Growth Experiment Reports

Clean SEM hit a double window low (1,748 sessions / 13 leads) and total leads a window-low 23 — Labor Day plus a five-week slide in the SoftwareExp long tail, with conversion rates holding; Direct – engaged fell a sixth straight week. Lead of the week — Juvo Technologies (BV 36.6), two serial payments founders 30 days from incorporating a fintech, asking for a PM + 3–4 devs at ~$50K/month, South America over India for the timezone.

AI / LLM posted its first zero-conversion week of the 14-week window (546 sessions) — a volume-driven drought: AI converts almost only via the homepage, whose AI arrivals fell 600→209/wk; clean Direct broke its floor (752 engaged sessions, 2 leads). Lead of the week — JUWI Inc. (BV 3.3), a Colorado solar+wind developer asking for full ownership of a PoC app a non-technical employee built with Claude Code; the intro ended on the engagement-model mismatch.

The Meta campaign started converting — two cohort leads this week, and all 4 of its lifetime conversions trace to a single creative, still filed under "Other"; clean Direct hit a window low while its excluded sub-120s slice converted for the first time in 14 weeks. Lead of the week — Minervini (BV 4.9), a profitable stock-market education platform: quote wizard in ~7 minutes behind a six-word comment, then a head-of-engineering ask ("Python just cannot keep up anymore").

The website's second conversion path read end-to-end: two QW-mirror capture flows (bot + human agent) whose only cliffs are greeting→interaction (0.85%) and the email ask — later steps complete at 87–100%. Mobile starts chats at ~3× desktop's rate but leads land 80% desktop; SEM /lp/ pages carry 65% of conversations and 67% of submissions — and 1 in 3 of their chat interactions is a job seeker. All 696 chat leads hand-classified: company buyers (29%) hold nearly all pipeline value (Rumble Cloud → 12 won openings); 20% are internal test leads, 5% recruitment-scam checks. Transcripts carry specs, RFPs and SOW links the QW never captures; pricing questions are always deflected. Inline SQL + labeled per-lead data.

MoM traffic + form-conversion on the 298 canonical SEM /lp/<slug>/ pages selling headcount in a named tech/skill (hire-<tech>-developers, <tech>-developers, <tech>-development, bare-tech), ranked by traffic. 59,586 sessions / 1,591 leads / 2.67% CR / 88% paid — ~3× the generic-staffing /lp/ CR on ~1/5 the traffic. Traffic ~halved at Feb 2026 with CR holding (42 pages went dark, 16 switched on between the two years); 106 of 298 pages produced zero leads; top-50 pages = 69% of sessions. <x>-developers pages (4.27% CR) vs hire-<x> (1.13%) on similar traffic, with 2–3 near-duplicate pages per technology. Part A family surface (MoM chart, concentration, sub-family split) + Part B per-page (24-mo sparklines + top-50 table) + Part C A/B measurability (min detectable win at pre-drop traffic, concurrent vs before/after, subset trade-off). Figures only — no revamp estimate; inline SQL.

Where copy on the 25 AI /lp/ pages that draw paid-search (SEM) traffic has fallen out of step with the current state of AI. Read from the website-frontend source, every stale item verified live in-browser. Pages are built from shared JSON templates, so the staleness concentrates in ~5 shared _sections files that render across 16 ai-solutions-template pages (7,280 SEM sessions, 72% of AI-LP SEM): the "AI Tools and Tech" wall frozen at 2023 (Mistral 7B, LLaMA 2, Stable Diffusion, MXNet, and Adept AI — a company Amazon absorbed in 2024); pre-LLM "NLP machine" phrasing; a superseded YOLOv5; and two rendered defects (an "intiaitives" typo heading + a "Schedule a call" string leaked into a skill chip). Plus the hire-AI page's 2022 client stat & non-AI case studies, and 5 llm-* pages sharing one identical success story. Balanced against what's already current (agentic / RAG / RLHF solution grids; the new ai-transformation stack). Exposure table, inline SQL, source file:line per item.

Performance + lead read of the two AI landing-page initiatives. BDWEB-7801 (AI Adoption Hub — 6 SEM LPs + SEO /ai-transformation/): ~873 sessions since mid-May 2026, SEM engagement 19% vs SEO 66%, 0 form submissions. BDWEB-6063 (AI Solutions SEM — 23 /lp/ template pages since Apr 2025): 14,104 sessions, 361 leads, 2.56% CR, ~90% Google CPC, volume tapering through 2026 (~400/mo). Plus a 6063 lead-quality layer: 202 CRM prospects, 59 first sales calls, 0 clients (vs 1,383 CRM-wide); avg BusinessValue by page (finance 4.0 / prompt-eng 3.8 high, generative-ai 0.44 / hire-AI 0.47 low); QW personas (Staff-Aug highest BV + call-reach; "build-my-product" founders the largest and lowest-value); standouts by BV and by logo-relative-to-page (NoRedInk, Solusfinance, National Fish & Wildlife Foundation, Quest Global, and ADNOC on oil-and-gas). Sources: GA4 BigQuery export + leads DB + growth_prospects/growth_voc + Gong. Companion first-call VOC read cross-linked. GA-figures only for perf; qualitative layer is descriptive.

Qualitative + quantitative context for redesigning /lp/software-development, /lp/hire-developers, /lp/staff-augmentation, /lp/offshore-development-center, /lp/software-outsourcing-services, and /lp/top-development. Per page: documented SEM keywords vs recorded search terms, 12-month sessions and channel mix, 2025+ lead funnel, every QW answer verbatim, intro-call transcripts analyzed buyer-side, buyer profiles at the top of each page (every lead assigned; per-profile BV + conversions computed from the cohort), and live-chat conversations categorized with substantive exchanges verbatim. The Software Outsourcing Services page (added 2026-06-25) covers 47 leads / 4 clients / 6 openings. The Top Development page (added 2026-06-25) is the biggest and most branded — 48,513 sessions, 320 leads / 2 clients / 15 openings, ~96% Google paid, "bairesdev" the dominant recorded search term; profiles + 14 calls focus on the 70 valuable leads, with value concentrated in enterprise nearshore-staff-aug while the 2 clients were returning buyers and startup pods drove most openings; 1,001 live chats. Two small low-yield pages added 2026-06-25 — IT Outsourcing (16 leads, 0 conversions; IT-ops/support + dev mix) and Overseas (12 leads, 0 conversions; solo founders chasing cheap overseas devs). Evidence-only — no conclusions, inline SQL.

Y Combinator logo added to the trustbar on /lp/top-development/ and /industries/startups/, deployed 2026-05-20. Matched 22-day windows: LP bounce 34.5% → 35.9% (p=0.30; placebo splits swing as far 23% of the time), CR and time-on-page flat — every metric within regular fluctuation. Startups page too small to read (~2 landing sessions/day). Placebo-split methodology, composition checks, inline SQL.

A/B engagement read (BDWEB-7460). v4 window 2026-05-06 → 2026-05-21 with an A/A noise baseline 2026-04-09 → 2026-05-05.

Dry MoM read of landing-session volume on the 109 main /solutions/<slug> and 111 main /technologies/<slug> pages (nested paths excluded), split by source bucket with SEO and LLM called out. SEO halved on both page types in Feb 2026 (solutions 201→77; tech 174→70) and has not recovered through May. LLM stable at ~11–24 sess/month per type. Includes top landing pages by SEO and LLM, source-bucket rules, leads attributable to these pages (strict via converting session + loose via any-session-touch + MarketingChannel mix — all sparse, 11 strict / 36 loose leads across six months), and inline queries. No content-update tracking — overall trend only.

CPRA + WCAG 2.2 compliance update. 6 B2B competitors (Globant, Accenture, Perficient, Thoughtworks, SoftServe, EPAM), 5 CMP references, 3 best-in-class (GOV.UK, Shadcn, DataGrail). Pattern analysis, P1–P3 opportunities, recommended direction (bottom bar + progressive disclosure + equal-weight CTAs + tertiary "Manage Preferences"), and emulate-vs-avoid guardrails. Output of a Figma design exploration.

Context for the case-studies collection refresh. Two page types in scope: the /clients/ listing (including industry sub-pages) and individual /case-studies/<slug> pages. Last 90 days (2026-02-23 → 2026-05-25). Eight chapters: volume, inbound/outbound journeys, conversion impact at same-session, multi-session user, and post-hit levels, engagement, return rate, per-study spread, industry breakdown, listing↔case-study cohort split.

Same method as the Top 1% analysis, applied to the company /about/ page (www.bairesdev.com only — the path is also served by 4 other brand microsites, 744 off-host sessions excluded). Last 90 days. 5,989 sessions = 1.02% of site (~5× the Top 1% page), flat. Well-discovered: 39% land directly, of which 29% organic search + 3% LLM/AI (vs Top 1%'s 14 SEO landings total); homepage feeds another 39%. But terminal: 62% exit (highest of the pages analyzed), only 1.5% reach the QW, with a visible job-seeker tail (careers 157 in / 300 out, contact 201 out). Conversion directional but inconclusive (9 of 5,989 hit-sessions converted; ~2× the 0.083% engaged baseline on n=9). 5-page trust-family entry comparison, opportunity-framed downstream (routing, not discovery), inline SQL.

Pure-GA before/after read of the five 2026 SEO page redesigns (Software Outsourcing, Software Development Services, Staff Augmentation, Dedicated Teams, Nearshore), refreshed with data through 2026-08-02. Over the full post-deploy period all five pages show engaged-reader time-on-page above pre-deploy — including the two that looked flat/worse at 21 days — while same-session conversion stays too thin to read. Weekly distribution box-plots, channel and landing/mid-session splits, deploy inventory, inline SQL.

Mar 18 – Apr 7, 2026 • Treatment drives +267% service-link CTR vs Control

QW contact-information viewport-height distribution by breakpoint, with above-the-fold height recommendations and a landing-page visitor reference.

The 68–77k/wk Direct-homepage flood, fingerprinted: a Mon–Fri, 5–9am-PT batch operation that rests on weekends, works US holidays, and converts exactly zero. Four independent projections put real traffic at 4.4–7.7k/wk — ≈90% of the channel is synthetic; GA's IsLikelyBot tracks the shape (r=0.99) but catches only a third of the mass.

Focused 13-month reconciliation: how much AI and SEO traffic GA4 misses by hiding it inside Direct, SEM, and SEO cohorts. Projected real AI ≈ 466 (GA captures ~23%); projected SEO ≈ 519 (GA captures ~72%).

CMO-oriented analysis of BairesDev marketing channel performance with GA4 × QW survey reconciliation across SEM, SEO-Site, SEO-Blog, AI, Direct, and Paid Social

The full read on /: 446.7k landing sessions in 13 weeks (72% Direct incl. a June bot wave, 13% Zoom-referral non-buyers), 26% of all site leads converting in homepage-landing sessions, ChatGPT visitors converting ~7× SEO's clean CR, and 983 leads → 11 clients ≈ $38.4M since 2025. Five buyer personas from 40 first-call Gong deep-dives, with verbatim quotes. Redesign input research; evidence only.

Competitive design benchmark for the homepage redesign (July 31, 2026), briefed by the 2026-07 homepage analysis. 18 live references (8 direct, 6 adjacent, 4 best-in-class) behind 8 evidence-backed strategic directions: make the hero a router not a statement, three doors for five personas, name the incumbent-vendor failure (unclaimed by all 8 direct competitors), put a real vetting number in the fold or drop "top 1%", design for AI-assistant readers (~7× SEO's clean CR), structurally separate the candidate funnel, claim the vacated nearshore category, and stop letting stated budget shape the funnel. Current-state audit, pattern analysis, and design translation by decision area.

Context-only read of what happens on bairesdev.com after an SEO, AI, or Direct visitor lands on the homepage and explores before entering the QW. 12-month window, no SEM / paid social / campaign landings. Numbers and named examples, no thesis.

Companies switching away from an existing software vendor, detected via QW comments (92 leads; avg BV 12.4 vs 5.4 baseline) and won clients' first-call Gong briefs (21 clients, $56.7M VOC ≈ 4.9% of all VOC — TRAFFIX, Geisinger, LendingTree). The story appears in 0.7% of QW leads but 4.3% of won clients' first calls. AltSource cluster (3 clients from one vendor's shutdown), QW themes, verbatim first-call quotes, channel mix, and pre-submission on-site behavior vs baseline.

Two parallel deep-dives: top 10 VOC clients won this year ($80.3M combined VOC, 4 marketing-channel wins + 2 attribution-special cases at the top) and top 10 high-BV leads that reached SQL but did not convert (closed-lost / cold-disqualified, in-progress deals excluded). Each client has its own sub-page with the full chronological story — sessions → page visits → form → openings → call timeline — and direct links to every Gong recording. Patterns surfaced: marketing-channel wins (MACU, Affinitiv, Avoca, ezCater) all ran the standard QW funnel; high-BV losses cluster on out-of-budget (5/10) and not-a-decision-maker (3/10); BV scoring is uncorrelated with VOC ceiling at this end of the distribution.

12-month read of leads that found BairesDev via LLM tools, detected through three methods (GA4 conversion-session referrer, GA4 pre-conversion session, QW survey self-report). 286 leads, $2.02M BV, 13 SQLs, 2 converted (TroopHR/Reprise, Avoca AI). Channel comparison vs SEM/SEO, on-site journey, full deep-dive on the two converted clients including all GA sessions, page hits, QW answers, and buyer POV from the intro calls.

Qualitative read of 19 Gong calls from the last 6 months where prospects probed how BairesDev engineers use Copilot, Cursor, Claude Code, etc.