Most agencies do one half. A development studio hands you a site and walks away from whether anyone converts on it. A marketing agency buys traffic and sends it into a page they are not allowed to change. We do both, which means when the numbers move we know exactly what moved them, and when they do not, we can go and fix the actual cause. Below is the work, with every number stating how it was measured.
Every business that has hired both separately knows this feeling. Something is not converting, and the two suppliers each explain why it is the other one's problem.
The build is judged on whether it looks right and loads fast. Nobody on that team ever opens the ad account, so nobody notices that the highest-intent traffic is landing on a page that answers a different question. The site is finished. It is just not earning.
They can see the drop-off and they can write the recommendation. They cannot deploy it. So the fix waits in a ticket queue while the budget keeps running, and the monthly report explains the same leak it explained last month.
We read the funnel, we change the page, and we watch what the change did. There is no handover, so there is nobody to blame, which is the point. Every case below has a section naming the exact moment the two halves met, because that moment is the whole argument.
This is the differentiator, laid out rather than claimed. 7 of them needed three or more. Read down a column to see depth in one discipline. Read across a row to see why a project needed one team instead of three suppliers who each own a piece of the answer.
| Project | Web build | Marketing | SEO | CRO | CRM | Dashboard | Total |
|---|---|---|---|---|---|---|---|
| Perth Pre-Purchase Inspection | ● | · | ● | ● | · | ● | 4 |
| US translation-device brand | · | ● | ● | ● | · | · | 3 |
| Suwasetha | ● | · | ● | · | · | · | 2 |
| Lotus Edge | ● | · | ● | · | · | · | 2 |
| AgrilHoTech | ● | ● | ● | ● | · | · | 4 |
| Sri Lankan motorbike parts store | · | ● | ● | ● | · | · | 3 |
| Sri Lankan sportswear brand | · | ● | · | ● | · | · | 2 |
| Sri Lankan fashion label | ● | ● | · | ● | · | · | 3 |
| Resort group, Maldives | · | ● | · | ● | · | · | 2 |
| Spear Systems Modular Solutions | ● | · | ● | · | ● | · | 3 |
| Kenny Dissanayake | ● | · | ● | · | · | · | 2 |
| Vigor Security | ● | · | ● | · | · | · | 2 |
| Ransewana Constructions | ● | ● | ● | · | ● | · | 4 |
| Event ticketing campaign | · | ● | · | · | · | · | 1 |
| Footwear brand, New Zealand | · | ● | · | ● | · | · | 2 |
| International e-commerce brand | · | ● | · | · | · | · | 1 |
| Education brand | · | ● | · | · | · | · | 1 |
| HNC Associates | · | ● | · | · | · | · | 1 |
Every figure here says how it was measured. Where a project has no number yet, it says that too, and shows the build instead. Filter by the discipline you care about.
Showing all 18 projects.
How it was measured. The site's own funnel report, reading the live booking database. Every step is a recorded event, not a modelled estimate. In the first three days after launch, that was 28 confirmed bookings from 162 page opens. Volume has grown since, which is why the rate above is the number we track rather than the count behind it.
The funnel report was naming the wrong drop-off. It still counted postcode entry as a step from an earlier version of the form, so it pointed at a problem that no longer existed. Fixing the measurement changed the decision: at three inspection slots a day, the fall-off between choosing a tier and picking a time is mostly a full calendar rather than a broken one. The advice was to check the bookings list before buying any traffic. An ads agency reading the same chart would have sold more clicks into a capacity ceiling.
The last step is the current leak. Half of everyone who starts entering their details does not finish, which is where the next work goes.
How it was measured. The same five-part marketing audit, covering content, conversion, SEO, competitors and brand, run against the live site in July and again on 10 September 2026. The client's developers shipped the fixes. We wrote the list and re-checked each item on the live site rather than taking it on their word.
Before any budget went into Google's Performance Max, the tracking check found that the order confirmation page believed whatever was typed into its address bar. A made-up order for $14,999.50 showed the "your payment went through" screen and fired a real purchase event. Performance Max learns from purchase events, so every fake or test visit would have taught it to find more people who never buy. The fix went in on the server, and seven fake orders were rejected on the retest before the campaign plan was written.
What we built. A nine page site, relaunched in September 2026 with elder care as the lead service alongside wound care, home visits and care at the centre. Every price, qualification and clinical statement was checked against a source before launch, and the release gate ran an automated accessibility scan and a timed page load. Enquiries open a prefilled WhatsApp message instead of an email form.
The person searching is usually an adult child, often working abroad, arranging care for a parent who is not the one doing the searching, so the site answers a stranger's questions about someone else's condition. When elder care became the lead service, the site's single set of opening hours contradicted every elder-care package on the same domain, because the evening hours belonged to the nurse's wound-care clinic. It now publishes two labelled sets of hours, in the copy and in the data Google reads. One combined span would have been false, since nobody covers the gap between them.
How it was measured. Load time is the largest-paint figure from five Lighthouse mobile runs against the live menu page, all inside a 2.5 second budget. The 85 dishes in 14 sections are read from the restaurant's own Google Sheet, refreshed every 60 seconds, with a saved copy served if the sheet is ever unreachable.
The printed leaflets already pointed at lotusedge.lk, and the page they reached still said "Opening" with a countdown, sample prices and a demo phone number, so fixing it came before any promotion. Nobody on staff is there to update a website, so the menu reads the sheet the owner already edits: untick a dish and it shows as sold out on every table. Table ordering was held back on purpose until the restaurant picks a till that can accept orders from the web, because two systems holding the same order is how staff end up abandoning one of them.
How it was measured. A live audit of the store run on 12 July 2026 against a fixed three-part rubric, with the evidence file saved. This is a starting score on our own property, published because a baseline you can check beats a result you cannot.
robots.txt explicitly allows GPTBot, PerplexityBot and ClaudeBot while blocking the scrapers that give nothing back. That is a server-level edit made for a marketing reason: being answerable when a buyer asks an AI assistant where to buy a plant in Sri Lanka. A build team has no reason to touch robots.txt for answer engines, and a marketing team is not allowed to. The same audit found the homepage title reading "Home - Agril" and the shop description still carrying an unedited plugin placeholder, one file away from an llms.txt that was thorough.
How it was measured. A full UI, CRO, SEO and technical audit run on 4 June 2026 against a scored rubric, with the report and eight screenshots saved. Same rubric on re-audit, so the next score is comparable rather than a fresh opinion.
The store had no main navigation menu, no meta descriptions anywhere, a pricing bug, a broken wishlist and thirty-one products with no descriptions. It also had no phone number, which in this market is the largest single trust barrier before anyone will order a part they cannot inspect. Three fixes, navigation, phone number and meta descriptions, were modelled to move the score past 55. Selling traffic into that store first would have spent the budget on a shop nobody could navigate.
How it was measured. The 54/100 starting score comes from a scored audit run before any spend. ROAS and AOV are drawn from the client's own Meta Ads Manager and store reporting over the 2.5-month engagement.
The account was rebuilt into six campaigns, but the first job was verifying that the Pixel and the Conversions API were actually reporting before a rupee of the new structure went live, since optimising against broken data is how budgets disappear quietly. Average order value moving from LKR 2,472 to 6,841, up 177 percent, is not an ads result. It is a merchandising and on-site result that the ads then get to spend against, and it is the reason the return on spend had room to move at all.
What we built. A custom HTML5 Canvas tool that renders an anatomical body silhouette from the shopper's own measurements, so they can see how a piece will sit before they order it. The label's Meta sales campaign is running now, and its results are not in yet, so none are shown here.
Online fashion here loses orders at the size question, and no amount of ad spend answers it, so the fit tool exists because the same team was watching where the drop-off happened. The ads audit found the same kind of gap from the other side: in the 31 days before the audit, the store's Pixel recorded about 100,000 page views, one add-to-cart and no purchases. A campaign told to find buyers would have had nothing to learn from, so the site's tracking went first on the launch plan, ahead of any creative.
How it was measured. Read from the account's own exports for 6 July to 28 September 2026, across three campaigns and 21,915 search terms, with every figure marked as either observed in the data or calculated from it. Underneath the ads sits the stack built earlier in the engagement: Meta Pixel, Google Ads and GA4, Hotjar session recordings and Klaviyo email automation.
Bookings finish on a third-party booking engine, so the campaigns launched optimising for the last step the site could see: a click on Book Now. Twelve weeks of data showed that stand-in was measuring browsing, not buying. It counted every click, and brand and generic searches converted at the same 30 percent, which real booking data never does. A real booking event from the booking engine is the fix now being set up. The same audit found one resort's only ad disapproved, which had switched its whole campaign off, and the blocked-search waste above going to competitor resorts, job hunters and a Careers link Google had added on its own. Earlier, the same weekly watch caught copycat domains impersonating the brand.
A static lead-generation site built for speed and for one job: getting a qualified enquiry into the owner's hands.
Built lead-generation first rather than brochure first. In this market the enquiry happens on WhatsApp, so the site hands people to WhatsApp with context already attached instead of burying a contact form on page seven and hoping.
A dark editorial single-page site with scroll-driven motion, built to hold attention on a phone.
A portfolio site is usually treated as decoration. This one had to satisfy two very different readers at once: a casting director scrolling on a phone, and a search engine deciding whether this name is a real, indexable entity worth ranking for.
What we built. A six page site covering six service lines, careers and contact, on a brand system sampled from the logo file rather than eyeballed. A holding page is live at vigorsecurity.lk while the full site is completed.
A new security company has no clients to name, no testimonials and no photographs of its guards, and this industry's templates fill that gap with invented numbers. The first detail we asked the client for was their Ministry of Defence registration, because it is the one credential a buyer can check. The same rule runs through the design: the brand gold measures 1.76 to 1 against white, unreadable as text, so it only appears on navy, where it clears 8 to 1.
Five pages plus service-detail pages, real photography, and a contact path that falls back to the visitor's own email client if the server ever fails, so an enquiry is never silently lost.
The client is newly established, so the site carries its founding year, an in-house team, a 24-hour reply commitment and a two-year workmanship warranty, instead of the fabricated project counts and certification badges that this industry's templates ship with by default. The homepage runs an honest FAQ where testimonials would normally sit. Trust signals that turn out to be false cost far more than having fewer of them.
How it was measured. Reported spend and bookings across Meta and Google for the campaign, pulled from the platforms' own ad accounts.
How it was measured. Site-wide conversion rate and blended ROAS, read from GA4 and Meta Ads Manager over a one-month engagement, alongside a 28 percent lift in average order value.
How it was measured. Cost per purchase reported in the ad account, tracked over a two-month engagement.
How it was measured. Course registrations and cost per lead, read from the platform's own lead-gen reporting over a one-month window.
How it was measured. Lead-generation campaign spend and attributed revenue for FY 2024-25, reported by the client against closed engagements.
Accounting firms sell on trust and referral more than on a landing page, so the ads are doing the work a website usually would: explaining the service and getting a name into a conversation. That funnel is currently running with no site behind it, which is the honest state of this engagement right now. The website is the next lever, once the same numbers are there to justify what to build.
Marketing results are shown by industry rather than by client name, because those numbers came out of accounts we do not own. Websites are named, because a live site you can open and inspect is a stronger proof than anything we could write about it.
The Growth Audit takes about three minutes and tells you whether your problem is the traffic, the site, or the gap between them. No call required to get the answer. If you already know what you need, every engagement above is scoped and priced on the services page.
