Website performance optimization that finds where the time goes
Send us a URL and tell us where it hurts, or that you cannot tell. We measure the real load path, and everything you share stays private.
Tell us where it is slow
One URL, one line about what feels wrong. What comes back is a measurement scope, not a price.
What this usually looks like from the inside
The last thing we bought moved a score and nothing else moved
A green number arrived, the complaints did not stop, and the money is already spent. Passing is common: the Chrome UX Report puts 53.2% of origins passing all three Core Web Vitals on mobile and 60.0% on desktop in July 2026. A passing grade still does not tell you what your own customers waited for.
Nobody can point at the slow part
Something inside the product is wrong and nobody can say what. Your team can name five suspects and not one of them has been measured this quarter.
The people who built it could not make it fast
That is an uncomfortable conversation to start, and it is why most owners end up shopping outside the relationship they already have.
An outsider near production is a risk somebody carries personally
Whoever holds the pager will be on the call if a change lands badly during a peak hour. No contract moves that call to anyone else.
You already paid for this product once
If this spend does not move the number, you are the person who paid twice with nothing new to show for either payment. That part rarely gets said out loud.
None of that is a reason to rebuild yet. It is a reason to measure first.
Show data table
| Segment | Value (%) | Share | Source |
|---|---|---|---|
| Passed every Core Web Vital in the field | 57 | 57% | HTTP Archive, September 2021 |
| Failed at least one Core Web Vital in the field | 43 | 43% | HTTP Archive, September 2021 |
A score in the nineties is a lab result, taken on a fixed device on a fixed network. More than two in five of those pages still failed their own users. Ask whoever is quoting you a score what their field numbers did.
Show data table
| Dimension | Good origins | Needs improvement | Poor origins |
|---|---|---|---|
| Server response | 600 ms | 1,360 ms | 2,270 ms |
| Delay before image fetch | 350 ms | 720 ms | 1,290 ms |
| Image download | 160 ms | 270 ms | 350 ms |
| Browser render | 230 ms | 310 ms | 360 ms |
Server response is the largest part at every tier and it grows fastest as sites get slower. On a poor origin the browser has waited 2.27 seconds before it has anything at all to draw. The rule that 87% of end-user response time sits in the front end comes from Steve Souders, The Performance Golden Rule, measured in 2012 across the top 50,000 sites, at 76% for the top 10 and 92% around rank 10,000. Front-end work still matters, and it cannot reach the tallest bar here.
How an engagement runs
01
Scoping
We agree what is being measured, what it costs and when it runs, in writing, before any of the work starts.
You get
A written scope for the measurement, with the schedule agreed alongside it before work starts.
02
Measurement
We take field data from your own origin, server traces, query plans and the request waterfall, on the paths that matter to the business.
You get
A load path with the time attributed to server, network, assets and render, at your customers' own percentiles.
03
Findings
We tell you where the time goes and what we would change, ordered by what returns most for least risk. That includes what to leave alone.
You get
A ranked list with an expected effect and an expected effort against each item on it.
04
The fix
We quote the tuning work against the paths the measurement found, then do it on your accounts alongside your own software engineers.
You get
Faster load paths live in production, and the instrumentation left running in your own account.
05
Handover
Your team keeps the measurement setup and the thresholds, and the diagnosis is written where your software engineers can re-derive it.
You get
The instrumentation, the alert thresholds and the written diagnosis, all in your own accounts.
| What is on offer | Published price | Who published it, and what it covers |
|---|---|---|
| A marketplace gig | Published price20 to 450 USD | Who published it, and what it coversFiverr live seller listings, retrieved 2026-08-27. Marketplace seller listings, not an agency scope. |
| A document, with implementation stated as out of scope | Published price300 to 1,000 USD | Who published it, and what it coversShopExperts, 2026 pricing guide, 2026-05-24. Shopify stores. |
| Shipped changes | Published price500 to 8,000 USD | Who published it, and what it coversShopExperts, 2026 pricing guide, 2026-05-24. Shopify stores. |
| Architectural rework | Published price5,000 to 100,000 USD | Who published it, and what it coversShopExperts, 2026 pricing guide, 2026-05-24. Shopify stores. |
| An ongoing retainer | Published price2,000 to 10,000 USD a month | Who published it, and what it coversShopExperts, 2026 pricing guide, 2026-05-24. Shopify stores. |
| An open rate card | Published price1,800 to 4,800 USD a month, plus 3,125 to 6,250 USD onboarding, inside a 50 to 12,000 USD span | Who published it, and what it coversWebFX, Site Speed Optimization Pricing. |
| A web development agency, measured across its clients | Published price25 to 49 USD an hour, a 66,499 USD average project, a 7,139 USD average month, nine months typical | Who published it, and what it coversClutch, Web Development Company Pricing Guide. Web development generally, not speed work alone. |
| Hiring it in | Published price134,129 USD average base, range 78,735 to 228,499 | Who published it, and what it coversIndeed, updated 2026-08-16, for a performance engineer in the United States, corroborated at 140,763 USD average by Glassdoor, 2025. A US in-house salary, not a project price. |
What our engagement includes
The measurement
Agreed at scoping, alongside the scope.
What it includes
- Field data from your own origin, at the percentiles your customers sit in
- Server traces, query plans and the request waterfall
- A load path with the time attributed to server, network, assets and render
- A ranked list of what to change
- A written answer if the honest finding is that nothing here is worth paying to fix
Widens when More load paths matter to the business, production access needs a security review, or the slow part sits in a platform we cannot change.
The tuning work
Agreed at scoping, alongside the scope.
What it includes
- The ranked changes, taken in the order you agree
- Work on your own accounts, alongside your own software engineers
- The instrumentation and the alert thresholds left running in your account
- Modernization without forcing rewrites where audits show incremental hardening will do
Widens when The slow part is how the page is assembled rather than a query and an index, or it sits in third-party script you bought.
Ready to find out where your time goes?
Send the URL and where it hurts. Two good outcomes: a scoped measurement, or a straight answer that this is not worth buying.
Five published experiments, including one where speed was worth nothing.
Kohavi, Deng, Longbotham and Xu (Microsoft), Seven Rules of Thumb for Web Site Experimenters, KDD 2014
A randomized experiment reported by Kohavi, Deng, Longbotham and Xu (Microsoft), KDD 2014
Jake Brutlag, Google Research blog, Speed Matters, 2009, cross-checked as reference 12 in the KDD 2014 paper
Google web.dev with Vodafone Italy, 2021-03-17: a 50/50 A/B test on a paid-media landing page, LCP 31% better
Measured across 37 brands, observational rather than experimental. Deloitte Digital with Google, Milliseconds Make Millions, 2020
*Disclaimer: Published experiments run by other companies, not our results. The Bing null held across almost 20 million users.
If it goes wrong
An outsider is in production and something leaks.
The work happens inside accounts you own, under access controls you set, on an audit trail you keep.
What we commit to
We work on your own accounts, under an NDA from day one, signing in through your own identity provider (SSO), so removing someone removes them everywhere, and with secure repositories. You fully own all of the code that we write.
Your compliance team asks for certifications we do not hold.
What we commit to
We do not hold SOC 2, GDPR, or HIPAA certifications ourselves. We engineer systems that pass your audits. We document architectural decisions in the TRS so your compliance team can verify the design intent.
You buy a senior conversation and get a junior team.
Our co-founders Tirth Bodawala and Ajay Patel still take production calls themselves, years into an engagement.
What we commit to
A core team assigned to your project, a lead software engineer you can talk to directly, no rotation and no handoffs you did not agree to. The same software engineers stay across years.
Something needs a decision and nobody on our side is awake.
What we commit to
We work IST 10am to 7pm, with all software engineers currently based in India, and we agree on overlap commitments at engagement scoping so expectations are clear from Day 1.
The scope moves quietly and the invoice moves with it.
What we commit to
Nothing in the build moves on a re-scope you have not explicitly accepted, and that acceptance is confirmed both in a meeting and in writing before it starts.
The delivery pipeline we build becomes the new fragile thing.
This is the delivery pipeline we build, not an availability commitment for your system.
What we commit to
The pipelines we build target 98-99% uptime under normal operating conditions, and we set the alert thresholds that prove it. Cloud-provider outages sit outside our scope.
One engagement, and what tends to happen next
Anonymised, as every case on this site is.
The situation
A hyperlocal SaaS marketplace operating across Europe, with a lagging frontend on a product people used every day.
What changed
Server-side rendering, state management and memory-optimized rendering, so the heavy work moved off the device and the page stopped fighting the browser for memory.
Where it landed
A high-performance shopping experience on the same product, without a rebuild.
The pattern across engagements
Several of our longest-running engagements began the same way. A client came to us for a single Scale & Optimize project, verified the work, and then asked us to take on more. That is a pattern across different clients rather than a footnote to the one above.
Faster load time on low-memory mobile devices
Read the full case studyWhen we are the right call
Where we fit
You have a live product or store and a number that is slipping.
You want to know where the time goes before you commit to paying for a fix.
The slow part might be the database, the server or a third-party script, and you want someone who will say that out loud.
Your own software engineers stay on it afterwards, and you want them holding the instrumentation.
A security review has to happen before anyone touches production, and you would rather that be planned than negotiated.
Where we are the wrong call
Your budget for this sits at the 20 to 450 USD marketplace level, where the work on offer is a plugin install.
The page or the product does not exist yet, so there is nothing to measure until it is built and has real users.
What you actually want is a certificate or a score rather than a faster product, which a monitoring subscription will give you for far less.
The product is being rebuilt next quarter anyway, so fix it inside the rebuild rather than tuning it twice.
Nobody can give access to the code, the infrastructure or the field data, so nobody can measure it, including us.
The questions people bring to the first call
Are you going to tell me to rebuild it?
What happens when it gets slow again in six months, and am I signing up for a permanent bill?
Who is actually doing the work?
What if the measurement says the problem is a third-party script we do not control?
What happens after you leave?
Can you work where we cannot change the platform code?
For the person who has to approve this
Send this to whoever signs it off. Every line stands on its own.
Forward this part
Performance measurement engagement, Atyantik Technologies
- What gets measured Field data from the live origin, server traces, query plans and the request waterfall, producing a load path with the time attributed to server, network, assets and render.
- What comes back first A ranked list of changes with an expected effect and an expected effort against each one, including the items Atyantik recommends not doing.
- Where the work happens On the client accounts, under an NDA from day one, signing in through their own identity provider, which means they remove us in one place (SSO), and with secure repositories. The client fully owns the code.
- Certifications Atyantik does not hold SOC 2, GDPR or HIPAA certifications. Atyantik engineers systems that pass client audits and documents architectural decisions in the TRS so a compliance team can verify design intent.
- Scope control, and the stopping condition Nothing in the build moves on a re-scope the client has not explicitly accepted, confirmed in a meeting and in writing. If the measurement finds nothing worth paying to fix, Atyantik says so and there is no second phase to buy.
- First contact Send a URL and one line about what is slow. Atyantik will reply within one business day with a technical person on the call.
Tell us where it is slow
One URL, one line about what feels wrong, and anything you already tried.
Send the URL and where it hurts, or where you think it hurts. What comes back is what we would measure, and what that measurement would take.
Two good outcomes. Either we scope a measurement and you know exactly what you are buying before you pay for anything, or the honest answer is that nothing here is worth paying us to fix, and that is the answer you get. Tell us what you need. We will reply within one business day with a technical person on the call.
- Everything you share stays private.
- An NDA from day one.
- A technical person on the first call.
- No price comes back from this form. A measurement scope does.