From Idea to MVP: Designing an AI-Powered SaaS Product from First Principles

A solo founder came to us with deep knowledge of a difficult corridor: EU exports into inland Southern Africa, where a single shipment crosses multiple jurisdictions, handoffs, and failure modes that standard visibility tools barely describe. Their target buyers were not global operators with dedicated regional desks. They were organisations moving cargo on this lane a few times a year, competent on general freight but without in-house specialists. Planning still ran through forwarder relationships, email, and spreadsheet rows that went stale once cargo left the port.

There was no commercial product that turned that corridor knowledge into something a buyer could use before committing to a move. Visibility tools track boxes. Forwarders sell carriage. What was missing was a decision layer: structured evidence a logistics or procurement lead could review with their team, with explicit limits on what the software could promise.

We combined advisory work with delivery across roughly eighteen weeks: discovery and positioning first, then a focused MVP build accelerated by AI coding tools. The output was not a freight forwarder in disguise. It was advisory intelligence with a deterministic core, a visible confidence score, and a human expert path when the data or the risk did not justify blind trust.

This write-up is anonymised. We have removed product name, scoring weights, and commercial terms. The approach and trade-offs reflect the engagement.

The corridor in numbers

Before discovery workshops, we pulled baseline statistics from public institutions. They explain why EU shippers without a dedicated regional desk feel exposed on this lane, even when their forwarder is competent.

• EU-SADC trade volume: Total trade between the EU and SADC EPA countries reached €52 billion in 2025, with overall goods trade up 26% since the EPA entered force in 2016 (EU exports up 16%, SADC exports up 38%) (European Commission).
• North-South Corridor transit time: The SADC Time Release Study, published in February 2025 with EU support under the Trade Facilitation Programme, measured average clearance and transit for northbound cargo from Durban to Kasumbalesa at 15 days, 21 hours, and 5 minutes, with Beitbridge and Chirundu identified as principal bottlenecks (SADC). This is one major regional corridor benchmark; EU-origin cargo may also enter via other gateways.
• Landlocked cost penalty: UN-OHRLLS estimates that landlocked developing countries face unit logistics costs 63% higher on exports and 75% higher on imports than transit developing countries (UN-OHRLLS, 2025). World Bank research cited by UNCTAD puts the landlocked freight premium at roughly 50%, with each additional transit day costing about 0.8% of cargo value; Zambia and Malawi fall into this category (UNCTAD).
• Road dominance and border pressure: In Zimbabwe's corridor network, road carries an estimated 80% of freight by volume (WFP Logistics Cluster, cited in UN-OHRLLS Zimbabwe transport analysis, 2025). Beitbridge is described in that same analysis as the busiest road border post in Southern Africa.
• Regional trade still below potential: In recent years, intra-SADC trade has been around 23% of the bloc's total trade, leaving room for corridor efficiency gains if border and transit friction comes down (SADC).

These are not abstract macro figures. For an SME importing project cargo, mining spares, or agricultural inputs a few times a year, multi-day border variance and a landlocked freight premium translate directly into committed spend before anyone can see where delay will bite.

The challenge

The pain showed up before the container stalled at a border, not only after.

An importer in Rotterdam or Hamburg planning a move into inland Southern Africa must weigh cost, dwell time, seasonal disruption, and border behaviour across a chain that can run roughly sixteen days on the North-South Corridor alone (per the SADC Time Release Study), before inland legs even begin. Cyclone season, election calendars, and month-end surges at major border posts are predictable in aggregate; they are just scattered across corridor bulletins, institutional reports, forwarder habit, and informal operator networks.

When we spoke with the founder’s target users, a pattern repeated. They were not asking for another map with a pulsing dot. They wanted to know: given my shipment, which corridor is the sensible default, what am I trading off, and where should I not trust the answer without a second opinion?

The competitive set was awkward. Enterprise visibility suites assume you already booked the move. TMS platforms belong to the forwarder. Generic AI chatbots can sound fluent about Africa and still invent port facts. The forwarder’s recommendation might be fine, but it is one opinion with a commercial motive. The startup’s wedge was independence: no carriage to sell, only corridor evidence, expert validation where needed, and outputs the user could interrogate.

A hard legal line followed from that wedge. No bills of lading. No carrier liability. No freight forwarding. Advisory outputs only, with disclaimers users see at the point of decision. That boundary was product design, not fine print buried in terms.

Discovery: the corridor before the code

Discovery took the first half of the engagement. It felt slow to a solo founder racing to launch. It saved the build.

We interviewed the beachhead user the founder had already sketched: logistics or procurement staff at EU SMEs (roughly five to two hundred people) shipping into inland Southern Africa a few times a year. Low-to-medium corridor expertise. High fear of an expensive, avoidable mistake on port choice.

Three jobs emerged, sequenced deliberately:

1. Decide (MVP): Compare port and corridor options for a specific shipment. Cheapest, fastest, safest, and what risks are real rather than generic.
2. Prepare (later): Documentation and transporter briefing so border time is not wasted.
3. Escalate (revenue engine): When risk or uncertainty is too high, reach a human who knows the corridor.

v1 lived entirely in job one. Everything else was backlog with written gate conditions.

We mapped why the Southern Corridor made a viable beachhead: less crowded by tech than some northern lanes, high advisory value because of competing ports and transit complexity, and enough public corridor data (regional maritime and corridor institutions publish performance and transit indicators) to stand up a first version without carrier API integrations. The founder still had to curate and validate that data. Nobody gets a trustworthy confidence score from copy-pasting a PDF once.

On AI we were strict early. Scoring and ranking routes had to be deterministic TypeScript, rule tables the founder specified before any codegen touched them. Models were scoped to plain-language rationale and risk explanation where templated text failed. Letting a model choose the port would have produced plausible wrong answers on a lane where plausibility is dangerous.

The strategy: intelligence layer, not forwarding

Positioning settled on a line the founder could defend in a sales call: help importers execute Africa-bound moves with independent evidence, not replace their forwarder.

We rejected several alternatives. A visibility aggregator competed with well-funded incumbents. A document-checklist product was useful but commoditised. A chat-first “AI freight assistant” failed the trust bar. A structured query → scored comparison → confidence label → optional expert flow matched the buyer, the legal boundary, and the data available at launch.

Commercial shape was advisory first, subscription later: paid expert sessions once trust existed, SaaS tier after repeat usage proved itself. We do not publish pricing here; the strategic point is sequencing revenue with validation, not maximising self-serve signups on day one.

Governance matched a customer-facing decision product, set out in the same spirit as our GDPR and AI guide: show provenance, avoid guarantees, log what was shown to whom.

What we built

The MVP was three connected surfaces, not a platform for everything.

1. Input: a short form (origin port area, inland destination, cargo type, container context, weight band, target date, priority, prior corridor experience). Designed to finish in a few minutes without jargon.

2. Decision engine: deterministic comparison of seven pre-defined Southern Corridor route variants across four dimensions: cost, speed, risk, and predictability. Weights shifted with the user’s stated priority (cost-first, speed-first, risk-first). No model participated in the score.

3. Output: a recommendation bar, side-by-side route comparison, plain-language port choice rationale, risk tiles (seasonal flags, known border dwell bands, transit complexity), a 0-100 confidence score built from explicit components (data freshness, expert validation status, cargo coverage, destination specificity, seasonal adjustment), and a short explanation of what limited the score. Plus next steps and an expert booking path when risk or uncertainty crossed thresholds (medium risk, low confidence, certain transit countries, first-time shippers).

AI appeared only in the narrative layer: one-sentence rationale and readable risk copy, grounded in the same structured fields the engine used. If the structured data did not support a claim, the template did not invent one.

Out of scope for v1: bills of lading, carrier booking, compliance certification, open-ended corridor expansion, autonomous email to forwarders, predictive ML on delays.

A decision before booking (anonymised)

A procurement manager at a European manufacturer needs to move mining equipment to Lusaka. Ocean freight from Antwerp is settled; the open question is inland entry. Their forwarder prefers one east-coast gateway. A colleague once used another and lost a week at border.

They complete the form: heavy cargo, cost priority but not at any risk, target arrival before the rainy season, first time on this corridor.

The engine compares seven scored variants in seconds. Two routes cluster on cost; one is faster but carries a seasonal elevation on the sea leg. The recommendation bar names a default with a clear runner-up. Risk tiles call out multi-day dwell at a major Zimbabwe transit point and show that the data behind dwell estimates was last validated weeks ago, not years. Confidence lands in the high seventies; limiting factors cite data age and specialised cargo coverage, shown explicitly rather than hidden behind a green badge.

Plain-language copy explains the trade-off without hype. Because risk is elevated and the shipper is new to the lane, the expert path is prominent, not buried in the footer. The user can still take the comparison to their forwarder as an independent brief. Nothing in the product implies the startup will book carriage.

That single journey drove architecture: structured corridor tables in the database, versioned scoring spec in code, LLM calls isolated behind a narrow interface, shareable output links for internal approval, PDF export for tender packs, payments and booking hooks for advisory tiers.

Architecture and AI

Stack choices were deliberately boring: modern web app, managed database with row-level security from day one, hosted deployment, payment and booking integrations for advisory SKUs, email for outputs and expert briefings, analytics on the funnel from form start to paid session.

The scoring function lived in an API route with no external dependencies for the decision itself. Corridor rows were curated manually from institutional sources and expert sign-off, updated on a published cadence (monthly for some fields, quarterly for border dwell bands). Expert validation was not cosmetic: it materially affected whether the product was honest enough to sell.

AI codegen accelerated scaffolding (schema, form UI, webhook handlers, email templates). It did not replace the founder writing the scoring spec or recruiting corridor experts. The lesson from the build was consistent: when AI compresses engineering, the bottleneck moves to data and people.

We evaluated outputs with scenario tests across all seven routes: cyclone-season flags, oversized cargo, remote destinations, first-time users. Target was sub-four-minute end-to-end on a typical query, including narrative generation.

Delivery and outcomes

Calendar-wise, discovery and strategy consumed the early weeks; the core MVP build targeted roughly four to six weeks of focused execution once the scoring spec and first corridor rows existed. AI tools shaved engineering time; they did not shrink expert outreach or manual QA on whether recommendations matched corridor sense.

Delivered: working form-to-output flow on all in-scope routes, confidence scoring with visible limiting factors, advisory booking with paid checkout, expert briefing automation, tender-pack PDF, disclaimers on every output, instrumentation on the conversion funnel.

Business: moved from founder intuition to a demonstrable product narrative: independent corridor decision support, legally distinct from forwarding. Revenue targets were advisory-led in month one; subscription tiers waited on repeat behaviour. We do not cite actual revenue; those were launch goals, not audited results.

Users: beta queries from the founder’s network and early design partners. Feedback centred on trust: showing why a route won and what data was stale mattered more than visual polish.

Strategic: the team stopped treating AI as the product. The product was curated corridor knowledge encoded as rules, with AI as translator and experts as backstop.

Lessons

The founder’s first instinct was to ship a broader platform. Discovery narrowed to one job: choose the corridor correctly.

Experts were a pre-launch dependency, not a Week 3 afterthought. Without validated rows, confidence scores collapsed and advisory upsell made no sense.

Forwarder distrust was a feature, not a bug. Users trusted the product because it did not sell the next container slot.

UI that exposed limiting factors (“data 61 days old”, “mining equipment coverage partial”) outperformed slick summaries. Anti-black-box design was conversion design.

Solo-founder bandwidth plus AI codegen tempts you to overbuild UI. The scope triage rule held: never cut scoring accuracy or the expert path; cut polish and secondary integrations first.

Why this matters elsewhere

The Southern Corridor is specific. The pattern is not: high-stakes route decisions, fragmented public data, expert tacit knowledge, and incumbents who sell the outcome they recommend.

The same sequence applies wherever buyers need an independent decision layer above systems of record: trade finance route choices, supplier selection with regulatory variance, infrastructure investment comparisons. Map the job, encode the decision table in code, use models only for language, show confidence honestly, wire humans where data is thin. Our mortgage advisory case study shows the same discipline in a regulated workflow. For leadership model questions, see fractional CTO versus technology advisor.

What's next

Post-handover priorities: capture structured outcome feedback (did the recommended corridor perform as expected?), expand curated routes only when demand and validation justified it, and gate a subscription tier on repeat queries from advisory customers. Northern corridor expansion stayed explicitly sequenced behind Southern Corridor proof.

Frequently asked questions

What kind of organisation was this?

A confidential solo-founder startup building advisory corridor intelligence for EU importers into inland Southern Africa. Not a freight forwarder. Strategy, architecture, and MVP delivery in one engagement.

How long did it take?

About eighteen weeks total, with discovery and positioning front-loaded and a four-to-six-week focused build once the scoring spec and initial corridor dataset existed.

Where was AI used?

Plain-language rationale and risk copy only. Route comparison and scoring were deterministic. No autonomous booking or carrier-facing actions.

Could this work outside logistics?

Yes, wherever buyers face opaque route or vendor choices, public data exists but is hard to use, and trust requires showing your working. Our AI adoption guide covers the wider “where AI belongs” question for smaller teams.

---

Vyrion Tech provides independent technology advisory, product strategy, AI strategy, and hands-on delivery for founders and growing businesses in the UK, South Africa, and beyond. If you are moving from idea to MVP and want decisions grounded in evidence rather than hype, book a free consultation.