle on day three, nothing improved. Pair this sheet with agency capacity planning.
What to do with the numbers on price
Three options:
- Keep the price, keep the gain. Works if you sell on outcomes and quality, and clients compare you on results, not hours.
- Lower the price to win more work. Works if you have a pipeline to fill. You need volume to make up the margin.
- Keep the price, add scope. Include something clients value, such as a monthly check, a faster launch or a care plan.
Price context matters. AgencyAnalytics published example ranges in October 2024: a responsive redesign from about $3,000 for 5 pages to about $10,000, and content retainers of $1,500 to $2,500 a month. Those are one vendor's examples, not a market average, so use them only to see if your price is in a sensible band. More on models in how to price website builds and retainers when AI cuts the hours.
Decision guide
| If your sheet shows | Then |
|---|---|
| Total direct cost down 15 percent or more and no rise in post-launch fixes | Roll the AI way out to similar sites, one client type at a time |
| Hours down but review and rework erase the gain | Fix standards (fact sheet, checklist) and re-test on 3 sites before you decide |
| Margin up but profit per hour flat | You may be paying for tools that replace little. Cut tools or move more tasks |
| Margin up, but you have no new clients to use the freed time | Fill the pipeline or add a paid service before cutting price |
| Margin down | Stop and find the cost line that grew. Do not scale a loss |
The 15 percent line is my planning rule, not a standard. Choose a line that fits your risk, and write it down before you look at the result, so you do not move the goalposts.
When to run the check
Run it at three points. First, before you adopt any AI tool, on your last three finished sites, so you know your true starting margin. Second, after your first AI-assisted site, with review and rework timed on their own lines. Third, every quarter after that, using the average of that quarter's sites, because tool fees, credit packs and staff mix change. Keep one shared sheet, not one per person, so the same definitions apply to every site. Write the definition of "loaded hourly cost" at the top, and do not change it mid-year, or your before and after columns will not be comparable.
Also split sites by type. A five-page brochure site and a 40-page site with a shop have very different cost shapes, and AI helps them differently. Mixing them in one average hides where the gain really is. If you only have a few sites of each type, say so in the sheet and treat the result as a rough guide.
Common mistakes
- Using price as the base for time saved, not cost.
- Counting only the best site, not the average of several.
- Leaving out your own time as owner.
- Ignoring tool fees because they are "small" at one site but real across thirty.
- Treating a vendor's survey margin as your target without checking its sample.
How I'd approach this
I would run the sheet on your last three finished sites first. That gives you your true starting margin, which is often lower than people think. Then build the next site the AI way with timed review and rework lines, and compare. Do this for three sites before you change prices or staffing, because one site proves nothing.
If your work needs something specific, such as AI agents that use your business knowledge, CRM and calendar connections, or finishing a half-built flow, Blur Studio builds custom versions with ongoing care. You can see what that covers here. The sheet above stands on its own without it.
First-week plan:
- Day 1: copy the sheet and pull time logs and invoices for your last three sites.
- Day 2: fill the old-way column and compute J and K for each.
- Day 3: choose your next site for the AI way and set up timers for review and rework.
- Day 4: write your pass line (for example, 15 percent lower direct cost, no extra fixes).
- Day 5: tell the team what you are measuring and why, so nobody feels watched.
Decision rule: keep a change only if total direct cost per site and profit per staff hour both improve across at least three sites, with no rise in post-launch fixes.
Related guide: replacing a freelancer bench with AI agents.
Questions people ask
What is a good gross margin for a web agency?
There is no single standard. Productive's BenchPress 2025 report put average gross margin for UK agencies with GBP 1 million or more in revenue at 39 percent for 2024 and recommends at least 50 percent. TMetric, a time-tracking vendor, publishes other ranges. Samples and regions differ, so compare yourself to your own history first.
How do I work out the cost of building a website for a client?
Add staff hours at loaded cost, freelancer invoices, review and rework hours, tool fees for that site, and extra revision rounds. Subtract the total from your price. The sheet on this page does it line by line.
Does AI always improve agency margin?
No. It cuts build hours but adds review, rework and tool fees. In the made-up example here margin rose, but a stress test with heavier review erased nearly half the gain. Measure on at least three sites.
What is loaded hourly cost?
The true cost of an hour of staff time: pay, payroll tax, benefits and a share of equipment and software, divided by billable hours. It is higher than the wage rate.
Should I lower my price if AI makes me faster?
Not automatically. Options are to keep the price, lower it to win volume, or add scope. See pricing website builds when AI cuts the hours.
How do I count AI tool fees per client?
Divide each subscription by the number of sites it serves, and use a three-month average for credit-based tools because unused credits usually do not roll over. Count hosting or plan fees you absorb as direct cost.
How long should I track post-launch fixes?
Thirty days is a practical window. Count every hour spent on fixes in that time as rework on the site's sheet.
Where does freelancer work fit in this?
List freelancer invoices on their own line. If you plan to replace freelancers with AI tools, read replacing a freelancer bench with AI agents and run a pilot first.
More in this topic: Websites: agency delivery, quality control and margins (10 guides) →