Eight years operating inside seven industries, among them hospitality, real estate, technology, marketing, fitness and construction. That is why the systems I build understand how the business actually works, instead of the other way around.
Each case opens its own page, with the role I held, what was solved and how long it took.
The operation ran on HubSpot, where everything hangs off the contact. But one trip holds several bookings, and whoever coordinates it needs to see them together. We built the CRM that puts them together: 792 bookings and 2,513 tasks migrated without loss, and 39 screens verified against the design down to zero differences.
The domain was already registered, but nothing was built on it: typing it in landed you on the registrar's parking page. It is now 13 pages, with ten articles written for two different audiences, and a homepage that weighs 23 KB compressed. Not one stock photo.
Site content could not be edited without going into the hosting. We rebuilt it with its own admin panel: 17 filters, 180 options and 2,487 listing values migrated from the previous site.
A creative agency's site is its own portfolio: if it does not look good, it has already answered the question. Eight services, nine projects and two languages where the contact button switches channel by country.
Twenty years of real work, split across two Google listings that never added up. Project in progress: half is already live, the rest is under way.
A training app built around a real coach, sold by a specific person who was nowhere on the site. Every AI-generated asset failed: it had to be rebuilt from real app screens and one real photo.
Twenty-nine years of reputation that lived only in past clients' memories. Selling a category nobody knows exists, and building authority without being allowed to promise a single outcome.
Three generations of contractors without a single page of their own. The brand's first real online presence, built on a symbol the family already used without knowing it was their identity.
You look for a plumber with water running across the floor. A site where the decision takes seconds, built on the system the plumber already used on the phone without knowing it was design.
A company billing in dollars, spending in pesos and operating in Mexico, run on a spreadsheet rebuilt every month. The problem was not the work it took: it was that the past would not hold still.
I do not sell hours. I sell things that already work elsewhere and get adapted.
The same method across all seven industries. What changes is the business, not the way in.
I operate first and advise second. I learn how the business actually works before proposing a single change.
I define the real process: who does what, in what order, by what criteria, and what happens when something breaks.
I set up the tools that hold that process. CRM, automations, dashboards. What is needed and nothing more.
I document it and get out of the way. The team has to keep it running without me, and that is the part almost nobody delivers.
If the system needs me to keep running, I am not finished yet.
Argentina, Mexico and the United States, where I live today. Twelve organizations, seven industries, and almost always the same brief: come in, fix something that was already failing, and leave it running.
None of this I learned in a classroom. I learned it at the front desk, in sales and in admin, before I built a single system. That is why I know where things break: I was on the other side when they did.
I reply the same day. If it is not for me, I will say so.
A luxury travel agency was coordinating entire trips on a CRM built to sell, not to coordinate. And a trip is not a sale: it is a dozen bookings that all have to go right, in order — the airport transfer, the villa, the second-day chef, the yacht, Saturday's table — each with its own vendor, time and commission.
Nobody could see a whole trip at once. To know how a family's weekend was shaping up you had to open their contact, go into each booking and read the body of every message. We built the system where the trip is the center and everything else hangs off it.
Inverting the model, even though it made the migration costlier. In the old system everything hung off the contact. Copying that shape made the data move direct and cheap; putting the trip at the center made it far costlier, because every booking needed a trip to belong to and many of those trips did not exist as records: they had to be built from loose data. We chose the custom model. All 792 bookings ended up with their trip and their vendor, without a single exception.
What the client sees and what the agency keeps. The business depends on this call: if the itinerary shows the chef's or the driver's name, the client can skip the agency next time and the vendor relationship — the actual asset — is given away. But hide the restaurant's name and the client cannot get where they need to be. Solving it case by case does not scale across hundreds of bookings entered by different people. So we wrote a rule: the vendor is shown only when the client has to travel to them — a restaurant, a villa, a yacht — and never when the vendor comes to the client. There, the title is the service: the meal, the route, the vehicle.
A chef and a helicopter do not ask for the same data. A villa needs check-in and check-out; a transfer needs origin, destination and flight number; a chef needs menu and allergies. An off-the-shelf system forces a choice between one form that serves nobody and a free text box where everyone writes differently. We gave each category its own fields, and every service type carries the same color throughout the system. The data stopped being a hand-typed sentence and became something you can filter, sort and total.
Moving eight years of operation without the business stopping for a day. The risk was not losing a record: it was switching the new system on halfway and working worse than before. We audited the whole account against its backup before writing a line, ran the full import dry — 5,171 rows — and turned nothing on until everything was in and verified. Along the way came fine repair: 105 trips arrived split across two records, 9 had the year mistyped — one showed travel through 2028 — and 2 departed before they arrived. All three were corrected before the system was switched on.
It should not look like a control panel. This is a system opened first thing and closed last thing. We ruled out the default look of every internal tool — system typeface, grays, borders — and used a paper ground, a single accent color and the agency's own brand typeface throughout, tables included.
A table row never runs to two lines. Status goes in a color or a pill. The easy way out when data does not fit is a second line, but with rows of different heights a list stops being scannable, and scanning lists is what people do all day here. The list scans top to bottom without stopping, which is what people do here all day.
The client gets a link, not a file. A file is born stale: change a time and the client is still looking at the old paper. The link reads from the database every time, so what they open is always what is true today. The vendor gets their own and sees nothing but their own bookings.
A trip's value is calculated, not typed. Fees plus commission from the bookings that actually happened. A hand-typed number cannot be totalled, filtered or audited: in the old system, 335 of 597 commissions were a sentence like "add 20% to the net price". Today all 597 are a number that totals itself, and commission is part of the model rather than a note in the margin.
The system now coordinates 2,572 contacts, 792 bookings and 1,067 trips, sharing its design and its login with the agency's public site. Open one card and the whole trip is there with the status of every booking. The client receives an itinerary that updates itself, the vendor sees only their own work, and quotes are built line by line with the total calculated.
Fields per service category, commission inside the model, and documents with their own rules about what they reveal do not exist in an off-the-shelf system: the model is this business's own, and the same model fits any operation that coordinates vendors on a client's behalf and charges for it.
An agency selling villas, yachts and private aviation had a site that read like a classified ad: heavy newspaper-headline type, content that could not be edited without going into the server, and a catalogue of hundreds of listings that existed for whoever already had the link and for nobody else.
The underlying problem was not aesthetic. The site did not match what the agency charges. Someone about to spend on a villa in Ibiza compares, without saying so, against the hotel and luxury-brand pages they already know. If the page looks cheap, the conversation starts at a disadvantage and the price feels high even when it isn't.
Translating the word luxury into concrete decisions, without the category's worn-out vocabulary. Luxury on screen has its clichés — gold, magazine serifs, dark grounds — and using them makes a brand look like every other. At the same time it had to look unlike the ordinary agency, which communicates volume: grids crammed with offers, discount badges, twenty filters at the top. We fixed three words before touching a single screen — calm, care, judgment — and validated every later decision against them instead of against taste.
One search criterion served several product types, and the structure would not allow it. The old site forced every filter to belong to a single category. But "Zone" is exactly the same criterion for restaurants, beach clubs, villas and experiences: duplicating it five times means five lists that drift apart and a search that returns different results depending where you entered. We redesigned the relationship so one criterion can belong to several categories at once. Whoever runs the catalogue defines "Zone" once.
The old catalogue's filters did not match between one place and another. There were 2,487 values across 17 filters and 180 options, and the mapping looked derivable from the name but was not. Assuming the obvious rule would have loaded values under the wrong criterion, and nobody would have noticed: the listings look perfect either way, and the error only surfaces when a client filters by one zone and gets places from another. We built the mapping by hand, one by one, before loading a single value.
Separating what must be found from what must never appear. The agency sends each client their itinerary through a private link. The reflex is to shut the search engine out, and that reflex does the opposite of what you want: block access and the engine cannot read the instruction not to publish the page, so the address can show up anyway. We did the counter-intuitive thing: those pages are deliberately left readable, precisely so the instruction arrives and is obeyed. And no measurement tool ever loads on them: a client's itinerary is not measured.
One typeface family across the whole site, rounded strokes and soft terminals, worked through its weights. We ruled out the usual two-family pairing and, above all, the newspaper headline that came before. A thick headline shouts, and a brand selling villas and yachts does not shout. Every extra family is also weight the visitor downloads before seeing anything.
A paper palette, not a screen one: warm white ground, an almost-black-but-not-black ink for text, two grays for secondary matter and a single accent, a deep jade green. We ruled out gold, the industry shortcut, which on screen muddies more than it adds. One accent forces everything marked to be marked for a reason: if everything can be highlighted, nothing is.
Fewer listings per screen and more room around them. On a catalogue page the temptation is to fit one more row above the fold. We did the opposite. Air is what tells the eye that what it is looking at was selected, not accumulated — and that is exactly what the agency sells: selection, not volume.
The public site and the admin panel are one piece, with one login and the same visual language. The short path was to leave the site alone and build administration separately; whoever loads the catalogue is in it every day, and two systems mean two places to look and content that drifts apart. That panel later became the base the CRM was built on.
We migrated the whole catalogue: 563 experiences with 3,612 photos, their sections and their filters, verifying that not one listing pointed at an empty place. The photos still lived on the old server, so we made a second pass to bring them across: all 3,612 now sit on the new side and the old site can be switched off with no consequence.
The agency edits its content from its own panel, without touching the server. And the map the site hands to search engines builds itself from the catalogue: every destination, category, experience and article enters without anyone adding it by hand. Load a new experience and it appears. That was the exact point where the old site failed silently: all 563 catalogue listings became findable, and every one added later enters on its own.
VEO Creativo is a creative agency in Tucumán: eight services — brand, strategy, communications, film, advertising, design, web and automation — for mid-sized companies in fields with nothing in common.
A creative agency's website has a problem others do not: it is its own portfolio. If it does not look good, it has already answered the question the visitor came to ask. And it has to answer it before they read a single word.
Proving design judgment without the project photography. The usual way to earn trust in this category is the photographic record of work done, and that material was not ready. We loaded the credibility onto what we did control: a very large type scale held by a rounded, thick-stroked face, the brand's own palette applied in blocks of solid color, a lot of air, and a single motion sequence at the opening. The site demonstrates judgment instead of announcing it: the composition stands on its own, and the photos drop into place without anything being redone.
The owner asked for flat, and it still had to be memorable. Her words were "if it's flat, better". An agency site usually leans on effects to get attention; strip them and all the weight falls on composition, scale and color, which is exactly what looks cheap fastest when it is badly resolved. Five devices were ruled out — striking through text on scroll, lighting words up like karaoke, tilting images in three dimensions, numbering the services, ringing the brand mark with a halo — and we concentrated all the motion into one opening sequence. The impact comes from the scale of the headings and one section in solid color.
An opening that has to land exactly on the logo, on any screen. This is not an animation that ends anywhere: a piece of the brand mark spins while four words appear one at a time, and at the end it must fall precisely onto the header logo. A small drift ruins the effect and reads worse than no animation at all. On phones the target moves too, because the browser bar hides on scroll and shifts the landing point mid-sequence. The landing point is recalculated every time the screen changes height, which is exactly when it shifts.
Two languages that do not sell the same way. Translating the text is the easy part. The contact channel is where the site converts, and the one used here is not the one a client in the United States uses: leaving the same button in both versions means half the visitors have no natural way to write. The language switch changes the channel, not just the words: in Spanish the buttons open a direct chat, in English they open a form.
Two typefaces with opposite jobs. A rounded, thick-stroked one holds very large headings without turning harsh and carries almost all the weight. The other is handwritten and appears rarely: numbers, a caption, a project name over a photo. We ruled out using it for headings too: an agency that writes everything by hand reads like a small side project. Kept to minimal appearances, it adds warmth without costing authority.
The temperature contrast, held with no transition. Whole sections in deep burgundy and then a whole section in solid orange, all at once. A cut that hard risks reading as something broken, and the obvious fix — a gradient — looked like a smudge over the orange. We held the hard cut and gave the orange a single job across the site: everything you touch. That way the color change reads as a change of section, not an accident. The three supporting colors appear sparingly so the orange stays the only strong accent.
Content never exceeds 1,216 pixels wide even when the screen is twice that, and headings arrive on their own before their text. Air is what makes an agency site read as expensive, and letting the heading land first sets the hierarchy before reading begins.
The project grid rearranges itself, two blocks at a time every few seconds, as if it were sorting itself out. The work comes from construction, technology, automotive, furniture, real estate and travel: side by side, that spread can read as a lack of focus rather than range. The motion makes the variety read as emerging order. On phones it switches off, because with less field of view the same device gets in the way.
Eight services that open one at a time — the visitor sees the full map first and the detail only of what interests them — nine projects with their category and what was done on each, and a profile of the founder. All in two languages, with the right contact channel in each.
The grid was verified with no gaps across seven screen widths, from 1,920 down to 390 pixels. Every stage was approved on the owner's own phone, through a link that opened with nothing to install, so the type scale, the air and the motion were validated first on the screen where the site is actually seen. Four rounds of corrections, all closed.
R&R Improvements is a family remodeling company in Fontana, California: kitchens, bathrooms, additions, accessory dwelling units and garage conversions. Twenty years of craft, and the whole business runs through a visit to the house and a hand-written estimate.
The proof of those twenty years existed, but it was scattered across places that never added up. They needed one place where the company could be seen whole and where someone decides to trust in a minute, from their phone.
Building trust for expensive, invasive work with seven reviews against competitors holding 78, 110, 137 and 416. The client is about to let strangers into their home for weeks and pay a lot of money. Review volume is the usual currency of trust in this trade and it cannot be acquired quickly. So we audited local competitors looking for where they were all weak, and two gaps appeared: twenty companies repeating "affordable" and "quality" without showing a single project of their own, and almost none publishing their license number. We changed the currency of proof: where we could not win on quantity, we won by being the only one showing papers and real work.
The gallery had to prove, not decorate. In a section called "Proof, not promises", a single image that is not the client's invalidates everything else. The archive had near-identical duplicates, titles that did not match the photo, and inherited material from other sources. Automatic similarity comparison cannot tell a kitchen from an addition, so the review was visual: contact sheets with the whole archive grouped by category, gone through in batches. Three duplicates surfaced that no tool had flagged, and the hard rule stood: only their own work, with anything doubtful set aside for the owner to confirm before use.
Fitting 55 projects into a page that can neither be slow nor overwhelm someone opening it on a phone. Visual proof wants quantity and the experience wants brevity: two opposing demands on the same page, and most of these visits arrive on mobile data. The gallery opens with six and adds six more when the person asks, with filters by type of job so whoever came to see a bathroom does not scroll through additions. Total photo weight dropped from 50.2 MB to 5.4 MB, none above 250 KB, leaving untouched the ones that already qualified rather than running them through the same mould.
Designing for an owner who does not work through digital tools, and still converting. The expectation of a modern site pushes toward online booking, forms and chat. But an open channel nobody answers is worse than no channel, because it loses exactly the person who had already decided to move forward. We ruled out online booking and concentrated everything on the phone, the one channel he actually answers. The site, the services, the FAQ and the contact all point to the same number.
A weighted serif for headings and a neutral sans for body text. The comfortable option — and what the competition does — was a single sans throughout, and that left the site reading like a template, which is the opposite of what you need when the sales argument is twenty years of craft. The serif gives the headings age and weight; the sans keeps body text readable on small screens, which is where it is actually read.
A scale of blues built on the logo's own blue, with a very dark blue for text instead of pure black. Black on white hardens reading and does not belong to the brand; the deep blue keeps the contrast you need and makes even body copy part of the company's color world.
The license number up top, in plain sight, repeated at every decision point. Almost every local competitor leaves it in the footer's fine print. It is the cheapest thing to show and the most reassuring to someone about to let people into their home for weeks.
The first six photos are the site's entire promise. The order photos arrive in is never the order to show them in. The gallery opens with the six strongest and most varied — marble island, marble shower, finished exterior, kitchen with professional equipment, the crew setting the countertop, a modern bathroom — then rotates between categories to show the range. Work in progress demonstrates activity, but someone about to spend serious money wants to see the end, not the middle.
A service gets no showcase until a photo exists that does it justice. The alternative was filling the grid with whatever was at hand — timber framing, a modest hallway — and a showcase built on work in progress lowers the perceived level of everything around it. The exact list of shots each section needs was written up for the next photo session.
The site is now the only place where the company can be seen whole. The homepage opens with the license number and the credentials in plain sight, the nine services are described in the words people actually search with, and the gallery shows 55 of their own projects across six categories with filters and progressive loading. A before-and-after comparator drags under your finger: it turns the proof into something the visitor operates with their own hand rather than something they are told.
And the ratings scattered out there appear gathered into a single band: 48 reviews across three platforms, scores in plain sight, all in one place. Reputation stopped depending on which door the person happened to knock on. The owner sends the link to anyone who calls, and that person sees, in a minute and from their phone, twenty years of finished work, proof that the company is legitimate, and one number to call. The gallery is built to grow: each new batch of photos enters its category without redesigning anything.
TR Fit is a strength-training app built around a real coach: you get a plan built by hand and, at the end of every set, you declare how hard it felt. That changes the weight of the next set. The site exists to get downloads, not to sell the coaching service the old site sold.
The product's differentiator is not a screen: it is a decision the user makes after every set. A video shows a screen, not a decision. And that was exactly what had to land in three seconds.
There was nothing to show the product the site existed to sell. Of the twenty-eight files delivered, fifteen were unique and five were usable: four app screenshots and the logo. The six presentation videos showed interfaces that are not the app's and garbled words where it should have said Consistency or History: none was publishable or salvageable by editing. Rather than filming the app or generating it again, we rebuilt its interface inside the site itself, screen by screen, using the four real screenshots as an exact blueprint. What the visitor sees is the actual app, with its real text, sharp at any screen size and with no separate version for phones. Production cost of the material: zero.
Turning the user's decision into something understood without reading. We made that sequence the site's signature section and tied it to scroll in four stages: the finished set appears, the effort selector lights up, the next set's weight changes on screen, and the rest timer starts. The choreography was audited at ten points along the scroll, on desktop and phone, so no stage is left half-finished and there are no dead zones where scrolling advances and nothing happens.
An app built around a real coach does not sell as software: it sells as a person. The coach was explicit: he builds every routine by hand, based on goal, age and available time, and talks to every client. We rewrote the whole site in first person — the app became the vehicle, not the star — and the signature section took his words, progressive overload, instead of the industry's jargon. In parallel we asked for real material with a precise brief rather than a generic request: side lighting and never frontal, because frontal flattens muscle; no selfies; every scene shot horizontal and vertical, because the site is seen more on phones. Two photos and eighteen technique videos arrived, all real, and they are the ones on the site today.
Deciding what to buy generated and what to build by hand, before spending. The brief for the opening piece had five parts: the phone rotating with the app running, the stamped logo, a droplet rippling the background, light streaks and better atmosphere. We separated what can hallucinate from what cannot: three of the five were built by hand, because there the app you see is the real one, the logo is the logo, and sharpness does not degrade. We only bought matter that represents nothing and therefore cannot be wrong — dust, light, optical flares. And we wrote down before spending a ceiling of pieces and a cut rule: three probes, and if two were not usable the whole route closed. Only what passed that test made it onto the site.
Headings in a wide, heavy display face; body in a modern neutral. The display shares the DNA of the brand's monogram — geometric, wide — and gives the product the athletic hit it needs. We ruled out the old site's typeface and its wordmark as a source of style: both belong to the brand being replaced. The two complete families weigh 90 and 29 KB.
A dark site, with the palette sampled from the app itself. We did not invent a palette or reuse the old site's. Here dark is not an aesthetic trend: it is the real identity of the product being sold. And there is a single accent, measured: on the app's actual screenshot, the blue runs 16 to 1 over the gold — 6.30% cool pixels against 0.38% warm — so giving gold a whole section would have inverted the brand's real hierarchy for a detail that occupies less than half a percent of the product.
The phone screen always renders dark, whatever the section behind it. The site alternates dark and light backgrounds so the scroll has pulse without introducing a second color. The finding was that a dark phone on a light section stands out far more, so the product becomes the focal point with no effort at all. And it always paints dark because on a light ground the interface would turn white and we would be showing a product that does not exist.
Asymmetric composition, with the phone large and cut off by the edge. A centerd composition reads like a slide. The phone nearly doubled in size and we let it run off the frame. The heading stays for the whole section instead of fading: one that stays is an editorial pattern, one that appears and leaves is a billboard. The frame reproduces a specific model rather than a rounded rectangle, because the client's screenshots are exactly that model's resolution: the proportion and the side controls are data, not taste.
The thirty-week bar leaves the phone and becomes a full-width graphic. Inside the screen each week measures six pixels: illegible, precisely when the point of the section is that thirty weeks should feel like a lot. Outside, it also gives breathing room after a section that is all device frame.
The cover photo runs whole, cleared with a linear gradient. The obvious route was to cut the figure out so it floated clean on black, and that cost three things that were adding up: the real gym behind him, which is credibility; the depth of field, which is depth; and the hair edges, where every cut-out fails and it shows. His shirt's black is also practically the site's black, so the torso melts into the ground and what emerges is the face, the arms and the logo on his chest.
The site is live, complete end to end: a homepage with the coach's real photo, six product sections, a closing and a footer. It replaced the previous site, which talked about a business that no longer exists. Editing is design too: the page was cut from 21.6 to 14.4 screens tall before publishing, removing everything that repeated something already said.
Everything you see of the app is the real app, rebuilt from its screenshots, so it reads sharp on any screen and there is no separate phone version. And the download button had an underlying problem we solved rather than hid: the app is exclusive to people already in the coach's service, so the site says so in its FAQ and adds a second path, a direct link to message him before deciding.
The work did not end with the site. The business coaching continues to this day, and over that period the project went from 20 to 100 monthly subscribers in a matter of weeks.
Affordable Document Services is a legal document assistance practice in Riverside, California, founded in 1997. Its owner is a registered and bonded Legal Document Assistant: she prepares and files with the courts the documents someone needs to get divorced, settle an estate, start an eviction or transfer a property, without being an attorney and without giving legal advice. Her client is someone in the Inland Empire with a concrete legal problem who cannot afford a lawyer.
What existed before worked like a business card: name, phone and little else. It did not explain what a Legal Document Assistant is, did not list the filings, and offered no way to make contact other than calling. Twenty-nine years had built a reputation that lived in past clients' memories, and when that network was not producing inquiries there was no other channel. What the client asked for, in her words, was: "that people understand I am not a lawyer but that I know what I am doing". That sentence became the standard for the whole piece.
Selling a category nobody knows exists. Most people believe there are two routes: pay a lawyer or manage alone. The middle category is unknown, and nobody searches for what they cannot name. We structured the site around the moment of need — "I'm getting divorced", "my father died", "I need to evict" — rather than around the profession, with the mechanism explained in four steps from the homepage and the content entering through the question, not the service.
Building authority without being able to promise anything. Regulation forbids guaranteeing outcomes, forbids giving advice, and on top of that requires a disclaimer on display. A site that builds trust normally leans on promises, and here they were off limits. We rested the authority on verifiable facts — year founded, registration number, bonded status, the fact that she files the documents herself — and designed the mandatory disclaimer as a visible, well-composed credential rather than fine print. What the law obliges you to say ended up working as a sales argument.
The forbidden word. Spanish carries a specific trap: "notario" means a lawyer across most of Latin America, and California law forbids a non-attorney from using it to advertise. But it is exactly the word the Spanish-speaking client searches for. We solved it by always naming the act — "notarización" — and never the person, auditing every line of the Spanish vocabulary against that restriction.
Two languages without splitting the site in two. A switch that changes the text on screen is invisible to search engines: there is one address and the Spanish never gets indexed. Duplicating the site doubles maintenance and risks a duplicate-content penalty. We gave each language its own address, made every page arrive already written rather than assembled in the browser, and had each version declare its counterpart. Twenty-seven indexable pages, seven of them in Spanish, one single piece to maintain, and a weight barely six per cent above the monolingual site.
Deciding where to invest with no historical data. There was no prior analytics and no idea which service generates the most demand, and investing in advertising like that is gambling. We did it the other way around: first the measurement of inquiries and calls, then a research programme — demand by service and by language, local competition, seasonality, cost benchmarks for the trade, exclusion lists and the regulatory frame for legal advertising — and only then the strategy.
Type. Headings in Hedvig Letters Serif, a single-weight serif, with 1.05 line height. Body in Libre Franklin, 300 to 700, at 1rem with 1.65 line height. We ruled out a heavy slab serif, which reads as litigation and aggression, the opposite of the positioning; and ruled out solving everything with one sans, correct but with no signal of longevity. The serif carries the twenty-nine years, the sans the clarity of paperwork.
Colour. A cream ground, #FAF7F1, with two further steps to separate sections. Institutional blue #003087, with a deeper variant for dark grounds. And a gold, #F0B000, as the single accent, with a darkened version for when it works as text on light and needs legible contrast. We ruled out pure white with cool grays, the category default, which produces clinical coldness: the cream makes the page read like paper, which is literally the product. The gold enters as warmth offsetting the hardness of courtroom blue.
Form, space and rhythm. A corner radius of 2 pixels across the whole system: we ruled out the large rounded radius, which reads as a consumer app and erodes the institutional signal. A 1200-pixel container with elastic side margins between 20 and 64 depending on width. And a single repeated element orders every section: the eyebrow in caps, with 0.22em letter spacing and a 34-pixel gold rule to its left.
Motion. Blocks arrive with a 26-pixel vertical shift over 0.8 seconds, staggered in 80-millisecond steps. The top bar is 84 pixels and moves from transparent over the photography to translucent cream with blur on scroll, switching the logo from white to blue in the same gesture. The rule, in one line: motion should accompany reading and never interrupt it.
Photography. Real rooms and real people. We ruled out the category's visual repertoire outright — gavels, scales, columns, shelves of legal volumes — because it is lawyer iconography, and the whole project rests on the distinction that she is not one.
The practice went from a digital business card to twenty-seven indexable pages in two languages, with its six services explained in detail and twelve articles answering the questions people ask before hiring. The Spanish is not a hidden translation: it is seven addresses of its own that a search engine can show to someone searching in that language, in a county where a large share of the market speaks it.
What also remains is a documented visual system — palette, type, spacing scale, photographic and motion criteria — that allows pages to be added without deciding anything again. And for the first time in twenty-nine years the business can answer what was previously impossible: how many inquiries arrive, through which channel, from which service and in which language.
DeLapp Builders is a family remodeling company in Riverside, California, run by the third generation of contractors in the family since 1981. Kitchens, bathrooms, cabinetry and full home remodels, with one crew and one project at a time. It lives on word of mouth: the owner's personal reputation is, literally, the product.
There was no website. Its only public presence was a handful of loose photos on a social page, with no brand criteria, plus its Google reviews. That gap cost the business money: every inquiry depended on someone already knowing him, with no way to show the real volume of work or build trust with anyone who had never heard of him. The brief was direct: a site on the level of another one he already liked, with an identity of its own, not a generic contractor template.
Ordering more than a hundred jobsite photos without turning them into noise. The family handed over an archive of a hundred-plus real photographs, shot on a phone and in no order. The problem was not showing them: it was deciding which ones tell the right story, keeping three angles of the same room from saturating the grid, and keeping the page light on a phone. Two curation passes — first discarding for quality and duplication, then classifying by room — produced a gallery of 54 photos in four categories, shown eight at a time.
Making the language match what the business can back up. A licensed contractor cannot afford a site that over-promises. The review surfaced precision fixes: a service listed but no longer part of the business was removed entirely, the description of the cabinetry work was corrected to reflect exactly how it is built and installed, and the way the crew is described was adjusted to how each job is actually staffed. Every one of those sentences changes what a client expects the day the contractor walks into their house, so it was reviewed line by line until the text said exactly what the business stands behind.
Building enough trust to let people into your home for weeks. It is a decision you cannot test before making it, and the category's language is worn out: everyone claims to be the contractor you can trust. We built it in concrete layers: the state license visible from the first scroll, three generations of family continuity as the central brand argument, and real reviews with none invented.
The identity already existed, and the family did not know it. They informally used a drawing of a wizard with a wand, with no system around it. Rather than design a brand from scratch, we took it as the starting point and built it out into a full identity. The palette followed the same logic: the initial sage green was dropped and the cobalt blue came from a blue-and-white Spanish tile in one of his own kitchens — the very photo the family had already chosen as their page's cover. The brand's color was inside the work.
Capturing inquiries without leaving the client tied to whoever built the site. From the start it was clear the site would end up on the family's own server, not on one of mine. That put a real constraint on the contact form: it had to keep working without depending on any outside tool they could not operate later. We solved it with a self-contained submission system, hosted inside the site itself, delivering each inquiry straight to the business's inbox, with no third-party accounts to administer.
Handing over control without handing over a construction manual. The family chose to run the domain and the server themselves. That meant leaving the site working on their own hosting with enough instructions to install future updates, without turning the documentation into a course on how every piece was made. It closed with the site migrated, verified on the final domain, and a guide kept to the operational essentials.
Type. Headings in University Roman, thin-stroked and full of character, chosen because it connects directly to the wizard mark. Body in Barlow, a neutral sans, so long paragraphs — reviews, family history — read light against a heading with that much personality. It is the opposite of the generic sans almost the whole category uses.
Colour. Cobalt blue #2D5F86, with a light variant #6FA3C7 and a dark one #1F3F5C, over a warm base of linen and cream (#F5F1E6 and #FBF8F1). Two dark blocks break the journey and give a long page room to breathe. The first palette's sage green was dropped: the blue was not chosen by taste, it was extracted from the client's own work.
Photography. Zero stock and zero recreations: every image is a real finished job, cropped to a consistent 4:3 so the grid reads tidy without losing the documentary character of a photo taken on site.
Motion. The homepage rotates through four photos with a soft crossfade every six seconds and a slow zoom on each, so the first impression does not feel static. The rest of the page reveals with a gentle entrance on scroll. All motion switches off for anyone with reduced motion enabled on their device.
Structure. One long narrative page: hero, track record, services, the family's method, kitchen specialism, gallery, history, reviews, coverage area and close. Content width was fixed at 1240 pixels so the photos breathe without losing reading intimacy.
DeLapp Builders went from having no site to a complete online presence: a curated gallery of 54 real photos organized by room, an inquiry system that arrives straight in the business's inbox with no manual steps, and a brand identity built from a symbol the family already felt was theirs.
The site is published and running, hosted and administered entirely independently by the family. That was the goal from day one, and it shaped decisions taken months earlier: the form that depends on nobody, and documentation written so they can carry on alone.
Ebmeyer's Plumbing is a family plumbing service in Apple Valley, California. It is run by one person, licensed by the state since 1994, across a wide, low-density area covering around ten towns in the High Desert. Leaks, drains, water heaters and pipework: emergencies and scheduled jobs alike.
Before, there was a site built on a generic template: a wine-colored top band, a background of pale blue waves with no relation to the trade, a circular "call now" badge, and three loose links over plain text. It did the job of being online, but it communicated nothing about the trade or the person behind it. The business needed something more urgent than aesthetics: a way to turn someone with a real household problem into a phone call, at the exact moment they are deciding who to call.
Urgency as a design principle. You look for a plumber with water running across the floor: this is a trade where the buying decision can take seconds instead of days. That meant inverting the usual order of a website. The phone number and the problem triage went above everything, ahead of any institutional introduction, and everything a hurried visitor does not need to decide right then was moved out of the way.
The system already existed, in the plumber's head. The one thing the client explicitly asked to keep was his five problem categories: the ones he already used to triage calls, because they worked in the field even though he had never thought of them as a piece of design. Instead of replacing them with a generic form, they became the site's first screen. Each category sends the request straight through, with the address and the urgency already attached.
Promising the price before starting, in a trade that does not. The site was built around stating the cost before the work begins, the opposite of standard practice in the trade. Holding that promise without quoting blind meant resolving it as a process — what steps follow a request — rather than as a fixed figure impossible to guarantee up front.
A one-person business, without it reading as a limitation. The same man has answered since 1994: an enormous trust asset and, at the same time, a real ceiling on capacity. We built the message on continuity and dealing directly with him, without the language of "team" or "crew" that does not exist, and without promising round-the-clock availability that is not real either. The service area was set with an explicit rule about which towns to list, rather than inflating it to look bigger than the operation can sustain.
Showing a trade that is mostly invisible. Much of a plumber's work happens under a sink or inside a wall, and there was no photographic record of real jobs. We solved it with a three-tier system: simple graphic motifs to set the scene, painted illustration to explain objects and processes, and real photography reserved exclusively for what proves the business exists — the truck, the public license, the genuine reviews. The rule was strict: illustration explains, it never simulates competence.
One of those illustrations arrived with an illegible passage in the engraving, and it had to be corrected without losing the etching texture of the rest of the piece. It was rebuilt by hand, repeating and adjusting clean sections of the engraving curve by curve, until the repair was invisible even up close.
Type. A pair of two families. Headlines echoing mid-century American commercial lettering — the same tradition the business's cursive logotype and the truck signwriting come from — with body text in a clean-lined gothic, in the vein of classic hardware catalogues: legible, unadorned, built to be read fast.
Colour. A palette held to five tones: a deep oxblood red as the brand color, a warm mustard as accent, a near-white paper ground, a dark ink for text, and a water blue reserved exclusively for the drip and leak motifs, never used as a general interface color, so it keeps its charge of meaning.
Motion. Deliberately moderate throughout, with one exception: a section that pins to the screen while the visitor scrolls, in which a water leak advances as they move down. It dramatises, at exactly the right moment, the urgency that brought someone to the site, without turning the rest of the experience into a spectacle.
Structure. Ordered from urgent to considered: the problem triage first, services next, the proof of trust — license, years, reviews — halfway down, and the contact form always one scroll away.
The site is live on the business's own domain, indexed in search, and receiving real requests through the form, which reach the owner immediately with the address and the urgency already attached.
Every stage was approved on real screens before publishing, and clear documentation of the service terms was left so there is no ambiguity about what is included and what is not. The business went from a template that merely existed online to a piece that works at the exact moment someone decides who to call.
FinancePro is a finance platform for a business that bills and pays in more than one currency. It was not a commission: it was built for a problem of my own, running personal finances alongside those of a company that bills in dollars, spends in pesos and operates in Mexico.
Before, everything lived in a spreadsheet rebuilt every month: a tab per account, one per category, one summary, and an exchange-rate column updated by hand. Three currencies, fourteen accounts, and a company whose entries had to be separated from the personal ones row by row.
What broke was not efficiency but trust in the numbers. The past would not hold still: with a single exchange-rate cell, updating it in August rewrote March, and that makes the most basic question in a business impossible, whether this month was better than the last. The close arrived well into the following month, when the decision that depended on that number had already been made blind. And with no written rule, a personal expense paid from the company account was resolved differently depending on the month.
Buying something off the shelf was not enough. Personal aggregators assume one currency and one country, and do not model a company. Accounting tools do model it, but they are systems of record: they look backwards, and their multi-currency solves the accounting conversion, not the business question. None accepts the founding premise: that the person and the company are the same economy, and you have to be able to see them together and apart without switching tools.
The currency of truth. Every entry is fixed to the exchange rate of the day it happened, and that value is never touched again. It sounds obvious, but it forces you to accept something uncomfortable: historical balances stop adding up against today's rate. The alternatives were discarded for what they broke. Converting everything at the current rate, which is what the spreadsheet did, gives balances that are coherent today but destroys comparison between months. A period average smooths the jumps, but invents a number that never occurred and makes it impossible to reconcile against a real statement. The entry's own date is the only one that allows an audit: every figure corresponds to something that happened. When today's value is needed, it is shown separately and labelled, without replacing the original.
Personal and business without a separate dashboard. Duplicating the application into two modes was the easy path, and it was discarded because half the useful questions cross both worlds. Ownership was solved as a rule of the model and not as a screen filter: an entry belongs to the company if it is marked as such or if it lives in an account the company owns. Being a single rule, no screen can contradict another. Above it sits one single scope control, and the user always knows what they are looking at because there is only one place where that is chosen.
Showing the future without promising it. Projecting forces you to separate what is committed, what is likely and what is assumed. The projection is built on what is already recorded and is always drawn dotted against the solid lines of the past: that distinction does the work a disclaimer does not, because nobody mistakes a dotted line for a fact. Beside it sits a simulator where you change one assumption, a client paying late or a new fixed cost, and see the effect. It is a tool for thinking, not a forecast to believe.
Density. A finance screen carries many numbers and turns illegible. It was solved in three levels: four large figures at the top answering "how am I doing" at a glance; below, collapsible sections that each answer one question and remember whether you left them open; and inside, the detail. The secondary is not removed, it is folded. Every section shares the same anatomy, so the position of the controls never changes.
Colour that already means something. In finance, green and red are taken: they are income and expense. That forced them out of play for everything else, so the series palette was built without either and validated as a set on white for the hardest case, a pie chart where every slice touches all the others, checking their separation under color blindness too. Where a number is good or bad depending on context, color never travels alone: a word or a percentage goes with it.
Making entry not feel like a chore. Every finance tool dies the day you stop entering things. The form proposes today's date, the most used account and the day's rate already resolved, suggests the category from what you typed, and attaches the invoice in the same step, reading its data.
Type. A single family for text and interface, Geist across four weights, with headings on tightened tracking. For every figure, Geist Mono. A display serif for headings was discarded: it gave character, but fought with the reading of dense tables.
Colour. Page ground #F6F7F9, an off-white leaning slightly cool rather than a neutral gray; cards in pure white so they float without needing a heavy shadow; ink #0A0B0F, borders #E1E4EA, secondary text #6C717F. One single accent, electric blue #0A60FF, with its wash #EBF2FF for active states. Semantics: expense #E4494C, income #0D9668. Chart series run in a fixed order and never cycle. The dark blue sidebar of the early version was discarded: it stole weight from the single accent and made everything else look muted.
The numbers. This is where the most detail was spent. Every figure is set in a monospaced face with tabular figures, so columns align by position and an amount does not change width when it updates. Amounts are always right-aligned. A negative is not shown in parentheses, as in classic accounting, because that is illegible to anyone who is not an accountant: it takes a minus sign and red. And an amount in a foreign currency is shown in its own currency, large and with its symbol, and below, in smaller type and secondary color, the dollar equivalent preceded by a ≈: that symbol says "this is an approximation as of today" and stops anyone reading it as the real amount.
Spacing and form. A 4 pixel scale with steps at 4, 8, 12, 16 and 24. Base corner radius of 12 pixels, 16 on cards and 8 on inner controls; every form field shares one radius so none looks like it came from another family. Minimal shadow: the border separates, not the shadow.
Navigation and motion. A fixed sidebar with three groups instead of one long list; top tabs were discarded, since with nine destinations they forced an overflow menu. Of motion, almost none: 150 ms color transitions and the section expansion, with no entrance animation on the numbers, which in a tool opened twenty times a day only gets in the way.
FinancePro has been in daily use since March, without interruption, holding three currencies, fourteen accounts and the separation between personal and company in one place.
The spreadsheet rebuilt every month no longer exists. Closed months stopped moving, and one closed today is worth the same a year from now, which made it possible for the first time to compare periods, from weeks to years, on one chart. Whether the business is profitable on its own stopped depending on a row-by-row review and became a rule of the system. And the month close stopped being a spreadsheet chore: today it is a report that generates itself with its receipts.
The domain prosperityworksllc.com was already registered and the client's email was running, but nothing was built on it: typing the address in landed you on the registrar's parking page.
There was no old site to rebuild and no structure to respect. What the business says, who it says it to, and in what words all had to be decided from zero.
It was not "they have no site". It was two things, and the first one shaped the vocabulary of every page.
The natural language of the field —accounting services, financial modeling— is reserved by law for licensed accountants in a number of states, and she serves clients nationwide. The site had to describe precisely what she does without crossing that line, in a field where the everyday words are exactly the regulated ones. This was not a question of tone. It was professional scope.
The second surfaced in a follow-up meeting. Nonprofits appeared as one of seven client types, in a list at the bottom of the page. But that work —grant tracking, restricted funds, reports a board and a funder both have to be able to read— is substantially different from work with a small business, and that difference showed up nowhere on the site.
The first typographic direction, editorial in tone, was dropped: the client felt it did not set her apart enough. A panel of six complete visual directions was built, and the one chosen came out of it.
Zero stock photography anywhere on the site. The one real photo, the founder's, was held back for a single section, after six blocks built from nothing but type and colour. A real face lands on its own, with nothing around it competing for attention.
The green field on the homepage, where the headline is cut out of the colour rather than painted on top of it, is the only full-bleed solid-colour surface on the entire site. If that treatment repeated anywhere else it would stop being a moment and start being a pattern.
"Who we help" went from a list at the bottom of the page to two columns of equal weight, businesses and nonprofits. The client spotted it; how to solve it was mine.
Every colour pairing was measured for contrast before it was used. One variant putting the brand's gold on small text fell short of the minimum and was dropped. The brand colour was not eyeballed off the logo either: it was sampled from the pixels of the original file.
And a visible line in the footer sets out the scope of the service, for the same reason as above. It is the kind of sentence nobody asks for and that protects the client.
Thirteen pages. A homepage with eight sections, a questions page with 37 questions across six categories, an article index and ten articles.
The ten were split across the two audiences without overlapping in topic, with each other or with the questions page, and they link to one another: each one closes with two more. The homepage weighs 23 KB compressed; the article index, 20; the questions page, with all 37, 33.
The site was connected to the domain in October and has not gone out to search engines yet, so there are no traffic numbers to show.
An off-the-shelf CRM is built to sell: everything hangs off the contact and the closed deal. This one is built to coordinate. The central object is the trip, with all its bookings hanging off it, each with its vendor, its time and its commission. It is the difference between a sales funnel and an operation that has to be executed without dropping anything.
Every booking on a trip in one screen, with its status. Nobody opens five records to see how a weekend is going.
A villa needs check-in and check-out; a transfer, a flight; a chef, menu and allergies. Each type shows its own fields and nothing else.
Each service with its net price and margin, and the total calculates itself. Nobody types a number by hand.
A trip’s value comes from fees plus commission on what actually happened, not from a field someone fills in.
Each opens their own and it updates itself, with rules about what is shown and what the agency keeps.
A client and a vendor are not the same record. Without that you cannot measure, filter or bill properly.
A spreadsheet with several currencies lies without meaning to: every time you open it, it converts everything at today’s rate, and March’s result changes. This platform fixes every entry to the rate of the day it happened. The past holds still. And on that base that does not move, instead of explaining the month that ended, it shows you the one coming.
Every entry is fixed to the exchange rate of its date. You close March once, and March never moves again, even if the dollar jumps in April.
Your accounts and the company’s live in one place without ever merging into the same figure. You look at the company, at yourself, or at both.
Not only what happened: whether you make it to month end, which week the hole opens, and what changes if a client pays late or you add a fixed cost.
What you planned and what you spent, by category, on one screen. It flags where you are overrunning while the month is still running, not after it ended.
The invoice stays attached to its entry and the system reads its data instead of you typing it. When someone asks for backup, nobody goes hunting.
The package for the accountant, with the period’s entries and their receipts, assembles itself. Nobody rebuilds the same spreadsheet again.
A local business almost always enters the fight outnumbered on reviews. R&R had seven; its competitors had 78, 110, 137 and 416. Winning on volume was not an option, so we changed the currency of proof. A site showing your own finished work, verifiable credentials and the phone where clients look for it does not compete on praise: it competes on evidence.
Before a line is written, we audit what the rest of the trade shows in the area. That is where the winnable ground appears, and it is rarely the one the owner assumes.
55 projects across six categories with filters, six at a time. Visitors see finished work and choose to look at what resembles their own job.
One client arrived with over a hundred job-site photos shot on a phone, in no order. Two passes left 54 across four categories: what proves stays, what repeats goes.
In one case online booking was dropped because the owner does not handle digital tools, and everything went to the phone. A channel nobody answers scares off more work than it brings.
A plumber’s triage is the five categories he already used on the phone. Each sends the inquiry with the address and urgency already in it, so the first call does not start from zero.
One site’s photos went from 50.2 MB to 5.4 MB with no loss of detail. A trade’s site gets opened from a job site or a parking lot, not from an office on fiber.
An ordinary website explains what a brand does. This one has to prove it before anyone reads a word. When you compete on design, the site is the first piece of work a client sees: if it looks generic, it already answered the question. Its own visual direction, motion used once, and one piece that exists nowhere else.
Colour orders, it does not decorate. In one site the blue outweighs the gold 16 to 1, measured on the product’s own screenshot. In another, orange marks one thing only: everything you can touch.
In legal services, regulation forbids promising outcomes. The required disclaimer was designed as a visible credential instead of fine print, on cream paper with 2 pixel corners so it never reads as a consumer app.
For one agency, five effects were cut so one could stay. The opening lands exactly on the header logo, and recalculates that landing point every time the screen height changes.
In one site, scrolling moves the visitor’s decision through four stages, checked at ten points along the way. It is the part no template produces.
One client had nothing to show their app with. It was solved inside the site itself: sharp on any screen, no photo shoot, zero production cost.
Switching language does not just switch the text. In one case it switches how they reach you; in another, each language has its own address. Two markets, not a translation.
This does not start from the solution. The problem gets defined first, and only then does anything get built. Before a line is written, whatever already exists gets audited: in one CRM the screen showed 16,456 tasks and the backup held exactly one. Starting there changes what gets built, and sometimes changes whether it is worth building at all.
On one site, 2,487 filter values were mapped by hand. The obvious rule would have loaded everything under the wrong criterion, and nobody would ever have noticed.
When a decision can be measured, it gets measured, and stops depending on anyone’s taste. That is what avoids the three meetings that go nowhere.
Before spending on production, we define how many pieces are in and when we stop: three attempts, and if two usable ones do not come out, that path closes. The budget has a written limit, not an intention.
792 bookings moved with their trip and their vendor, without a single exception. Before anything was touched there was a dry run over 5,171 rows. The business did not stop for a day.
On a site that was going to end up on a family’s own server, the form was built without depending on outside services. Nobody inherits someone else’s account that stops working one day.
The system is written down: palette, type, spacing scale and usage criteria. The team adds new pages without re-deciding anything, and without calling me again.
