Every system we have shipped, open for inspection.

Overview

IntellReach needed an outbound system that writes like a person and knows when to stop. SBD built a multi-tenant platform that ingests and verifies contacts, writes each email with Claude from a knowledge base, runs follow-up sequences, classifies replies, and routes interested prospects to a human.

• 11product areas in one platformDashboard, campaigns, contacts, lists, personas, sequences, knowledge, inboxes, domains, meetings and privacy.
• Every rowisolated by tenantRow-level security scopes every read to the caller's workspace.
• 3scheduled jobs run it allA ten-minute tick, hourly alerts and a daily run. No separate queue service.

Walkthrough

IntellReach AI · AI outbound sales platform

1. 01 One funnel, sent to bookedEvery lead’s path from first email to booked meeting, on the first screen anyone sees.
2. 02 Health you can readDatabase, email writing, sending and mailboxes, each with a state. If something stops, the dashboard says so before a customer does.
3. 03 Written from real knowledgeCase studies and service documents are indexed into searchable passages. Every email the AI drafts is grounded in them.
4. 04 Sequences, written onceSteps, timing and the audience each sequence is for. Teams reuse them across campaigns instead of rewriting.
5. 05 Nothing sends half-readyA readiness check names exactly what is blocking a campaign, with the fix beside each item.
6. 06 Replies become meetingsInterested replies end on a calendar, in the lead’s own time zone.

Workflows

1. Contact ingest WhenA list is uploaded ThenAddresses verified, contacts deduplicated and assigned to a campaign Idle
2. Email generation WhenA contact enters a sequence ThenA personalised email written from the knowledge base and queued Idle
3. Reply handling WhenA reply or bounce arrives ThenClassified, sequence paused, interested prospects routed to a person Idle
4. Readiness checks WhenA campaign is about to start ThenDomains, inboxes and content checked, blockers listed before sending Idle

Stack

• Supabase
• Claude
• OpenAI
• Instantly
• Vercel
• Resend

Every account in the client’s name. Tools with a page link to how we integrate them.

Build log

1. 01
   Product design

   A product reference document and kickoff that fixed what the first phase would and would not do.

2. 02
   Platform

   Next.js and Supabase with row-level security on every tenant row, so each workspace only sees its own data.

3. 03
   Writing and retrieval

   Claude writes each email using retrieval over a knowledge base, with personas and proof points the client controls.

4. 04
   Run it simply

   No separate worker or queue service. The queue is a table and three scheduled jobs do all the work.

Overview

Ruyi Flux runs a US residential rental portfolio of around 120 properties. SBD shipped six fixed-scope automations covering mortgage reconciliation, owner tax summaries, utilities, maintenance scheduling, move-in and move-out capture and welcome letters, and is now building an agent fleet with a control plane and human approval.

• 6automations liveMortgage, tax summary, utilities, maintenance, move-in and out, welcome letter.
• ~120properties coveredAcross the client's US residential portfolio.
• 12agents in buildBehind one control plane with approvals and a kill switch.

Walkthrough

Ruyi Flux · Property operations automations and agent fleet

1. 01 The agent officeEvery department’s agents at their desks, live: working, waiting for approval, idle or stopped.
2. 02 One control planeLog, queue, gateway and policy at the centre. Every action an agent takes passes through it.
3. 03 Agents propose. People approve.Routine items proceed on a timer. Hard stops never do, and every decision is logged with who made it.
4. 04 The Monday scorecardSends, approvals, corrections, failures held and what the week cost, read in one minute.
5. 05 A kill switch for everythingEvery channel and every agent can be stopped in one click, and the policy says which actions always need a person.

Workflows

1. Mortgage scan and reconciliation WhenA mortgage statement arrives ThenRead, matched to the property and reconciled Idle
2. Owner tax summary WhenYear end ThenA summary per owner generated from the records Idle
3. Move-in and move-out capture WhenA tenant moves in or out ThenCondition captured on a phone with photos and filed Idle
4. Agent decision queue WhenAn agent proposes an action ThenApproved automatically within policy, or queued for a person Idle

Stack

• n8n
• QuickBooks
• Twilio
• Zendesk
• ClickUp
• Resend
• Claude

Every account in the client’s name. Tools with a page link to how we integrate them.

Build log

1. 01
   Fixed scope

   Six automations, each scoped and priced on its own, so the client could see exactly what each one did.

2. 02
   Build on n8n

   Deterministic workflows for the repeatable work, with documents read where needed.

3. 03
   Then agents

   A control plane with an event log, a decision queue, a tool gateway and a policy engine, so agents act only within rules.

4. 04
   Human approval

   Routine actions can be approved in bulk, risky ones always go to a person, and a kill switch stops any channel or agent.

Overview

SBEE Cables encodes every product attribute into positional article numbers across 39 categories, so finding the right part meant reading a 24-table catalogue. SBD decoded the scheme into a four-step part finder, a searchable product catalogue and master data, with an enquiry-to-documents flow following.

• 6,000+article numbers decodedEvery finished-goods article mapped to its attributes.
• 39product categoriesGrouped into about fifteen product families.
• 24lookup tables encodedThe catalogue's full numbering dictionary, now in software.

Walkthrough

SBEE Cables · Part finder and product catalogue

1. 01 Start from the requirementStandard, voltage, conductor, insulation, armour, cores and size. Sales enters what the customer asked for.
2. 02 Matches, as you narrowThe finder returns the product families that fit, ranked, before anyone opens a catalogue.
3. 03 6,000 article numbers, readableEvery finished-goods code decoded into its attributes and category.
4. 04 The dictionary, in softwareThe catalogue’s 24 lookup tables and product hierarchy, maintained in one place instead of in someone’s head.

Workflows

1. Part finder WhenA user enters requirements ThenMatching categories and article numbers returned Idle
2. Article decoding WhenAn article number is entered ThenEvery attribute decoded from the lookup tables Idle

Stack

• Vercel

Every account in the client’s name. Tools with a page link to how we integrate them.

Build log

1. 01
   Decode the catalogue

   The full numbering scheme analysed: 39 categories, 24 master lookup tables, and the order each category combines them in.

2. 02
   Clean the data

   Over 6,000 article numbers checked against the scheme, with missing numbers for control cables generated by the same rules.

3. 03
   Build the finder

   A four-step wizard: requirements, category, configuration, result, with free-text search alongside.

4. 04
   Ship and iterate

   Version 1.0.2 live on Vercel, with enquiries and customer modules following.

Overview

A clickable prototype of a white-label property management app: one portal with four modules, a resident surface, property and vendor directories, and integration points shown in the interface. Built against the client's written requirements, with an architecture proposal alongside.

• 4AI modules, one portal
• 1queue of what needs a person
• White-labelfor the client’s customers

Walkthrough

A property management software company · White-label property management portal

1. 01 What needs you, firstThe portal opens on the most urgent item across all four modules, with the reason it is urgent.
2. 02 A maintenance agentEach tenant request is triaged, classified and sent to the right vendor, against a response target.
3. 03 Rent reminders that escalateA cadence per lease. Late payments step up on their own, and a person can pause it.
4. 04 An AI receptionistCalls answered, summarised and routed, or handed to a person when the agent should not decide.
5. 05 Books with a confidence scoreTransactions categorised automatically. Anything the agent is unsure of waits for review.

Modules

• Maintenance agentA request queue where an AI agent triages and dispatches tenant requests to vendors within SLA.
• Rent remindersA reminder schedule with escalation, by lease and by property.
• AI receptionistA call log of calls answered, routed or handed to a person, with summaries.
• Bookkeeping categorisationBank transactions categorised with confidence, low-confidence items queued for review.
• Resident surfaceEach resident sees their own lease, rent and requests.
• DirectoriesProperties and units, residents, and vendors with SLA performance.

Stack

• React
• Vite

Every account in the client’s name. Tools with a page link to how we integrate them.

Overview

Revotec receives requests for quotation by email in every format. SBD built an enquiry platform that reads each RFQ with AI, matches requested parts against the SAP item master, and tracks each enquiry through to order. A review queue for uncertain matches is the next phase.

• EmailRFQs read automaticallyEvery format, without re-typing.
• SAPitem master matchingRequested parts matched to the system of record.
• Ordertracked from enquiryEvery enquiry followed through to order.

Workflows

1. RFQ ingest WhenAn RFQ email arrives ThenRead by AI, items extracted Idle
2. Part matching WhenItems are extracted ThenMatched to the SAP item master Idle
3. Enquiry tracking WhenAn enquiry changes stage ThenStatus updated through to order Idle

Stack

• OpenAI

Every account in the client’s name. Tools with a page link to how we integrate them.

Build log

1. 01
   Read the inbox

   Emailed RFQs ingested and read by an AI step, whatever the format.

2. 02
   Match the parts

   Requested items matched against the item master, including bearing part numbers in their many written forms.

3. 03
   Connect SAP

   The platform works with SAP so enquiries and orders stay in the system of record.

4. 04
   Extend

   Attachment parsing, spreadsheet quotations and a review queue proposed as the next phase.

Overview

OneMagnate Consulting ran quotations, brands, architect firms and contracts in a SmartSuite workspace that had outgrown itself. SBD built a purpose-made operations tool on Next.js and Supabase with nine modules, a pipeline dashboard, and roles enforced on the server, with every login tied to a real person.

• 9modules in one toolDashboard, quotations, brands, firms, architects, contracts, projects, team, settings.
• Serverenforced rolesMembers edit only the quotations assigned to them.
• 0shared or demo loginsEvery account is an invited, named person.

Workflows

1. Quotation pipeline WhenA quotation is created or revised ThenLinked to brand, firm, architect and project, with a revisions timeline Idle
2. Access control WhenAny change is saved ThenRole checked on the server before it is written Idle
3. Dashboard WhenAny record changes ThenPipeline value, win rate and meeting SLA refreshed Idle

Stack

• Supabase
• Vercel
• SmartSuite

Every account in the client’s name. Tools with a page link to how we integrate them.

Build log

1. 01
   Map the workspace

   Every table and relationship in the old workspace mapped to a proper data model.

2. 02
   Build the modules

   Quotations with revisions, brands, architect firms and architects, contracts, projects, team and settings.

3. 03
   Lock it down

   Administrators and members, with roles checked in every server action and sign-up switched off.

4. 04
   Migrate

   Data moved across from the old workspace before switch-over.

Overview

A copper wire manufacturer tracked production on paper, spreadsheets and phone calls. SBD built a multi-user production dashboard with ten modules, role-based access, automated reports and direct printing. Order processing time fell by 45% and data accuracy rose by 60%, with every production stage visible live.

• 45%faster order processingTime from order to dispatch-ready, before and after the dashboard.
• 60%better data accuracyErrors between floor records and the system, before and after.
• 10+modules in daily useProduction, stages, orders, cashback, dispatch and more.

Workflows

1. Production stage entry WhenA batch moves between stages ThenRecorded once against the order, visible to every role that needs it Idle
2. Order movement tracking WhenAn order changes stage ThenDashboard updated, dispatch sees what is ready Idle
3. Report and print WhenA report is due or requested ThenGenerated from live data and printed directly on the floor Idle

Stack

• Bubble
• Cloudflare
• PrintNode
• Render

Every account in the client’s name. Tools with a page link to how we integrate them.

Build log

1. 01
   On the floor

   Several rounds on the factory floor with operators, floor managers and dispatch, mapping where work broke down, not where management assumed it did.

2. 02
   Process and data model

   Business process mapping and a database design covering production, input and output, drawing stages, orders, cashback and dispatch.

3. 03
   Build

   A multi-user web app with role-based access, report generation on Cloudflare Workers and direct printing through PrintNode.

4. 04
   Live and extended

   Ten-plus modules in daily use, with traceability across every production stage.

Overview

A fast-growing watch reseller updated every product by hand on Shopify, eBay and Chrono24. SBD built one product record on SmartSuite that syncs to every channel with per-channel mapping. Update time fell by 80%, launches got twice as fast, and inventory mismatches dropped to zero over the following 90 days.

• 80%less time on product updatesHours spent updating listings across platforms, before and after.
• 0inventory mismatchesReported mismatches in the 90 days after go-live.
• 2xfaster launchesUnder 20 minutes to launch across platforms, against two to three hours before.

Workflows

1. Product create and update WhenA product record changes ThenTranslated and pushed to every channel Idle
2. Sales sync WhenAn order is placed on Shopify ThenStock updated in SmartSuite and on every other channel Idle
3. Per-channel mapping WhenA listing is pushed ThenTitles, categories, images and metadata fitted to the platform's rules Idle

Stack

• SmartSuite
• Make
• Shopify
• eBay
• HubSpot

Every account in the client’s name. Tools with a page link to how we integrate them.

Build log

1. 01
   Map every channel

   Worked through each platform's listing rules with the client team. eBay, Shopify and Chrono24 differ on titles, categories and images.

2. 02
   Clean and unify

   Existing listings reverse-engineered, cleaned and tagged into one schema flexible enough for new channels.

3. 03
   Build the sync

   SmartSuite as the source of truth, Make scenarios translating each change into every channel's format, with image resizing and metadata per platform.

4. 04
   Close the loop

   Shopify sales synced back so stock reflects real orders.

Overview

A building materials supplier ran enquiries, orders and dispatch from Excel, with transporters and warehouse staff updated by phone. SBD built one Retool system from enquiry to dispatch, with invoices raised in Zoho Books and WhatsApp updates to every stakeholder. Manual time fell by 40%.

• 40%less manual timeHours spent on order coordination and data entry, before and after.
• 5+stakeholder types kept informedTransporters, warehouse, accounts, customers and management.
• AutoinvoicingInvoices raised from the order, removing manual accounting errors.

Workflows

1. Enquiry to order WhenAn enquiry is confirmed ThenConverted to an order with nothing re-typed Idle
2. Dispatch notifications WhenAn order changes status ThenWhatsApp message to the transporter, warehouse or customer Idle
3. Invoice from the order WhenAn order is confirmed ThenInvoice generated in Zoho Books Idle

Stack

• Retool
• WhatsApp Business API
• Zoho Books

Every account in the client’s name. Tools with a page link to how we integrate them.

Build log

1. 01
   Watch the lifecycle

   Time at the warehouse and office watching order to dispatch, with transporters, warehouse and accounts interviewed separately.

2. 02
   One record per order

   Enquiries convert to orders without re-entry, and every stage writes to the same record.

3. 03
   Notify the next person

   Each status change messages whoever acts next on WhatsApp: transporter, warehouse or customer.

4. 04
   Close the books

   Confirmed orders raise the invoice in Zoho Books through its API.

Overview

An industrial manufacturer sold only through distributors and had no relationship with the people who used its products. SBD built a QR cashback system in the packaging, multilingual mobile forms, automated payouts and a customer database. Sales rose by 5 to 10% and end-user engagement tripled.

• 5 to 10%sales liftSales on participating products after launch.
• 3xend-user engagementDirect interactions with end users, against near zero before.
• AutopayoutsCashback paid by rule, with no manual processing.

Workflows

1. Scan to claim WhenA customer scans the QR in a pack ThenPurchase and customer recorded, eligibility checked Idle
2. Automated payout WhenA claim is approved ThenCashback paid without manual processing Idle
3. Customer database WhenEvery claim ThenCustomer profile built for segmentation and retention Idle

Stack

• Make
• Cloudflare
• Razorpay
• Google Sheets

Every account in the client’s name. Tools with a page link to how we integrate them.

Build log

1. 01
   Know the customer

   Built around the target users, with forms in several local languages and a very simple interface.

2. 02
   Capture

   QR codes in packaging lead to mobile-first forms that record the purchase and the customer.

3. 03
   Pay out

   Cashback calculated by rule and paid automatically through a payouts API.

4. 04
   Expand

   Rolling out to more product lines, with tiered cashback for premium products.

Overview

A B2B sales team handled leads from forms, email and social messages by hand. SBD built one capture layer with enrichment, AI qualification against the ideal customer profile, CRM sync, follow-up and booking. Manual CRM entry fell by 90%, first replies arrived in under three minutes, and conversion rose 3.4x.

• 90%less manual CRM entryFields entered by hand, before and after.
• <3 minaverage first replyTime from lead arriving to first personalised reply.
• 3.4xconversion rateLead to qualified meeting, with 40% more qualified calls booked.

Workflows

1. Lead capture WhenA lead arrives from any source ThenCaptured, deduplicated and enriched Idle
2. AI qualification WhenA lead is enriched ThenScored against the profile, objections predicted Idle
3. First reply and booking WhenA lead qualifies ThenPersonal email sent, meeting link offered, CRM updated Idle
4. Call summary sync WhenA call ends ThenSummary written to the CRM and posted to Slack Idle

Stack

• OpenAI
• Notion
• Gmail
• Cal.com
• Slack
• n8n

Every account in the client’s name. Tools with a page link to how we integrate them.

Build log

1. 01
   One capture layer

   Forms, email replies and social responses captured through webhooks and inbox polling.

2. 02
   Enrich and qualify

   Each lead enriched from public data and scored by an AI step against the ideal customer profile, with likely objections noted.

3. 03
   Act

   Qualified leads written to the CRM, follow-up tasks weighted by value, a personalised first email sent, and a meeting link offered.

4. 04
   Close the loop

   Call summaries synced back to the CRM and Slack.

Overview

A wholesale jeweller took orders and updated retailers by hand, with a catalogue that changed constantly. SBD built a gated portal for approved retailers with a live catalogue, order tracking and automatic WhatsApp and email updates. Orders processed 30% faster and inbound enquiries halved.

• 30%faster order processingTime from order to confirmation, before and after.
• 50%fewer client enquiriesInbound status questions, because updates now arrive unasked.
• Approvedaccess onlyTrade catalogue visible only to verified retailers.

Workflows

1. Retailer approval WhenA retailer signs up ThenAccess granted only after approval Idle
2. Order to status WhenAn order is placed or moves stage ThenThe retailer is messaged on WhatsApp and email Idle
3. Live catalogue WhenStock changes ThenThe catalogue reflects it Idle

Stack

• Bubble
• WhatsApp Business API

Every account in the client’s name. Tools with a page link to how we integrate them.

Build log

1. 01
   See it run

   Visits to the showroom, warehouse and back office, with interviews across departments.

2. 02
   Gate it

   Approved-user access so only trade accounts see the catalogue and place orders.

3. 03
   Keep it live

   Inventory management behind a catalogue that updates as stock changes.

4. 04
   Tell the retailer

   Order and dispatch updates sent on WhatsApp and email without anyone typing them.

Overview

A five-person photography and video studio ran enquiries, projects and tasks from memory and could not justify enterprise project software. SBD built the system on Google Sheets with a lead form, per-person task sheets, a live admin dashboard and automation kept cheap by design. Admin time fell by 60%.

• 60%less manual adminHours on tracking and updates, before and after.
• 5people on one systemEvery team member and the owner, from one dashboard.
• Near zerosoftware costBuilt on tools the studio already paid for.

Workflows

1. Enquiry to project WhenA lead form is submitted ThenProject created with client details Idle
2. Task to dashboard WhenA team member updates a task ThenAdmin dashboard reflects it live Idle
3. Client updates WhenA deliverable changes status ThenThe client is informed Idle

Stack

• Google Sheets
• Fillout
• Make
• Cloudflare

Every account in the client’s name. Tools with a page link to how we integrate them.

Build log

1. 01
   Design to the budget

   The constraint shaped the build: real capability on tools the team already had.

2. 02
   Capture

   A lead form feeds project data straight into Sheets.

3. 03
   Track

   Each team member gets a task sheet that syncs to the admin dashboard automatically.

4. 04
   Keep it cheap

   Heavier processing moved to Apps Script and Cloudflare Workers to keep automation costs down.

Overview

An operations control system for a garment manufacturer, delivered white-label through a technology firm. Phase one covers the order from receipt to payment: costing and approval, procurement, pre-production tracking, production, dispatch, goods receipts and payments.

• 8stages from order to payment

Modules

• OrdersEvery buyer order as one record from receipt to payment.
• Costing and approvalCost sheets built and approved before production starts.
• ProcurementFabric and trims ordered against the approved costing.
• Pre-production trackersSamples, approvals and readiness tracked against dates.
• Production and dispatchProduction progress and dispatch recorded on the order.
• Goods receipts and paymentsReceipts matched to orders and payments tracked to close.

Stack

• Supabase
• Vercel

Every account in the client’s name. Tools with a page link to how we integrate them.