Morvion · Field notes
Field notes from live CRM, AI and dashboard work. One brief a week, most Mondays.
Browse the archive
Latest issue
Open your own checkout and count the steps. Then count the fields. This is a teardown of the last screen of the sale: what it must collect, what it usually collects, and the field-by-field rebuild that stops a documented majority of buyers from walking at the counter.
Issues 1 to 12 of 29; the latest is above.
Glossary
Supplier invoices and wage floors move on their own schedule. The rate card usually moves once a year, after the margin has already gone. A field note on the quarterly pricing review: three inputs, thirty minutes, and one recorded decision per line, even when the decision is to hold.
Most operations have bought the security measures. Far fewer have ever watched what happens when one of the things they run on simply stops answering. The gap between owning a measure and staying open is not closed by buying more. It is closed by rehearsal, and the rehearsal fits inside one quiet hour.
The week that breaks a founder-run operation is never the average week. It is the one where the quarterly bill, the repair and the slow fortnight land together. Days of cover says how many of those weeks the business could absorb, and most operators have never worked it out. A working session on the measure, the baselines behind it, and the moves that add days without touching a price.
A growing share of buyers ask a chatbot one question and read one assembled answer instead of browsing ten links. That answer is built from your data exhaust, not your design. What the machine reads, why inconsistent facts hurt more than missing ones, and the inventory that keeps the answer about you accurate.
Most founder-run businesses have never run a full week without the founder, and the survey numbers say the test rarely happens by choice. The absence audit lists the processes that only work when the operator is in the room, then moves them one at a time into systems that run on their own.
Since June 2025, EU law sets an accessibility floor under online shops. Much of the web fails it at a documented scale, and most of what fails is ordinary craft. Where the legal line sits, how far a typical storefront falls below it, and how to build the floor into everyday releases instead of buying an audit.
Every reservation is inventory sold on a promise, and a documented share of those promises quietly break. No-shows as a ledger entry: what the empty chair costs, evidence that the rate falls when venues act, and the policy settings that price the risk without punishing regulars.
Most founder-run operators treat online reviews as a marketing chore, chased when a bad one lands and forgotten otherwise. The research says the rating moves real revenue for independents, and that responding measurably lifts both rating and review volume. How to run reviews as an operating signal instead of a campaign.
Most premium storefronts are designed for the buyer who already made up their mind. Research on luxury shoppers finds four distinct groups arriving with different intent, and the phase where the website matters most is the one operators consistently under-build.
Decades of research show what makes a dashboard usable, yet the typical business dashboard is wallpaper. Goal-anchored design, F-pattern scanning and cognitive-load discipline lift usability by as much as 70%. The version an operator can apply on Monday.
McKinsey's 2025 State of AI report puts 88% of companies using AI and only 6% capturing serious value. The gap is the pilot loop. A two-week sprint structure that ends in a build verdict backed by evidence.
The word agent has been stretched until it means nothing. Strip the marketing away and an AI operator is a narrow, supervised worker doing unglamorous jobs inside the stack you already run: reconciling, following up, routing. What that looks like on an operator's screen, and the engineering that makes it safe to leave running.
Swiss SME AI adoption rose from 22% to 34% in a single year. The tools are landing fast; the skill to connect them into one working system is not keeping pace. What that gap looks like on the ground, and what changes when integration becomes the brief.
Independent restaurateurs and boutique hoteliers are spending more on software than ever, while ranking integration as their top operational pain. The fix is connecting what is already in the building, without ripping anything out.
Most hospitality CRMs are enterprise hotel software with a marketing layer welded on. What small venues actually need is a memory layer: who the regulars are, what they like, what to do when they walk back in. The minimum viable hospitality CRM and where to grow it from.
Every marketplace transaction depends on a guess: is the supplier real, will they perform, is the buyer worth the time. Trust is the invisible layer that makes the guess feel like a decision. Six signals that actually move marketplace conversion in 2026.
The default 2026 instinct is to break every AI workflow into specialised agents. Most of the time that is the wrong move. Where the split actually pays off, and where it just adds latency, cost, and failure surface.
Retrieval-augmented generation is the most common AI architecture in 2026 and the most-misimplemented. The model gets blamed; the retrieval layer is the bottleneck. Where it actually breaks and how to design it instead.
In services marketplaces, the request-to-quote loop is the actual product. A buyer describes what they need; suppliers reply with what they will do for what. The loop is rarely treated as a designed object. Where it actually wins, with Dreilokale as the reference case.
The reservation system is the load-bearing layer of every venue. When it fails (a double booking, a no-show without a deposit, a Friday night that nobody told the kitchen about), the operator absorbs the cost. Where venue reservation systems actually break and what to build instead.
Demos pass. Production fails. The bridge is an eval harness, measurable, repeatable, written before the agent. Notes from the engagements where we built one, and the ones where we wish we had.
For ten years the goal was to see your business in one place. The next ten will be about systems that act on what they see, and the operator becomes the editor, not the analyst.
Buying a SaaS AI agent is cheaper at the start and more expensive after month six. A short field guide on when to build, when to buy, and where the line is moving as model costs collapse.
Most AI MVPs are too big to ship and too small to learn from. The fix is structural: one decision, one surface, one fact. A guide to two-week proofs that actually move the founder forward.
Email is a sunk cost: every reply breeds two more. A client portal turns the same hour into compound retention. What to put in the portal, what to leave out, and the smallest version that pays back.
Reservations, events, communication, and customer journeys live in one connected layer, or they fight each other. Notes from inside two engagements.
WebGL heroes win awards and lose users. The studios shipping conversion-grade immersive interfaces are the ones who treat motion as a meaning system, not a finish. A field test of what works and what's just expensive theater.
From request flows to operator dashboards: what we learned, and why most marketplaces fail at the shape stage, long before any code is written.
Three agents, a sharp dashboard, and a workflow that actually closes deals, built in eight weeks. A walkthrough of the engagement shape we've repeated three times.
The Signal Brief
Field notes from live CRM, AI and dashboard work. One brief a week, most Mondays.