A build log on how B2B teams win customers.
Sequor AI is where I study revenue tech from the inside: mapping how a stranger becomes a lead, a deal and a customer, and prototyping a simple sales OS for small teams along the way.
A personal learning build by Smrity, shared in the open. There's nothing to buy and no demo to book. I'm building, writing down what I learn, and talking to the people who do this work every day.
From signal to renewal, in nine stages
Select a stage to see how it works and where it breaks.
What I'm building, and why
Three things I kept seeing while studying the market.
Small teams run on a patchwork
A database for contacts, an enrichment tool, a sequencer, a CRM and a few spreadsheets. Leads go cold in the seams between them, and nobody owns the seams.
The tools find leads well. Turning them into replies is harder.
Across six tools, list building mostly works. Outreach is where they diverge: sequences stay rigid once launched and rarely adapt to how a prospect responds.
Autonomy without control erodes trust
Fully autonomous AI SDRs promise a lot, and reviews show the gap between promise and execution. Reps want AI that does the work, and a clear way to steer it.
The bigger picture, in four phases
A simple sales OS for small companies, built one layer at a time. Each phase is a study first and a prototype second.
The revenue landscape, stage by stage
How a company goes from not knowing a buyer exists to renewing them, what each stage produces, which tools play there, and where things usually break.
Who covers which stage
Mapped from each tool's own positioning and its reviews. No single tool spans the journey for a small team, so the gaps become manual work.
The Reply Gap.
Where AI-powered outbound loses revenue: from the first signal to the moment a prospect answers, and what happens next. Three acts, from market research to a solution I can measure.
Market research
What the tools promise, and where their users say they fall short.
Three candidates, compared before committing to one
| SDRs at SMB and mid-market B2B SaaS Selected | Founder-led, early-stage startups | RevOps at 200 to 500 person scale-ups | |
|---|---|---|---|
| Reach | High, active on G2, Capterra and Reddit | Medium, scattered and less vocal online | Medium, smaller pool, B2B forums |
| Pain intensity | High, daily quota pressure | High, time-starved generalists | Medium, process pain rather than personal |
| Willingness to pay | Medium, needs manager buy-in | Low, bootstrapped budgets | High, owns the tooling budget |
| Fit with research | Strong, matches the known landscape | Partial, a different buyer motion | Partial, the buyer isn't the user |
What the tools cost, and who they're built for
Published price ranges on one log scale. Per-seat plans multiply with every rep on the team.
What users say each tool does better, and worse
Finding leads works. Turning them into replies doesn't.
What the reviews showed
Prospecting and list building work well across all six tools. Data drift exists, but it's minor and well known.
Where the real gap sits
Outreach execution is where tools diverge. AiSDR and Reply.io show limited branching and no A/B testing. Instantly has both, but usability and trust issues blunt it.
What that means
The unmet need isn't finding leads, it's turning them into replies. Sequencing needs adaptive intelligence, not more contact data.
As an SDR at a small or mid-market B2B SaaS company, I need outreach sequencing with adaptive branching and built-in A/B testing, while keeping meaningful control over the AI's decisions, so I can turn more well-prospected leads into real conversations, instead of losing them to rigid, one-path sequences that can't adjust to how a prospect responds.
User research
The hypothesis, tested against independent evidence. G2 reviews shaped act one, so they were left out here on purpose.
Five patterns, in five numbers
Rigid, one-path sequences and missing control are real, independent of G2.
A test button won't work at one SDR's volume. The need is a learning loop that pools results and judges on meetings, and the sharpest pain sits at the moment a prospect replies.
Meet Maya, a composite SDR built from the research
Maya Brooks, 26
SDR at an 80-person B2B SaaS company selling into US mid-market
- In role: 1.5 years, eyeing the AE seat
- Stack: Apollo, HubSpot, LinkedIn
- Her day: about 110 touches
- Measured on: meetings her AE accepts
"I don't need more leads. I need my sequence to notice when someone actually answered."
- Hit her meeting number every month, not just the good ones
- Spend her hours on conversations, not admin
- Prove she can run a book well enough to move up
- A "not now" or "talk to Priya" pulls the lead into manual work
- She can't tell which message works, the numbers are too small
- AI drafts that sound wrong, found after they're sent
- Happy to hand over research and first drafts
- Wants to see why it picked a step before anything goes out
- Auto-send only for situations she's watched it get right
When a prospect finally responds, help me give them the right next step quickly, without rebuilding the sequence by hand or losing control of what goes out in my name.
One sequence through Maya's week, and where it drops
Seven problems, scored out of 20
Frequency, severity, market gap and feasibility, each scored 1 to 5. Gap scores come from act one.
When a prospect replies with anything other than a clear yes or no, Maya's sequence has no next step. She handles it by hand or the lead goes quiet, and the AI tools that try to help don't show her why.
Route every reply to the right next step, while Maya stays in control of what goes out in her name?
Solution and proof
Reply-aware sequencing: every reply gets a next step, and the SDR decides how much the AI decides.
Reply-aware sequencing, in five moves
- Intent detection on every reply, across six intents
- Branch library: not now resumes on their timeline, referral drafts a warm intro, out of office pauses until return
- Approval queue with reason and confidence on each card
- Autonomy dial per intent
- Hard exit on opt out and clear negatives
- Pooled message testing, judged on meetings booked
- Branching on signals before a reply
- Thread summary for the AE at handoff
- A contact database or enrichment, Apollo and Clay own this
- Own sending and warmup
- A fully autonomous AI SDR
One revenue number, guarded on four sides
Meetings booked per 100 prospects who reply
Versus control, over an 8-week pilot with 5 to 8 SDR teams. Counts only meetings the AE accepts.
- Suggestions approved without edits
- 70%+ by week 6
- Intents switched to auto
- 2 of 6 by week 8
- Median time from reply to next touch
- Under 4 working hours
- Manual reply handling per SDR
- Down 50%
- Intent accuracy, weekly audit of 100 replies
- 90%+
- Spam complaint rate
- Below 0.1%
- Unsubscribe rate
- No higher than control
- Sends after an opt out
- Zero
Read the full study
Every slide, every source, in one PDF. And if you think I've got something wrong, I'd love to hear it.
What I'm prototyping
Two ideas answer the control gap: sequences that adapt to what a prospect does, and AI that earns the right to act on its own.
Adaptive sequencing
Step one tests two openers. Then the next step depends on what the prospect actually does. Select a behavior.
Opens with the prospect's recent hiring push.
Opens with a peer company's result.
Confidence-Tiered Autonomy
Every next step lands in a lane. High confidence and low stakes run on their own, the rest waits for the rep, and anything odd gets flagged. Autonomy is earned.
Runs on its own
Waits for the rep
Flagged to look at
Design principles
The rules I'm holding every Sequor AI idea to.
Confidence and stakes are separate
A message can be very likely right and still too important to send unseen. Both scores decide the lane, not one.
Every action shows its reason
The rep sees why a step was chosen and why it landed in its lane, so they build judgment instead of losing it.
Edits are the best training data
Each approval loosens the bar for that kind of step; each edit or rejection tightens it. The system learns the rep, not just the market.
Build log
What I've done so far, newest first. Short, honest, and updated as I go.
- NEXT
Conversations with SDRs, founders and RevOps
Moving from secondary research to real conversations, testing the P0 against live workflows: how people build lists, run sequences, and decide what the AI should and shouldn't do on its own.
- DONE
Published the discovery study
Finished user research and shaped the solution: five insights, a composite persona, a journey map, a P0 and a north star metric. Read it here, or open the PDF.
- NOW
Building Phase 1: prospecting and outreach
A working prototype of the first layer, starting with reply-aware sequences and the autonomy dial.
- DONE
Widened the vision to the full lifecycle
Decided Sequor AI is a study of the whole journey, not just outbound: prospecting and outreach, then pipeline and data, enablement, and post-sales.
- DONE
Named the core design idea
Confidence-Tiered Autonomy, borrowed from how automated reordering works in inventory planning: routine moves run, uncertain ones wait, anomalies get flagged.
- DONE
Turned an AI SDR on this very site
Tested an autonomous AI SDR hands-on by pointing it at Sequor AI itself, to feel the experience from the prospect's side.
- DONE
Finished the market study
Compared three segments, priced and positioned six tools, read their reviews, and landed on a hypothesis about the control gap in outreach.
I'd love to hear how you sell.
No pitch, no demo, no sales call. I'm studying how real teams find and win customers, and the best way to learn is to listen. If you're facing a problem in this space, or just enjoy thinking about it, let's talk.
- SDRs and AEsHow you build lists, run sequences and decide who to follow up with.
- Founders who still sellWhat you've stitched together, and what you wish just worked.
- RevOps and sales leadersWhere data breaks between tools, and what you'd never let AI do alone.
- Builders in revenue techWhat you're learning, and what I might be getting wrong.
Questions on my mind
If any of these spark something, I'd love your take.