I’m writing this because I’ve read enough ‘from zero to $10k MRR in 30 days’ posts to know what they leave out. This is the version I wish I’d found before I started.

Month 1: The Build Phase Nobody Warns You About

We launched nothing publicly in month 1. I spent it building, talking to 18 potential customers, and testing assumptions.

Revenue: $0. Conversations: 18. Paid commits: 0.

The 18 conversations were the real product of month 1. I came in with a hypothesis about what outreach automation should look like. I left with a completely different problem statement.

My original idea: a tool that personalizes the first line of outreach emails automatically. What the conversations told me: founders didn’t need a better first line. They needed to know WHO to contact first. The personalization problem was real, but it was downstream of a messier problem: researching and prioritizing the right targets.

I almost pivoted. Instead, I kept the personalization angle but added a research layer on top. That decision cost 3 extra weeks of build time. It probably saved the company.

Month 2: The First Money Is Wrong

We started charging in month 2. Three customers by the end of it.

Revenue: ~$900 total. Churn: 0 (too early). NPS conversations: 3.

Here’s what nobody tells you about your first 3 customers: they’re all wrong. Not in a bad way. They’re the customers who were early enough, patient enough, and interested enough to pay before it worked properly.

My first 3 customers: A solo founder who replied to my cold email within 4 hours. He used the tool for 8 months before we fixed a bug that annoyed him and he churned. A sales lead at a 12-person startup who wanted to test AI tools. She signed up, ran one campaign, got distracted, and never logged in again. A consultant who wanted it for a client. She left after 2 months when the engagement ended.

The solo founder taught me that patience isn’t a virtue in a customer. Waiting 8 months to complain means I fixed the wrong things for 8 months. The sales lead taught me that ‘I want to test AI tools’ is not a buying reason. It’s curiosity. The consultant taught me that single-use cases are bad bets.

Month 3: The First Time I Almost Quit

Month 3 is when the math got scary.

Revenue: ~$1,400 MRR. Burn rate: ~$4,200/month. Runway: about 6 more weeks.

I made a spreadsheet that could justify either decision. Column A showed why I should keep going. Column B showed why I should stop.

The number that stopped me from stopping: one customer sent me an unprompted email saying our tool had changed how they think about prospecting. Not that it saved them time. That it had changed how they think.

That’s a different category of response than ‘it’s useful.’ Useful things get replaced. Things that change how you think become infrastructure.

I stopped the spreadsheet exercise and focused on finding more customers who would send that kind of email.

What Actually Moved the Number in Month 4

Month 4 we hit $3,200 MRR. What changed:

I stopped calling it ‘outreach automation.’ Started calling it ‘prospect research.’ Same product. Different customer entered the conversation already understanding why they needed it.

I did 12 direct outreach calls in one week. Not demos. Conversations about their current process. 4 became customers.

I added one line to our onboarding email: ‘What does a successful first week with Genie look like for you?’ The answers rewrote our feature priority list.

The tool we build today is Genie 007 — genie007.com. The outreach research and personalization layer we started with is still the core, but the problem it’s solving has clarified significantly since month 1.

The Part Worth Saving From This

If you’re in months 1–3 of building something: the numbers will be wrong. The customers will be wrong. Your hypothesis will be wrong.

What’s not fine is hiding from the wrongness. Every conversation your first customers have is a clue about where the real problem is. Follow the clues.

The founders who make it through month 8 are almost always the ones who treated their early customers as teachers rather than revenue.

— -

I’m Bill Kiani, founder of Genie 007 — an AI system that handles prospect research and outreach personalization so founders can focus on the conversations that actually convert. If you’re still doing this manually, check out genie007.com..Test save trigger

I remember thinking that was huge. It felt like proof.

By month 6 we had $2,200 MRR. Still zero outside funding. Two of us working out of my home office, my co-founder remote in another time zone. We were paying ourselves nothing.

By month 14, $11,400 MRR. Enough to pay one salary. Still not two.

I’m writing this because I keep seeing bootstrapping content that either skips the numbers entirely or shows the highlight reel: the moment someone crossed $10k MRR and the chart went parabolic. The actual slog in between gets compressed into “we worked hard and figured it out.”

So here’s what the math actually looked like.

The 14-Month Grind

Our CAC (customer acquisition cost) in the first 6 months was $0 because every customer came from a direct conversation. I was doing 8–12 cold outreach calls a week, qualifying by hand, following up by hand. Our close rate was around 22%. The problem wasn’t leads. The problem was that I was the bottleneck.

Every founder I’ve talked to who bootstrapped past $10k MRR hit some version of this wall. You build it, you sell it, you support it. You’re in every part of the value chain. The business can only grow as fast as you can personally.

What broke it for us: we stopped treating sales as something we had to do manually because we hadn’t earned the right to automate it yet. That’s a story we tell ourselves. The reality is that manual-only sales is a choice, not a constraint, and it’s often the wrong one.

When we built the first version of Genie that could handle prospect research and personalized outreach automatically, our weekly output went from 8–12 qualified conversations to 35–40. Same two people. Same hours.

Revenue went from $11,400 to $28,000 MRR in the next 4 months.

The Churn Reality

Here’s the number nobody wants to talk about: our churn in months 7–10 was 8.3% monthly.

That means we were losing almost 1 in 12 customers every month. We were growing, but we were also leaking.

We spent three months convinced the problem was the product. We kept adding features. The churn didn’t change.

When we finally did proper exit interviews (something I should have done from month 2), the answer was simpler and more embarrassing: customers weren’t activating. They signed up, got confused in the first 48 hours, and quietly left before they’d seen any value.

We didn’t have a product problem. We had an onboarding problem.

Fixing onboarding dropped churn from 8.3% to 2.1% in two months. That one change was worth more than every feature we built in the three months we were chasing the wrong problem.

The Real Cost of Bootstrapping

Everyone tells you bootstrapping means “you own it.” What they don’t tell you is that ownership costs you two things most people underestimate.

First: time. Not startup-working-hard time. Specifically the time spent on things that don’t compound. Invoicing, manual follow-ups, spreadsheet updates, responding to the same support question 14 different times in a month. That time doesn’t get you closer to product-market fit. It just keeps the lights on.

Second: certainty lag. When you’re funded, you have a board, advisors, and a team of people who force you to look at the numbers regularly. When you’re bootstrapped, you can go weeks without looking at your real MRR if things feel busy. Busy is not the same as growing.

We had a month where I genuinely didn’t check our metrics for 19 days because we were “heads-down building.” When I finally looked, we’d actually declined 4%. The features we’d built hadn’t shipped to production and two customers had churned quietly. We lost 19 days of response time.

What Actually Worked

I don’t want to write a “bootstrapping is hard, here are 5 lessons” piece because that’s not how it felt. What worked wasn’t lessons. It was small, specific decisions that changed the math.

Decision 1: Automate the thing that’s blocking you from selling, even if you think you can’t afford to. We couldn’t afford not to.

Decision 2: Talk to churned customers before you talk to happy ones. Happy customers tell you what you did right. Churned customers tell you what matters.

Decision 3: Pick one metric to own every week. Not 5. One. Ours was “number of customers who completed 3 outreach sequences using Genie.” When that number went up, everything else followed.

We’re now at $47k MRR as of September. Still bootstrapped, both founders paid, small team. The journey from $147 to here took 28 months.

The math isn’t glamorous. But it’s real.

If you’re building something in the sales/outreach space, or you’re early and trying to figure out the acquisition bottleneck, I write about this regularly and we built the tool we wished existed when we were stuck: genie007.com.

— -

Bill Kiani is the co-founder of Genie, an AI-powered outreach platform for SMBs. He writes about bootstrapping, sales systems, and what actually works at the early stage.

Eight conversations. Zero closed deals. Not because the partners weren't interested, but because I was pitching the wrong thing to the wrong person in the wrong way.

Here's what I changed, and how the next 3 partnerships closed in 6 weeks.

The mistake: I was selling when I should have been structuring

When you approach a potential partner as a salesperson, you're asking them to buy something. They have to evaluate risk, justify budget, get internal sign-off.

When you approach them as a collaborator, you're offering a shared upside. The question shifts from "is this worth paying for?" to "can we both win here?"

I was making partnership pitches that sounded like sales decks. Bullet points about features. ROI projections they had no reason to trust. Discount offers. None of it addressed what the other side actually needed, which wasn't Genie 007. It was their own business problem.

The reframe that changed everything

A partnership isn't a purchase. It's a structure.

The person on the other side isn't buying your product. They're evaluating whether working with you helps them do their job better, serve their audience more effectively, or reduce a risk they're already managing.

Once I started thinking about it that way, I stopped leading with my product. I started leading with their problem.

The 3 questions I ask before every partnership conversation

Before any conversation now, I answer three questions:

What does this partner actually want? Not "they want more revenue" -- specific. Are they trying to grow their email list? Launch a new product? Retain subscribers? Build credibility in a new vertical?

How does Genie 007 help their audience specifically? Not generally. If they have 4,000 productivity-focused subscribers, what exact thing does Genie do for someone in that audience? I need to be able to say it in one sentence.

What's the minimum viable offer that reduces their risk? Don't start with a rev-share model that requires trust you haven't built yet. What can you offer that requires almost nothing from them and still delivers something real?

The deal that made this concrete

The partnership I'm most proud of from that 6-week run was with a productivity newsletter. Around 4,000 subscribers, very engaged, niche audience of operations professionals.

The editor wasn't looking for revenue. He was looking for content. His newsletter went out every Tuesday and he was starting to feel the pressure of producing something interesting every week.

I offered to co-create one issue with him, built around a real workflow I'd automated using Genie 007. No money involved. No rev-share. Just a piece of genuinely useful content for his audience that happened to show exactly what Genie does.

The issue went out. 127 trial sign-ups in 4 days. His audience loved it because it was specific and practical. He loved it because it took the Tuesday pressure off.

After that, the conversation about an ongoing partnership was easy. We'd already built something together.

What I do differently now

The sales deck is gone. I don't use slides in partnership conversations at all.

Instead I show up with three things: a clear understanding of what they're trying to accomplish, a specific idea for how we could work together, and a first step that's small enough that saying yes is almost no risk for them.

Most partnerships that don't close aren't failing on trust or interest. They're failing because the ask is too big relative to how much credibility you've built.

Start smaller. Build the thing together. Then talk about the long-term structure once both sides have seen it work.

Bill Kiani is a UK-based entrepreneur building Genie 007 (genie007.com). He writes about sales, marketing, AI, and the founder grind.

Most teams building voice-first products think they’re solving a recognition problem. Get the transcription accuracy high enough, reduce the latency, handle accents better — and you’ll have a product people love.

They’re solving the wrong problem.

What most teams think voice AI is

The mental model is simple: voice = microphone → transcription → text → whatever happens next.

That model isn’t wrong. It’s just not where the real challenge lives.

Every major platform has gotten transcription remarkably good. Whisper is open source and accurate enough for most use cases. The recognition problem, for general speech in reasonable audio conditions, is basically solved. Yet voice products keep failing. Teams spend months on WER (word error rate) improvements while users churn over something else entirely.

The companies still struggling are the ones who optimized the input layer and assumed the rest would follow.

Where the actual complexity lives

Voice is fundamentally different from text in one way that doesn’t get discussed enough: it’s non-linear.

When you type, you revise. You backspace, restructure, delete entire paragraphs. The user’s intent and the document’s final state converge in a way you can control. There’s a clean handoff from thought to text.

When you speak, you revise mid-sentence. You trail off, restart, use filler words as cognitive load markers. “Take that meeting with — actually, just clear my Thursday after 2pm” is two instructions that contradict each other and depend on knowing the first one’s still pending.

Most voice AI treats this as a transcription edge case. It’s actually the core design problem.

The three things voice AI actually needs to handle

Intent resolution under ambiguity. “Call John” when you have four Johns in your contacts and different ones you call for different contexts. A voice AI that matches text and presents a disambiguation menu has already lost. The right answer needs to come from behavioral context.

Temporal context across a conversation. If you said “remind me to follow up with Sarah” ten minutes ago, and then say “actually, make that Tuesday instead,” the system needs to know what “that” refers to without being told explicitly. Most don’t.

Interruption and correction at the model level. Users don’t finish sentences the same way in voice as they do in text. Good voice AI doesn’t need a “done speaking” signal to know you’ve changed your mind. It models the shape of intent, not just the words.

Why product teams keep getting this wrong

Partly it’s incentives. Transcription accuracy is measurable. You can benchmark it. You can show a 94% → 97% improvement in a product review. The intelligence layer is harder to quantify and slower to build.

Partly it’s because the demo always looks better than the product. Demos are scripted. Real users are not. A founder can stand on a stage and ask their product a question they’ve rehearsed, and it’ll look like magic. The same user at 8am, coffee in hand, asking the same thing with slightly different phrasing and a background noise source from their kitchen, gets a worse experience.

I’ve watched teams spend six months optimizing recognition while users churn because the product doesn’t handle context. The transcription was good enough on day one. The reasoning wasn’t.

There’s also a demo-to-product gap that’s specific to voice. The people who build voice AI products are often heavy keyboard users. Their fingers are faster than their mouths for most tasks. They use voice in demo mode, not in actual daily workflow mode. The failure states don’t emerge in rehearsal.

The people who would actually benefit from voice-first interfaces — people who prefer talking to typing, people managing complex schedules in motion, people whose hands are occupied — these users hit the edge cases on day one. Their feedback rarely makes it back to product in a useful form before the team has already locked in the architecture.

What good voice AI actually looks like

The products that get this right don’t feel like voice interfaces. They feel like you’re talking to someone who actually knows what you meant.

That distinction sounds obvious but it’s hard to build. It requires the system to hold state across a session without being prompted. It requires confidence calibration — knowing when it’s certain enough to act and when to confirm. It requires modeling the user’s goals, not just their utterances.

The shift from “voice input that happens to work” to “ambient intelligence that understands context” is where the category is heading. Tools like Genie 007 (genie007.com) are betting on exactly this: not that they’ll transcribe better than OpenAI, but that they’ll understand what you meant across an entire session better than any single-turn interface can.

That’s the right bet. The companies that try to win on recognition are competing with OpenAI and Google, who will win that fight on infrastructure alone. The companies that compete on contextual understanding are building something those giants are structurally slower to ship.

What to watch for in the next 12 months

The voice AI products that break out will be the ones that treat voice as a continuous stream of intent signals rather than a sequence of commands.

They won’t be judged on whether they understand a single utterance. They’ll be judged on whether they can carry a coherent thread across an entire workday — remembering what you said this morning when you ask a follow-up question this afternoon, knowing that “that thing I mentioned to the team” refers to the Slack message you sent last week, not performing a search but retrieving context.

That’s a fundamentally different product than what most teams are building right now. It requires a different architecture, a different data model, and a different definition of what “working” means in user testing.

The transcription was fine. The memory wasn’t.

That’s the problem. And it’s one the market hasn’t solved yet.

I’d spent seven years in enterprise software sales. Good money. Stable career. No real complaints except one: I kept watching other people build things and wondering why I wasn’t doing the same.

So I quit. Built Genie 007, a voice-to-action system that works everywhere you already work. And then spent the next 18 months learning things no startup article had prepared me for.

Month one feels like freedom. Month four feels like a test you didn’t revise for.

The first three months were brilliant. I was shipping fast, talking to anyone who would listen, and genuinely loving the work. I had no boss. No pointless meetings. I was building something mine.

Then month four hit. The initial energy wore off and I found myself staring at a product I’d built features into that no one had actually asked for. Two weeks wasted on an onboarding flow that looked beautiful and solved a problem that didn’t exist. That’s when the self-doubt started showing up uninvited.

Progress is not a straight line. It is a staircase where most of the steps are invisible.

There were 14 weeks in months five through nine where I genuinely didn’t know if things were working. No big wins. No clear signal. Just showing up every day and doing the work anyway.

I almost quit three times. Once at month five when runway felt tight. Once at month seven when a bigger company copied the core feature and launched it with VC backing. Once at month nine when I couldn’t clearly explain what made Genie 007 different in a single sentence. Each time I stayed, things eventually became clearer. But that clarity always came after the dark patch, not before.

The hardest thing wasn’t the hours. It was the isolation.

People ask how I worked 16 to 18 hour days for months. The hours were fine. What I didn’t expect was how quiet it gets when you’re doing this alone. No team to bounce ideas off. No one to notice when something works. No one to blame when it doesn’t.

Month seven was the worst. I hit a wall for three weeks straight. Couldn’t ship. Couldn’t think clearly. Kept starting things and abandoning them. Eventually I took five days off completely, no laptop, no product. Came back with the clearest head I’d had in months. I learned that rest isn’t a reward for finishing. It’s a requirement for continuing.

Something unexpected happened when YC-backed companies copied the product.

Two funded competitors launched while I was still in early growth. Both had bigger teams, more polished interfaces, and real PR budgets. I watched them announce on Product Hunt and felt genuinely sick for about two days.

Then something interesting happened. Their users started asking for things that were already in Genie 007. Their roadmap looked like my backlog. Competition forced me to get sharper about what made us different: we work across every tool you already use, we support 140+ languages natively, and we’re GDPR and HIPAA compliant from day one. I’d known those things before. The pressure made me say them clearly.

The thing that actually moves the needle isn’t what startup Twitter says it is.

Startup Twitter says to build in public. Launch fast. Get press. Go viral.

None of that moved the needle for me. What moved the needle was doing 200+ one-on-one calls with real users in the first year. Not surveys. Not email sequences. Actual conversations. Half an hour each. Asking what they were struggling with, what tools they were using, what drove them mad about their current workflow. That feedback shaped almost everything Genie 007 became. The product that’s now live on all major platforms with 1,200+ clients across 140+ countries didn’t come from viral launch posts. It came from listening.

Where things stand now.

Genie 007 is live across iOS, Android, Mac, Windows, and Chrome. We have 1,200+ clients in 140+ countries. GDPR and HIPAA compliant. And we did it without a single pound of outside investment.

18 months in. No regrets. But I’d tell any first-time founder: ignore most of the startup advice you read online. Talk to real users instead. The truth is in the conversations, not the content.

Bill Kiani is a UK-based entrepreneur building Genie 007 (genie007.com). He writes about sales, marketing, AI, and the founder grind.

This isn't a review. It's a workflow breakdown.

Why I Started Tracking This

Two weeks ago I gave 12 specific AI tools one job each. Not "help me with X" — a real task with a measurable output. A draft that needed to go out. A report that needed filing. A decision that needed a response.

Seven of the twelve required more time with AI than without.

Not because the tools were bad. Because they were optimised for conversation, not for completion. There's a gap between "generates decent text" and "does the actual thing in the actual app where the task lives" — and most tools stop short of that gap and hand the output back to you like a gift.

I published a post about this on Indie Hackers and got a lot of responses. The most common question was: which tools actually worked?

Here's the honest answer.

The Stack That Actually Runs My Day

For writing first drafts of anything text-based: Claude or GPT-4. I don't have a strong preference between them right now. The important thing is that I treat them as a starting point, not a finish line. I write a sentence of direction, they produce a paragraph, I rewrite it in my voice. The back-and-forth takes about 40% of the time solo writing takes. That's the real number — not "2 minutes instead of 2 hours."

For structured thinking: Perplexity for research where I need cited sources I can verify, Claude for synthesis where I need something reasoned through. I've stopped using search engines for most research tasks. I still verify anything that will be published or shared.

For daily task management: I've tried almost everything. What works for me is voice. I talk through what needs doing out loud — a habit I picked up when I started building Genie 007. When you have to say a task out loud to capture it, you find out immediately whether it's specific enough to actually do. Vague tasks don't survive narration.

For everything that happens on my computer — filing, sending, updating, moving: Genie 007 (genie007.com). I built this because I couldn't find a tool that closed the gap I kept hitting. Every AI tool would get me to the output — a drafted email, a completed form, a filled-in CRM entry — and then hand it back to me. I'd have to open the app, navigate to the right place, paste the thing, and send it. Four steps that should be one.

Genie 007 is voice-to-action on your computer. You say what you want done, it does it wherever it lives — browser, desktop app, web form, anywhere. No integrations required, no setup, no copy-paste. I use it for the handoff from "AI produced this" to "it's actually done."

The Honest Limits

AI doesn't know what's urgent. Every AI task assistant I've tried has the same problem: it can help me with the task I describe, but it has no sense of what matters right now versus what can wait. Priority still comes from me. That's probably right, but it means AI tools haven't actually reduced the cognitive load of knowing what to work on — only the load of doing it.

It's still easier to ask than to specify. "Write me a follow-up email" is a bad prompt. "Write a 3-sentence follow-up email to a prospect who said they're interested but needs sign-off from their CFO, in a tone that's confident without being pushy" is a good prompt. The difference is that good prompting requires you to know what you want, which is sometimes the actual hard part. AI doesn't skip that step; it just punishes vagueness more visibly.

The tools that work have one thing in common. They don't hand the output back. They complete the action in the context where the action lives. A tool that writes an email is not the same as a tool that writes and sends it. Sounds obvious. Most tools stop at step one and call it AI.

What My Day Looks Like

Morning: I talk through what needs to happen that day. Not into a notes app, not typed — spoken out loud. Genie 007 logs it. I spend about 8 minutes on this instead of the 20 it used to take to review emails, a task list, and a calendar separately.

Mid-morning: Writing tasks. Whatever needs drafting — posts, emails, briefs — I use AI for the first pass. I've got a rough rule: if it's going to take me more than 15 minutes to write, AI gets the first draft. If it's under 5 minutes, I just write it.

Afternoon: Everything operational — CRM updates, file management, sending things, scheduling. This is where voice-to-action saves the most time, because it's all high-frequency low-stakes tasks that add up to 2-3 hours if you do them manually. I've gotten that down to about 45 minutes.

The One Question Worth Asking

Before you add an AI tool to your workflow, ask one thing: does it finish the task, or does it hand the task back to me?

If it hands the task back, it's a better starting point, not a better workflow. Both are useful. They're not the same thing.

The tools that finished the task for me: Genie 007 for computer actions, Claude for synthesis, Perplexity for research. Everything else has been useful occasionally and unreliable regularly.

That's the honest stack.

Building Genie 007 (genie007.com) — voice-to-action for everything you do on a computer. No integrations, no setup. You say it, it does it.

Building Genie 007 (genie007.com) — voice-to-action for everything you do on a computer. No integrations, no setup. You say it, it does it.

We weren’t lazy. We were busy being busy — optimising channels before we’d validated whether the underlying message was right.

Here’s what I wish someone had told us earlier: most early-stage marketing fails not because of channel selection, but because of message-market fit. You can have perfect distribution of the wrong idea.

This is the framework that finally worked for us at Genie — and the one I’d use again if I were starting from scratch today.

Step 1: Fix the message before you fix the channel

The instinct is to blame distribution. “Our content isn’t getting enough reach.” “Our ads aren’t converting.” “Cold email is dead.”

Usually the problem is upstream.

We spent six months pushing “AI-powered voice assistant for productivity.” Nobody cared. The same product repositioned as “dictate anything, anywhere — no typing required” started converting at 3x the rate within two weeks.

Same product. Same channels. Different message.

The test I use now: can a stranger explain what you do and why it matters in one sentence, after seeing your homepage for 10 seconds? If not, your marketing problem isn’t a channel problem. It’s a clarity problem.

Practical fix: write three versions of your value proposition — one for someone who’s never heard of you, one for someone comparing you to alternatives, one for someone who already tried something similar and it didn’t work. Which one is actually on your homepage?

Step 2: Choose one acquisition channel and go embarrassingly deep

We ran twelve channels simultaneously in year one. None well.

The temptation to diversify early is strong — it feels like risk management. In reality, it’s a tax on your attention. You end up with mediocre results everywhere instead of great results somewhere.

The founders I’ve watched build fastest picked one channel and made it work before adding a second. Not because other channels are bad — because going deep is how you find the insights that compound.

When we finally committed to content + community (primarily this platform and Indie Hackers), something changed. We weren’t just distributing — we were learning in public. Every response told us something about who our most engaged buyers were and what language made them lean in.

By the time we added paid as a third channel at month 18, we had message clarity, social proof, and a feedback loop. The paid channel worked immediately. It wouldn’t have in month three.

Step 3: Treat pricing as a marketing decision, not a financial one

Most founders price based on cost-plus or competitive benchmarking. Both are the wrong starting point.

Pricing signals positioning. It shapes who shows up, what expectations they bring, and whether they trust you enough to commit.

We learned this the hard way. We launched with a £4.99/month starter plan to lower friction. What it actually did was attract users who were price-sensitive enough that they churn at the first sign of friction. Our £24.99/month plan had a 40% lower monthly churn rate. Not because the product was different — because the buyer profile was different.

Counterintuitively, a higher price can drive higher conversion rates when it signals that you’re serious, that you attract serious buyers, and that the outcome is worth paying for.

The question to ask: who do you want to attract, and what price point signals membership in that group? Pricing isn’t just how you extract value — it’s how you attract the right people to begin with.

Step 4: Content marketing works — but only when you’re specific

The content graveyard is full of “10 tips for startup growth” articles that went nowhere.

What actually works is content that is specific enough to be genuinely useful to one person, not broadly relevant to everyone. The more specific your angle, the more deeply the right people connect with it — and the more likely they are to share it with people exactly like them.

Our highest-performing pieces weren’t the broad “how to build a SaaS” articles. They were things like “I removed our pricing page to force demo calls — conversions dropped 34% in week one” and “We got to 1,200 users before realising we had the wrong ICP.”

Specific titles. Real numbers. Honest failures. These perform because they’re genuinely rare. Everyone has generic advice. Almost nobody shares the actual data from their specific mistake.

The distribution strategy that follows from this: write something specific enough that when the right person reads it, they immediately think of two colleagues who need to see it. That’s your distribution engine. You don’t need reach — you need relevance.

Step 5: Fix the funnel before you fill it

This one sounds obvious. It never is.

Every time we added more top-of-funnel traffic before fixing a conversion problem, we amplified the problem. More visitors exposing a broken signup flow doesn’t result in more customers — it results in more wasted attention and inflated CAC.

We had a 68% drop-off at our mic permission prompt. Not because people didn’t want to give permission — because we asked for it before they’d seen enough of the product to trust us with it. Moving the permission prompt to after first use dropped exit rate from 62% to 38%.

That single change, not paid acquisition, was our biggest growth lever in Q3.

The audit I do before any new acquisition push: where is the highest exit point in our funnel, and have we actually tested a fix for it? If the answer is no, that’s where the next month’s focus goes — not on bringing more people to the top of a leaking bucket.

Marketing isn’t just about getting people in. It’s about understanding why they leave.

—

Bill Kiani is the co-founder of Genie, a voice AI tool for note-taking, dictation, and hands-free productivity. Building in public at genie007.com.vived 3 Years of Building (Most Don’t)

They picture someone talking and text appearing on a screen. Maybe a bit faster than typing. Maybe more convenient when your hands are full.

That’s a fine use case. But it’s not what voice AI is actually becoming. And if that’s the mental model you’re using to evaluate voice AI products right now, you’re going to miss what’s actually happening.

The Dictation Frame Is Too Small

Dictation is about input speed. Voice to text. You speak, it types. The product is measured by accuracy, latency, and how little you have to correct afterwards.

That’s a narrow problem. And it’s a solved problem. The accuracy is good enough. The speed is there. The tools exist.

So why are dozens of companies still launching “voice typing” products in 2026?

Because dictation is the entry door. The thing people search for. “How do I type faster with my voice?” “Voice-to-text for email.” “Dictation software that actually works.” These are real searches from real people with real problems.

But the people searching for voice typing are often looking for something bigger than they can name. They’re not frustrated because they type too slowly. They’re frustrated because their inbox is a nightmare, or their post ideas die before they become drafts, or they spend 20 minutes writing an email that should take 2.

The real frustration is not typing. It’s the gap between the thought and the finished thing.

The Shift From Input to Action

Here’s the frame that changes everything: voice AI isn’t a faster keyboard. It’s a way to close the gap between thought and action.

When you use Genie 007, you don’t dictate text. You say what you want to happen, and it happens. You say “write a reply to this customer complaint, professional but warm, keep it under three sentences” and you get the reply. You say “fill in this form with my standard details” and it fills in the form. You say “post this thought to LinkedIn” and it drafts and queues the post.

No integrations. No setup. Works in every app, on every website, across Windows, Mac, Android, and iOS. It doesn’t matter whether you’re in Gmail, Notion, a CRM, or a bespoke internal tool you built last year. It works everywhere because it’s not an integration — it’s a layer on top.

That’s a different product category. Not dictation. Action.

Why This Matters for People Looking for Voice Typing

Here’s the counterintuitive thing: the most compelling voice AI experience right now is one that looks like dictation on the surface and turns out to be much more.

You search for “voice typing for email.” You try a tool. It lets you dictate. Fine.

Then you try Genie and you say “draft the email I was thinking about to my client about the delay in project delivery” and it writes the whole email. Not just transcribes your words into text. Writes it. With the right tone. In the format you’d actually send.

That’s the moment where the mental model shifts.

This is the product category worth paying attention to in 2026 — not better dictation, but voice-to-action. Think-to-text. The gap between the rough thought in your head and the finished output, closed in seconds.

What’s Real vs What’s Hype

There’s a lot of noise in the voice AI space right now. Here’s the quick guide:

Hype: “AI voice assistants that can do anything.” Most of these are just Siri/Alexa with a better UI. They still require you to have the right app installed, the right integration set up, and the patience for things to go wrong.

Real: Specialized voice AI that’s narrow, reliable, and genuinely eliminates a specific category of work. If a voice AI tool can reliably clear your email backlog faster than you can do it yourself, that’s a real product. If it requires 45 minutes of setup and still fails 30% of the time, it’s demo-ware.

Hype: “Voice AI is about accessibility.” It is, but that framing undersells it. The second-language writing use case alone is massive — someone writing in English as their third language, getting native-level output because they can explain what they mean in their own language and the AI delivers it in English. That’s not accessibility. That’s a superpower.

Real: The workflow shift from keyboard-first to voice-first for specific high-value tasks. Email. Posting. Data entry. Commands. These are the categories where voice-to-action works now, not someday.

Where Voice AI Is Headed

In three years, the question won’t be “should I use voice AI?” It’ll be “why would I type this when I could just say it?”

But the products that win won’t be the ones that transcribe the fastest. They’ll be the ones that understand the intent behind what you said and deliver the finished result.

The race isn’t about accuracy anymore. It’s about understanding.

And understanding what you meant — not just what you said — is what separates dictation from action.

Bill Kiani is a UK-based entrepreneur building Genie 007 (genie007.com) — a voice-to-action AI that works everywhere: Windows, Mac, Android, iOS, Chrome. He writes about AI, sales, and the founder grind.

Not disliked. Hated. The kind of visceral avoidance where you’d rather answer every email in your inbox, reorganise your CRM, or sit in meaningless meetings than pick up the phone and dial someone cold.

For my first three months on a sales floor, I hit 80 dials a day and closed almost nothing. My close rate was around 4%. My manager was patient about it, which almost made it worse.

Then one afternoon he pulled me aside and said something that didn’t make sense at first.

“You’re not selling. You’re auditioning.”

He was right.

What auditioning looks like

Every cold call I made at the time had the same arc.

I’d open with a rehearsed line. Try to sound confident. Pitch the product early because I’d been told “get to the value fast.” Hope the person didn’t hang up. And when I hit an objection, I’d defend.

I was treating every call like a performance review. Like there was a panel of judges scoring how well I delivered the script.

That’s not selling. That’s auditioning for a part you’re afraid to get.

The tell is defensiveness. When someone says “not interested” and your gut response is to justify the product, that’s ego, not strategy. You’re trying to prove you weren’t wrong to call.

The three changes that moved my close rate from 4% to 11% in 6 weeks

None of these are “magic opener” tactics. They’re changes in how I thought about what the call was for.

1. Exit framing

Before I dialled, I told myself: the best outcome is a short, honest conversation. Not a booked meeting, not a closed deal — just a real exchange where both of us know within three minutes whether there’s something worth talking about.

That framing removed the pressure to perform. I stopped trying to hold people on the call. The strange side effect was that people stayed longer.

When you’re not trying to trap someone, they don’t need to escape.

2. Question first, pitch never

The only thing a cold call needs to establish is whether there’s a real problem worth solving. So I stopped leading with what the product did and started leading with what I’d observed about their situation.

Not “we help companies like yours with X.” But “we noticed you’re doing [specific thing]. Is [specific problem] something your team is dealing with right now?”

If yes, we talk. If no, I move on. No pitch, no objection handling. Just a genuine question.

This requires real research before every call. The upside is that when someone says yes, the conversation is already real. You’re not fighting for credibility — you’ve demonstrated it.

3. The 60-second hard stop

I started telling people in the first 10 seconds: “I’ve got a specific reason for calling and I think this could be worth two minutes. If it’s not, I’ll be the first to say so and let you go.”

This does two things. It respects their time explicitly, which immediately shifts the dynamic. And it puts me in control of the exit. I’m not clinging to the call hoping they don’t hang up.

Most salespeople are terrified of the call ending. The best ones treat it ending quickly as a neutral outcome.

What actually changed

In six weeks, my close rate went from 4% to 11%. Not because I found a better script. Because I stopped treating cold calling as a performance and started treating it as a research activity.

The calls I didn’t close were just as valuable as the ones I did. When a qualified person tells you clearly “this isn’t a fit,” that’s data. What changed is I started caring about what I learned from the no, not just what I could do to prevent it.

My manager had a line he used at the end of most sessions: “The best salespeople find people who don’t need convincing.”

That took me longer to understand than it should have. Cold calling terrified me because I was making it about me. When it became about the person on the other end of the phone, it stopped being terrifying.

I’m now building Genie 007 (genie007.com) — a voice-to-action tool that lets you say something once and have it turn into the finished action. Sales is still the skill I use every day, not just to talk to customers but to understand what problem we’re actually solving.

The principles from those early sales floors haven’t changed. Lead with a real question. Know when to exit. Find the people who don’t need convincing.

Bill Kiani is a UK-based entrepreneur building Genie 007 (https://genie007.com). He writes about sales, marketing, AI, and the founder grind.

Not because things are bad. Sometimes because things feel completely stuck. Sometimes because something outside you changes the game. Sometimes because you’re just exhausted.

This is what my six looked like.

Month 3: The positioning crisis

Three months in, I had a product that worked. I knew it worked because I’d built it to solve my own problem. But nobody was clicking the landing page, and the few who did weren’t converting.

My first instinct was: the product is wrong. Build something different.

My wife asked me a simple question: have you asked anyone why they didn’t sign up?

I hadn’t. I’d been assuming. Four customer conversations later, I realised the product was fine. My positioning was completely off. I was describing the solution instead of the problem. Changed the headline. Conversion rate changed within a week.

Lesson: the product failing and the positioning failing feel identical from the outside. Find out which one it actually is before you quit.

Month 6: The copycat moment

Six months in, I got a notification that a YC-backed company had just launched something that looked a lot like what I was building. I spent a day convinced it was over. Their brand budget was bigger than my entire revenue for the year.

Then I actually used their product.

It was a stripped-down version. It did one thing. Mine did ten. The gap between our visions wasn’t a threat — it was clarifying. They’d validated the market. They just hadn’t built the full product.

Lesson: when a well-funded competitor launches something similar, you have two options. Panic or study. Studying is almost always more useful.

Month 9: The plateau

First paying customers arrived around month seven. That felt like everything.

Then nothing for two weeks. Then another customer. Then three quiet weeks.

The plateau is the part nobody warns you about. You have proof the thing works. You have real users. But growth isn’t compounding the way you assumed it would once you had evidence.

What I didn’t understand: early customers come from you pushing hard. Sustainable growth comes from building systems. The plateau is usually a signal that the first method has run out of road and the second isn’t ready yet.

I spent month nine building the second method instead of doing more of the first.

Month 12: The isolation wall

Twelve months in, I hit something that wasn’t a product problem or a market problem. It was a mental one.

Solo founding is genuinely lonely in a way that’s hard to explain to people who haven’t done it. You have no team to debrief with. No colleague to complain to. Your wins are private. Your losses are private. And the people around you who want to be supportive have no reference point for what you’re going through.

What saved me was two things. First, I started treating my working hours as non-negotiable — not in a hustle way, but in a boundaries way. Eight hours of work was eight hours of work. Not twelve. Not fifteen. The quality of the decisions I made in those eight hours mattered more than cramming in extra time.

Second, I started connecting with other founders online. Not to network. Just to have conversations with people who understood the specific texture of the problem. IndieHackers helped more than any business book in that period.

Month 15: The feature crisis

A competitor shipped something that looked exactly like the thing I’d been planning to build next. Better UI, bigger team, press coverage.

I spent three days in a strange kind of paralysis — half wanting to pivot immediately, half wanting to give up entirely.

What I eventually realized: their feature was a demo. Mine was production-tested with real customers who’d paid real money for it. The gap between a features page and a thing that actually works reliably is enormous. They had the first. I had the second.

Lesson: depth beats breadth at the early stage. One thing that works really well is harder to replicate than ten things that kind of work.

Month 18: The calendar problem

By month eighteen, I had something real. Genie 007 (genie007.com) had passed 1,200 clients. Revenue was real. The team was growing. And I hit a new version of the quitting feeling I didn’t expect.

Not despair. Something more like: when is this allowed to be enough?

I’d been moving goalposts since day one. First paying customer — not enough. First ten customers — not enough. Profitability — still not enough. I’d trained myself to treat every milestone as a stepping stone rather than a thing worth pausing on.

The calendar problem: I couldn’t point to a single day in the last year that I’d marked as done. Every evening had a tomorrow. Every week had a backlog.

I fixed it by scheduling a proper review at the end of every month — not a planning session, a review. What happened. What worked. What I’m proud of. Thirty minutes. Calendar blocked. Non-negotiable.

The pattern

Looking back at all six moments, the thing that kept me going wasn’t resilience or mindset. It was timing.

I quit thinking about quitting right before each one resolved. Not because I knew it was about to resolve — I didn’t. But because I had one more thing to try, and trying it took enough time for conditions to shift.

The practical version of that: keep the thing alive long enough to find out what it actually is. Most founders who quit do so just before the learning that would have changed everything.

I don’t know if you’re in month three or month fifteen. But if you’re reading this at one of those inflection points, the only question worth asking is: what’s the next thing I can try?

Try that. Then see where you are.

—

Bill Kiani is a UK-based entrepreneur building Genie 007 (genie007.com). He writes about sales, marketing, AI, and the founder grind.