Original research

30 web-provider sites serving Colombia: a mobile study

What should you verify before commissioning a website? We tested a fixed sample of 30 public provider sites and published the observations, failures and method. Use the evidence to ask better procurement questions.

Read in Spanish

Turn evidence into a project brief

Choose one customer task and agree how the finished website will be verified. Inspect delivered work, define the scope and bring the current URL into the conversation.

01

The observations, with their denominators

On 27 September 2026, we attempted one mobile Lighthouse run for each of 30 selected homepages. Twenty-six produced usable reports; four remained unavailable. Eighteen of the 26 measured pages had a simulated Largest Contentful Paint above four seconds in that run. Automated contrast checks flagged 17 of 26 pages, and link-name checks flagged 11 of 26. These are page-level observations from a small, deliberately selected cohort. They are not national prevalence estimates, a ranking of suppliers or evidence about the quality of a provider's client work. The table retains every selected site, including unsuccessful measurements, so the reader can inspect exactly what supports each count.

27.09.2026 · One run per siteMobile lab loading, across the fixed cohort
26 / 30usable reports

Largest Contentful Paint (LCP), in seconds. Counts include unavailable attempts.

  • Up to 2.5 s2 / 30
  • Over 2.5 to 4 s6 / 30
  • Over 4 s18 / 30
  • Unavailable4 / 30

One simulated mobile run per homepage. These are lab observations, not field Core Web Vitals or a supplier ranking.

Automated accessibility checks flagged

Each fraction uses only eligible binary results for that check.

  • Contrast17 / 26
  • Heading order12 / 26
  • Link names11 / 26
  • Image alternatives2 / 26
  • Button names2 / 21
  • Form labels0 / 9

An automated pass is not a WCAG conformance assessment. Not-applicable results are excluded, not counted as passes.

Inspect all 30 observations

Alphabetical order. LCP is rounded for display; exact values, warnings and settings are in the download. Contact channels reflect delivered HTML only. A dash means unavailable, never zero.

Selected homepages, 27 September 2026
Provider / sourceLCP (s)ReportHTML contact signals
Acuarela WebSelection source2.73AvailableContact link, Email
Agencia 724Selection source—UnavailableUnavailable
ArtedinámicoSelection source8.81AvailableContact link, WhatsApp
BangluxorSelection source9.3AvailableContact link, Email, WhatsApp
BarranquillaWebSelection source2.85AvailableContact link, Email, Telephone, WhatsApp
BranchSelection source6.8AvailableContact link, Email, Telephone
Cangrejo DigitalSelection source3.04AvailableContact link, Email, WhatsApp
Contra StudioSelection source7.1AvailableEmail
CREA StudioSelection source4.16AvailableWhatsApp
Creativos WebSelection source2.75AvailableContact link, Email, Telephone, WhatsApp
DAMOSSelection source9.3AvailableContact link, Email, Telephone
Deconma DigitalSelection source3.36AvailableWhatsApp
DistecnowebSelection source13.49AvailableEmail, Telephone, WhatsApp
Dweb3dSelection source11.68AvailableContact link, Email, Telephone, WhatsApp
IDEOXSelection source3.44AvailableContact link, Email, Telephone, WhatsApp
INJOESelection source9.18AvailableContact link, Email, WhatsApp
Juan C. TamaraSelection source4.61AvailableContact link, Telephone
Medellín Web StudioSelection source1.8AvailableContact link, Email
MillorsSelection source—UnavailableUnavailable
MKS PublicidadSelection source25.19AvailableContact link, Email, WhatsApp
MoobaSelection source16.18AvailableContact link, WhatsApp
Orbit Lab CoSelection source4.24AvailableContact link, Email, WhatsApp
Página Web ColombiaSelection source8.04AvailableNone detected
PANCA StudioSelection source1.37AvailableContact link, Email, WhatsApp
Platino WebSelection source5.31AvailableContact link, Telephone, WhatsApp
SierrezSelection source5.57AvailableContact link, WhatsApp
SingleClickSelection source6.24Available · redirect warningContact link, Telephone, WhatsApp
SodecompanySelection source—UnavailableUnavailable
TuataraSelection source—UnavailableContact link, Telephone
WebMedellin.coSelection source7.57AvailableContact link, Email, WhatsApp
02

How the sample was selected

We fixed the 30 URLs before collecting measurements. The manifest combines providers already documented in Dardo's official-source research with three recorded searches for web services in Colombia and its cities. Each entry includes the provider's own source used for selection. Offering services for Colombia does not establish that the business has an office in a particular city. The cohort excludes Dardo and its clients. This is a purposive sample of web-service providers, not a random sample of all Colombian companies or an exhaustive agency directory. Search discovery favors visible providers, and the chosen homepage may not represent a provider's strongest work. We therefore report counts within this cohort and avoid generalizing them to a market share, nationwide defect rate or best-agency claim.

03

Performance: turn a number into an acceptance task

Two of 26 measured pages had lab LCP at or below 2.5 seconds, six were above 2.5 and at or below four seconds, and 18 exceeded four seconds. Four attempts have no LCP observation. These buckets describe a single simulated mobile load; they are not a field Core Web Vitals assessment. A browser run can vary with hosting response, third-party scripts, machine load and the page state. Before choosing a remediation, inspect the actual LCP element and loading waterfall, then repeat on representative pages and devices. In a website brief, request a recorded baseline, an agreed test setup and a comparison after the change. The meaningful deliverable is an explained loading improvement under comparable conditions, followed by real-user monitoring where sufficient traffic is available. A category score alone does not identify the cause.

04

Accessibility: inspect the failed check, then use the page

Contrast was flagged on 17 of 26 eligible pages; heading order on 12 of 26; accessible link names on 11 of 26; image alternatives on two of 26; and button names on two of 21. The form-label check flagged zero of nine eligible pages. Seventeen reports marked that check not applicable, so zero of nine must not be described as all 26 forms passing. Automated checks cover only part of accessibility. They cannot certify WCAG conformance or tell us whether a person can understand and complete the whole task. Ask the implementation team to reproduce each issue, document the affected component and test the repair. Add manual keyboard navigation, visible focus, zoom, meaningful reading order, error recovery and assistive-technology review appropriate to the project. Preserve both the automated result and the manual acceptance evidence.

05

Contact links are an invitation, not proof of delivery

The separate HTML retrieval produced 27 usable responses. We detected explicit WhatsApp links in 19, mail links in 17, telephone links in 11 and contact-language links in 22. Channels overlap: adding those counts would double-count sites. The collector reads delivered HTML without running JavaScript, and the contact-language rule is a heuristic. A missing detection does not establish that the rendered site lacks a channel; a detected link does not establish visibility, correct routing, response time or successful delivery. No messages or forms were sent to these providers. When commissioning your own website, separately verify the complete inquiry path with an authorized test: context retained, validation usable, submission stored, notification delivered and a named person responsible for follow-up. Count a link click as intent, not as a received lead.

06

A concrete journey to inspect

The A medio tono project provides a separate example of delivered Dardo work, outside this study cohort. In our dated public check, choosing Piano opened a filtered teacher list, and opening Sara Alfonso showed instrument, location, class formats and a contact link carrying the teacher's name. That is a reproducible product decision: the visitor moves from an interest to a relevant person while retaining context. The verification record appears in the sources and the full project case is linked in the research panel. It does not establish that a message was delivered, a class booked or enrollment increased. For a comparable service business, use the same acceptance pattern: choose a service, inspect relevant evidence, retain the chosen context and reach the correct inquiry destination. Commercial outcome claims require separate, approved records with comparable periods and clear definitions.

07

Reproduce the method and understand the gaps

We used Lighthouse 13.5.0 through Google's documented CLI, with a fresh headless Chrome profile for each serial run. The default mobile setup uses a 412 × 823 CSS-pixel viewport, device pixel ratio 1.75, simulated 150 ms RTT, 1,638.4 Kbps throughput and 4× CPU slowdown. The dataset preserves each timestamp, settings, browser environment, final URL, warnings and raw-report hash. A separate HTTPS HTML request records explicit source signals. Two runs encountered Chrome security interstitials and two encountered document-request errors; they remain unavailable without retries or bypasses. SingleClick's report warns about redirection to its English page, and that warning is retained. Our initial PageSpeed API probe had no public quota, so no API or field data are included. Download the manifest, observations and collector from the research panel. Run the collector with Python, curl, Node/npm and Chrome; expect the measurements to change as sites and environments change.

08

Disclosure, correction and a usable project brief

Dardo sells web design and development services and therefore has a commercial interest in this subject. We excluded our own site and clients from the sample, kept failures visible, published the source manifest and listed sites alphabetically instead of ordering suppliers by score. These controls do not remove selection bias. Treat this edition as an inspectable diagnostic snapshot. A provider can request a correction through Dardo's contact page by identifying the URL, field, date and supporting evidence; a corrected edition should describe the change rather than silently replace inconvenient observations. For your own brief, choose one important customer task, agree which devices and languages matter, name the content and inquiry owners, and require a dated acceptance record. That gives a design proposal something concrete to deliver and a future commercial evaluation something honest to measure.

Sources & editorial notes

Sources

How we built this guide

EvidenceThis guide combines 4 linked sources, Dardo’s analysis and clearly labeled project examples.

LimitsVisible design and content decisions are not presented as measured ranking, traffic or conversion gains.

Read our editorial policy