Aisle - the AI wedding agent (pitch)

Why this, why now

The category is crowded. The experience is not.

Zola, Joy and The Knot own wedding software, but they are all the same thing: dashboards you operate. 90% of couples call digital tools indispensable, yet 59% still describe planning as "overwhelming." That gap is the whole opportunity.

The shift in 2026 is not a better form. It is an agent that operates the software for you. Nobody has shipped that well for weddings, and no serious player has localised for South Africa at all.

The unfair advantage is us.

We already run the hard parts: a Lovable-style in-browser site editor, a template library, cPanel hosting, automated domain provisioning and per-service email sending. A random startup builds those from zero.

For us this product is mostly an agent wrapper plus wedding-shaped UX over infrastructure we own. That is why it is viable for this shop specifically, and defensible.

The market · South Africa

Money already moves here. Software just hasn't caught it.

R27bn

SA wedding industry per year (SABIA), planning segment about 16% CAGR

$1.4bn

SA wedding services 2026, heading to about $2.4bn by 2035

R10-110k

Typical SA planner fee, roughly 10 to 12% of the wedding budget

AI-native full-stack leaders in SA, and no Zola/Knot localisation

Existing SA tools (ForeverAfter, Happly, Plan My Wedding, Confetti) are shallow directories or free basic builders. The whitespace is real, but it comes with the central risk: every SA planning tool today is free to couples. See the money model and the validation plan below.

What Aisle does

One chat thread. Real write-actions behind it.

The agent is not a search box. Every message can change something real: a guest, a table, a page, a budget line, an email in the outbox.

CORE · lead feature

Conversational planner

"Move the Bothas to table 6, drop cousin Piet." "Add an FAQ section to the site." One thread edits guests, seating, website and budget.

CORE · lead feature

Constraint-solved seating

Auto-seats by RSVP and relationships, keeps divorced parents apart, groups the university crew, then adjusts by chat. Nobody does this at seat level.

CORE · differentiator

DJ handoff portal

Couple and guests build the playlist (Spotify), set must-play and do-not-play plus timeline. DJ gets one live link. The end-to-end handoff nobody owns.

reuse existing kit

Chat-to-build website

Describe the site, it builds it on a real domain we host. Approve edits inline. This is our template editor, wedding-skinned.

approve-first

RSVP drafts, sent for you

Agent writes invites and chases non-responders. Every send is preview-and-approve, never blind-fires on your behalf.

the #1 pain

Budget brain

Tracks spend vs plan, warns "you are 20% over on flowers," suggests cuts. Budget is couples' biggest stressor, and the retention hook.

also on the roadmap: contract red-flag scanner · vendor inbox with AI chase · day-of runsheet · cash-fund / registry

The wedge · three things nobody executes

Don't fight on the commodity layer. Win on the gaps.

Registry, basic RSVP and website building are solved and free. Walking into those is suicide. These three are genuinely unsolved across every incumbent we checked.

Constraint-solved seating unsolved

Zola has the RSVP data and still ships manual drag-drop. "AI" seaters only optimise at table level. True seat-adjacency conflict-solving is open.

Live DJ handoff unsolved

Vibo is closest, but most "integrations" are one-way export the DJ re-imports by hand. A live couple to guest to DJ thread does not exist.

Real conversational agent unsolved

No one has shipped one chat that edits guests, seating, site and budget with real actions. David's Bridal "Pearl Planner" is the only near-peer, closed beta, unverified. We are not late.

Who pays

Two-sided, and the couple side funds itself day one.

Side A · demand

💍 Couples buy a package

R2 500-4 500 / once-off wedding package

Sold as a deliverable (their live wedding site plus agent), not a naked subscription, dodging the "won't pay for SaaS" wall. Couples already pay one-off fees elsewhere.

Live wedding website on their own domain
Hosting and domain bundled (recurring markup we own)
The AI agent with an included token allowance
Optional add-ons: printed invites, day-of coordination

Side B · supply

🏛 Vendors buy in

R400-1 200 / month or per-lead

Venues, photographers, DJs pay to sit on the AI's recommended list and receive matched couple leads. This is Zola/Knot's roughly 90%-revenue model, grown once couples exist, never cold-started first.

Featured placement in agent recommendations
Verified badge plus matched, qualified leads
Self-serve listing (free tier), then paid tiers
Analytics: views, enquiries, conversions

AI usage · token model

The agent runs on tokens. Package includes an allowance, top up any time.

Every chat action costs LLM tokens, a real cost we pass through with margin. The package ships with an allowance good for a typical planning cycle. Heavy planners top up in-app (R99 / R199 bundles). This protects margin and turns usage into upside instead of a loss. Route routine edits to a cheap model, escalate only creative work, and cache repeat queries to keep burn low.

allowance used · 62%top up →

Competitive whitespace

Where everyone else stops.

Player	Strength	The gap we take
The Knot	Vendor network, free, seating auto-syncs RSVPs	Form-heavy, no conversational agent, US-only, budget tracker retired
Zola	Registry revenue, drag-drop seating, meal RSVP	Registry-first, seating still manual, not localised
Joy	Best "Smart RSVP", privacy controls	Website plus RSVP only, no planning brain, seating has no RSVP sync
AIWeddings / Nupt.ai	AI venue/vendor matching (thin)	No chat-to-manage, no site builder, no vendor verification
Vibo	Real DJ playlist sync	Point tool, no wedding platform around it
SA tools (ForeverAfter, Happly)	Local, free	Shallow, no AI, no seating, no monetised depth

MVP scope

The narrowest loop that shows the magic.

One vertical slice, one demo couple, sell the "it just did that" moment. Reuse the editor, do not gate launch on a venue scraper.

Chat agent over build-site, add-guest, seat-guest, draft-email, track-budgetthe whole product is a tool-calling agent over our existing APIs

Constraint seating, auto-seat plus chat-adjustthe standout demo moment

Chat-to-build site on a real hosted domainreuse of template-library editor

Approve-first RSVP draft and sendpreview every send, never blind-fire

DJ portal link, Spotify export plus must / do-not-playcheap, and nobody owns it

Budget chatthe retention hook

Venue discovery, Google Places / Yelp live layerv2, legit sources first

Own-site venue crawler, self-updating DBv2, crawl venues' own public sites, not aggregators

Build estimate · time

How long, honestly (buffer already baked in).

Phased for a small team leaning on AI-assisted dev and heavy reuse of the existing editor, hosting and email stack. Weeks below already carry a contingency buffer.

PhaseWhat it coversWeeks

0 · Validation + design

Couple interviews, UX, data model, agent tool spec

1 · Agent core

Tool-calling orchestrator, session memory, guardrails scaffold

4-5

2 · Site builder

Wire the existing editor as agent tools, wedding skin, provisioning

2-3

3 · Seating engine

Constraint solver, relationship model, chat-adjust

2-3

4 · RSVP + email

Guest list, approve-first outbox, chase automation

1.5-2

5 · DJ portal

Spotify integration, must / do-not-play, shareable DJ link

1.5-2

6 · Budget module

Spend tracking, alerts, agent awareness

1-1.5

7 · Fencing + POPIA

Auth scoping, spend caps, injection defense, privacy compliance

1.5-2

8 · Payments + tokens

PayFast, package checkout, token metering + top-ups

1.5-2

9 · Vendor side

Self-serve listings, recommended-list, leads, analytics

2-3

10 · QA + polish + launch

End-to-end testing, security pass, soft launch

2-3

Market-ready MVP

Realistic calendar time for a small team with buffer

~5-6 mo

Faster first light: a lean demo slice (chat agent + constraint seating + hosted site + one RSVP flow, no payments or vendor side) can be standing in roughly 6 to 8 weeks to validate the magic before committing the full build.

Build estimate · what we'd charge

The build cost, in three commitment levels.

Project build figures in ZAR at a blended SA dev rate, reflecting the reuse of infrastructure we already own. Ranges, not promises, until Phase 0 pins scope.

Prove it

Lean demo slice

R120k-180k

~6-8 weeks

Chat agent, constraint seating, hosted site, one RSVP flow. Enough to sell the moment and validate willingness to pay. No payments, no vendor side.

Recommended

Market-ready MVP

R380k-560k

~5-6 months

Everything in the slice plus budget, DJ portal, payments and token metering, security and POPIA pass, basic vendor listings. A product you can charge real couples for.

Full platform

Two-sided at depth

R750k-1.1M

~8-10 months

Adds venue discovery + self-updating DB, vendor lead marketplace, contract scanner, vendor inbox, analytics, registry / cash funds and scale hardening.

figures assume reuse of existing editor / hosting / email. building those from scratch would roughly double phase 2 and add 6 to 8 weeks.

Build estimate · running costs

What it costs to keep on the air.

Cost line	Model	Rough cost
Hosting (own infra)	Already paid, marginal per wedding	~R0 marginal
LLM tokens	Per active couple per cycle, pass-through with margin	R40-120 / couple
Google Places / Yelp	Usage-based API calls, budget-capped	~$17 / 1k calls
Email sending	Own infra, marginal	~R0 marginal
PayFast fees	Per transaction on the package sale	~3.5% + R2
Domain registration	Per wedding, charged on to the couple	~R100-150 / yr
Spotify API	Free tier	R0
Maintenance	Ongoing dev / support retainer	retainer

Unit economics look healthy: a R2 500-4 500 package against roughly R300-600 of direct cost per couple (tokens, domain, payment fee, support) leaves a strong margin, before any vendor-side revenue on top.

Keeping the agent fenced

An agent with real write-actions has to be caged properly.

The moment a chatbot can send email, spend tokens and edit data, it becomes an attack surface. These controls are in scope from Phase 1, not bolted on later.

AUTHORIZATION

Scope every tool to one wedding

The agent can only touch the signed-in couple's own data. Every tool call is authorised server-side against their account, closing IDOR / data-leak paths to other couples.

COST ABUSE

Hard spend and token caps

Per-account token budgets and daily rate limits stop a malicious or runaway session from running up a bill. Overages pause the agent, they do not silently drain money.

PROMPT INJECTION

Treat all content as untrusted

Guest names, vendor blurbs and venue pages can carry injected instructions. Untrusted text can never trigger a tool call or override the system prompt. Tools run behind an allow-list, never free-form code or SQL.

HUMAN IN THE LOOP

Approve-first on anything outbound

Sending email, publishing the site or spending money always surfaces a preview the couple confirms. The agent drafts and proposes, the human commits.

GENERATED OUTPUT

Sanitise what it builds

AI-built pages are escaped and CSP-guarded (no inline handlers), so a crafted prompt cannot inject script into a couple's live site or the dashboard.

CRAWLER SAFETY

SSRF-guarded venue fetches

The venue crawler is fenced to public web targets only, blocking internal-network and metadata-endpoint fetches, and it respects each site's robots and rate limits.

PRIVACY · POPIA

Guest data is regulated data

SA POPIA applies to guest lists and dietary / contact info: consent, data minimisation, retention limits and delete-on-request are built in, not optional.

ACCOUNTABILITY

Every action is logged

A full audit trail of what the agent did, when and on whose behalf, powers trust, support and one-click rollback of any change.

Risks · the honest bits

What could sink this, and where we're being careful.

Will ZA couples pay?

Every local tool is free today. This is the central risk, bigger than any tech gap. Selling a deliverable (site plus domain plus agent), not a subscription, is the hedge. Validate with real couples before writing code.

Vendor lead-gen needs scale

The high-revenue side cannot cold-start. Grow it only once couple traffic exists. Never build for it first.

Scraping aggregators is a trap

Zola/Knot ToS ban mass harvesting and actively bot-block. We use Google Places/Yelp plus crawl venues' own public sites, never the aggregators.

Short LTV

Couples churn after the wedding. Offset with vendor recurring revenue and a path into engagements and events later.

Cheapest possible validation: 5 couples, 5 questions

How did you find and choose your venue, and what was the worst part?tests the venue-matching value and the real pain

What are you paying for right now that you would happily stop doing by hand?tests willingness to pay and which pillar

Would you trust an AI to draft and send invites if you approved each one first?tests the approve-first agent model

What would a done-for-you wedding website plus planner be worth to you as a one-off?tests the price point directly, the make-or-break number

Show the seating and DJ demo: does this feel like magic or a gimmick?tests whether the wedge actually lands emotionally

Open questions · still to decide

Calls we should make before Phase 0 closes.

Name and brand. Is "Aisle" the one, or do we want something more distinctly ZA? (alternatives on the table: Vow, Betroth, Amara.)

Do we build the lean demo slice first to de-risk the pay question, or commit straight to the market-ready MVP?

Exact package price point, the single most important number, to be set by the 5-couple validation, not guessed.

Do we also white-label the planner for wedding planners and venues as a B2B tool, a warmer, more recurring customer than couples?

Full-auto vs approve-first is settled (approve-first). Do guests get a light login, or purely magic-link RSVP with no account?

POPIA: do we register as a responsible party and appoint an information officer up front, or handle at launch?

Next steps

Talk to couples before touching code.

This week

Run the 5-question script with 5 real SA couples. One afternoon of coffees. If the price point lands, everything else is execution.

If validated

Scope the MVP against existing template-library code, a reuse map of what the editor already gives us vs what is net-new agent tooling.

Then

Build the vertical slice, chat agent plus constraint seating plus hosted site plus DJ portal. One demo couple, end to end.

In parallel

Hand-recruit 10 vendors in one city onto free listings, seeding the recommended-list side without a scraper.