Envoy CRM: What I Learnt Getting 20 SMEs to Actually Use a CRM
Contents
I've spent the last year trying to get SMEs in Singapore to use a CRM. If you've ever tried this, you know the punchline: they won't. Not HubSpot, not Salesforce, not Zoho. The databases stay empty, the pipelines stay untouched, and everyone goes back to WhatsApp and memory.
Envoy is our attempt at building something they'll actually use. It's an AI-native CRM, WhatsApp-first, built specifically for how Singapore SMEs operate. We now have 20+ paying customers, ranging from wellness clinics to fashion brands to energy companies. This post is about the product decisions that got us there, the ones that worked, the ones I killed, and what the data told me along the way.
#Why I bet everything on WhatsApp
Early on I had to make a call: build omnichannel from day one (WhatsApp, email, Instagram, Telegram) or go deep on WhatsApp and add the rest later.
Every competitor does omnichannel. Wati, Respond.io, they all treat WhatsApp as one channel among many. The pitch sounds great on a slide deck. The reality is that spreading across five channels means you're mediocre at all of them, and for SMEs in Singapore, WhatsApp is where 90% of their customer conversations happen anyway.
So I chose WhatsApp-first. Not WhatsApp-only (we added email later), but WhatsApp as the centre of the entire product. Every feature was designed around WhatsApp conversations first, then adapted for other channels.
The data validated this pretty quickly. When we onboarded our first ten customers, every single one of them used WhatsApp as their primary channel. The email integration was nice to have. WhatsApp was the reason they signed up.
The non-obvious consequence of this decision: we could build a feature called "CRM from History." When a customer connects their WhatsApp Business number, Envoy syncs their full message history and auto-populates contacts, companies, and opportunities from past conversations. No manual data entry. You plug in WhatsApp and your CRM already has data in it.
That's the real insight. The biggest failure mode of CRMs in SMEs isn't missing features. It's empty databases. If the system requires manual entry, it stays empty. So I made the AI do the entry.
#Dogfooding, and why two weeks wasn't enough
Before we sold Envoy to anyone, we ran Voltade's entire sales pipeline on it. Phase 1 of our Q1 plan was literally "strongly satisfy one customer," and that customer was us.
I set specific criteria. Not "use it and see how it feels." Specific, measurable things:
- 100% of sales leads created in Envoy, not Telegram, not a spreadsheet
- Zero deals tracked outside the system
- 80% of follow-ups created automatically by the AI
- The sales team uses Envoy daily for two straight weeks
If any of these failed, we'd fix the product before selling it.
What dogfooding actually surfaced was uncomfortable. The AI was auto-tagging contacts with labels that looked right but weren't. It would tag someone as "Interested" based on a casual enquiry that was obviously going nowhere if you read the conversation properly. The follow-up reminders fired too aggressively, sometimes within minutes. And the mobile experience was slow enough that our sales team would just open WhatsApp directly instead of going through Envoy, which defeats the entire purpose.
We fixed all of it. But in hindsight, two weeks of internal use wasn't enough. Some bugs only surface with volume. A real customer with hundreds of daily WhatsApp messages generates patterns that a ten-person team using it for sales never will. If I did it again, I'd dogfood for a month.
#The Speed Test
This was probably our scrappiest experiment.
We'd message SMEs pretending to be potential customers, from five different WhatsApp numbers and email addresses. We tracked how fast they replied, during office hours and after hours, across different days of the week. Then we compiled the results into a report by sector, starting with wellness and education.
The pitch: "Find out how your reply speed compares to your competitors." Free report in exchange for an email signup.
It worked because it surfaced a problem most SMEs don't know they have. They think they reply quickly. The data usually said otherwise. And the natural next question is "how do I fix this?" which is exactly what Envoy does.
Not scalable. We tested it by hand from actual phones. But it generated real leads and, more importantly, gave us sector-specific data about SME behaviour that informed the product. We learnt that wellness businesses (clinics, spas) tend to reply fast during hours but go completely dark after 6pm. Education businesses are the opposite, slow during the day because teachers are teaching, faster in the evenings. That shaped how we configured default AI response windows for different sectors.
#Grants as a GTM channel
This is the part that doesn't show up in Silicon Valley playbooks.
Singapore's government subsidises software adoption for SMEs through the Productivity Solutions Grant (PSG). Voltade is one of the approved vendors by IMDA. An SME can get Envoy at the $4.6K tier with significant government subsidy.
I built the entire customer journey around the grant mechanics. The flow isn't "sign up, enter credit card." It's: qualify for grant, apply on the Business Grants Portal, connect WhatsApp, go live. We automated the tedious parts: the AI pre-fills application fields, generates claim documents, sends invoice reminders, and tracks the grant timeline.
We also pursued the GenAIxDL grant for larger $88K pilots (30% subsidised) and signed a reseller deal with Singtel at 20% commission. Singtel brings enterprise relationships, we bring the product.
The grant path sounds like ops overhead. It is. But it's also a moat. Competitors who aren't approved vendors can't offer subsidised pricing. And the grant onboarding is complex enough that customers need hand-holding, which creates a relationship from day one that's hard to break.
#Driving to Ex-Style's office at 4pm
One of our customers, Ex-Style (fashion brand), had their WhatsApp Business number banned by Meta. Overnight. No warning. Their entire customer communication channel, gone.
We drove to their office within two hours. Helped them set up a new number, migrated conversations, reconnected the chatbot, got them back online the same day.
I keep coming back to this as a competitive advantage. Being in Singapore, being able to show up at a customer's office on short notice, that's something no SaaS company serving SMEs from San Francisco can replicate. When I wrote about what differentiates us from the big CRM players, I listed "trust, proximity, and accountability." The Ex-Style incident is what I meant.
They stayed. And became one of our strongest referral sources.
#The tickets graveyard
There's a page in our Notion workspace called "Tickets Graveyard." It's where features go to die.
MCP server authentication. File upload toggles for the web widget. Broadcast with CRM filters. Super admin summary on Telegram. Conversation fetching optimisations. All killed.
Each one was a reasonable request. Some came from customers. Some from the engineering team. Some from me. But every feature you build is a feature you maintain, and at our stage (ten people, 20+ customers with different needs), focus is the only advantage we have over larger competitors.
The hardest one to kill was broadcast with CRM filters. Three customers asked for it. The engineering team had partially scoped it. But when I looked at the data, those three customers were using broadcast for fewer than 50 messages a month each. The feature would have taken two weeks to build and test. The ROI just wasn't there, not yet.
Saying no to a customer who's paying you is genuinely uncomfortable. But it's the most important product decision you make.
#The architecture call
By March 2026, Envoy's Rails backend was slowing us down. More time on maintenance than new features. Client-specific edge cases bleeding into the shared codebase. Engineers spending hours debugging cross-domain dependencies.
I wrote a decision memo with three options: full TypeScript rewrite, hybrid modernisation (keep Rails core, extract high-pain domains into TS), or stabilise Rails and defer.
The team had strong opinions. Some wanted the clean slate of a full rewrite. I pushed for the hybrid approach, and the reason was data, not preference. I looked at the last three months of delivery metrics: lead time, PR throughput, defects. The pain was concentrated in two or three domains, not spread evenly. A full rewrite would address everything but take months. Extracting the pain points could show results in weeks.
We chose hybrid with a hard 90-day criterion: if the first extraction pilot doesn't improve median lead time by 20-30%, we stop and revert. No sunk cost bias. Measure and decide.
That framing, scoring options against criteria instead of debating preferences, is something I keep using. It's the same approach I take for any product decision where the team is split. Define what success looks like, define how you'll measure it, set a kill condition, then go.
#How I run the team
Monday 10am I lead planning. Break epics into features, assign ownership. If a feature is too big to finish in a week, it's too big. Engineers spend the rest of Monday exploring the code and concretising what each feature actually requires: endpoint changes, model changes, frontend components.
Tuesday through Thursday: build. Push straight to staging, no PRs. Customer success and I test staging continuously. If something's broken, new ticket with a screenshot and a one-line description.
Thursday 9pm: deploy to production. Thirty minutes of end-to-end checks from the full team.
Friday is "Chaos Day." Customer success sends updates to every customer before lunch and explicitly asks for feedback. Engineering ships bug fixes and tech debt. I plan next week.
The thing that makes this work isn't the cadence itself (lots of teams do weekly sprints). It's the Friday feedback loop. Forcing customer-facing updates every single week means we hear about problems in days, not months. And the team knows that whatever they ship Thursday night is going directly to customers who will be asked about it Friday morning. That creates a quality standard that no code review process can match.
#The numbers
$2.5M total revenue across Voltade, bootstrapped, ten-person team. Envoy specifically: 20+ signed customers across $4.6K, $17K, and custom tiers. Goldlion, Mary Chia, Union Energy, Overmugged, Happy Fish, Ex-Style, Georges, Cake Inspiration. Every one of them running real integrations: WESS, Plato, Shopify, Xero, WooCommerce.
These aren't trial signups. They're businesses running daily operations on Envoy. That distinction matters.