Launching today
Pebbles Ai
AI sales platform for modern B2B teams
392 followers
AI sales platform for modern B2B teams
392 followers
The only GTM orchestration platform you will need to successfully take your products & services to market. Pebbles AI is a Go-To-Market Operating System built for B2B revenue teams. It brings strategy, lead generation, outreach, sales, & shared company knowledge into one AI-powered workspace. Using neurosymbolic AI trained on your business, it helps teams plan campaigns, personalize outreach, generate qualified leads, & execute without switching between disconnected GTM tools.
Pebbles Ai
Hey Product Hunt! 🙋🏻♂️
My name is Dmytro Antoniuk, and I'm the Chief AI Officer at Pebbles Ai. I'm incredibly excited to introduce Pebbles Ai to the community today.
What are we solving?
Getting your first customers is the hardest part. You instantly get pulled in many different directions: market research, finding target leads, writing cold outreach, and managing follow-ups.
Before you know it, your day is swallowed by a chaotic pile of disconnected AI tools. Teams end up wasting time switching between tabs, losing context, and burning budgets on a dozen different subscriptions just to align sales and marketing. It shouldn't feel this fragmented, and it definitely shouldn't feel this overwhelming.
We built Pebbles Ai to collapse that heavy, expensive stack into one secure, unified workspace for your entire growth team.
Who It's For?
B2B growth teams and professionals across sales, marketing, and RevOps trying to land their first or next customers.
What is the solution?
Pebbles Ai brings your strategy, audience targeting, and outreach into a single, connected loop. It combines advanced neurosymbolic AI with real B2B growth expertise so teams can work together in one place.
Learns your product and tone of voice from your brand docs, so you never re-explain your business
Reads B2B market signals to find your ideal customers and split them into clear segments
Acts like an in-house strategy consultancy and content agency in one
Crafts multi-touch campaigns tailored to each segment automatically
Keeps your data private in tenant-isolated architecture (CASA II certified)
We are live in the comments all day! Check out trypebbles.ai, give it a spin with your team, and let us know your thoughts. We would deeply appreciate your support and feedback today! 🚀
@dima_antoniuk On the buyer side, most "AI sales platform" demos fall apart the moment they hit a messy CRM and a sales cycle that doesn't match the template. How much manual mapping does it take before the AI is actually useful, and does it augment the reps' judgment or try to own the first touch?
Pebbles Ai
@dima_antoniuk @artem_fedorovich
Good question! Most tools bolt AI onto a messy CRM, assume it is clean, and fall over on contact. We went a different way.
Pebbles does not depend on you fixing your CRM first. Fresh Leads brings its own enriched data through a three provider waterfall, so the AI is useful on day one rather than after a mapping marathon.
The reasoning layer also reasons about your actual motion, so a long, unusual cycle becomes the input it plans around, rather than the thing that breaks it.
On your second question, it augments. The assistants are built to make you win, so they push back rather than flatter, and the rep keeps the final call.
It's the Dutch-direct approach. We cannot stroke your ego AND make you successful. Choose one. We choose to build an honest, science-led approach that ACTUALLY moves the (commercial) needle.
Where it does act on its own, it is deliberately narrow. Auto SDR takes the first touch only on inbound, once someone has already raised their hand. It qualifies and warms that reply in three to seven minutes, then hands a human a ready conversation.
It never cold blasts strangers, and it never gets to close. Farming, not hunting.
Happy to show you on a live demo.
Interested?
@dima_antoniuk congrats on the launch Dima. Tell me, can Pebbles trace the underperforming campaigns back to a specific assumption: ICP, segment, offer, channel etc?
Pebbles Ai
@zolani_matebese Thank you very much, this launch means a lot for us as a team.
We not only provide the unified chat interface with UI/UX enhanced experience for human in the loop activities to delegate the work to be done to Assistants on the platform. They can have access to out communication channels and evaluate the performance and pick-up proactively to keep warm leads warm increasing the overall campaign success.
The other way around there is a possibility to qualify your strategy and the direction with Strategy Assistant that have access to high quality GTM knowledge base constructed from years of experience in the domain. This is what we use for our product to handle GTM needs and be able to compete with much bigger teams.
Pebbles Ai
@zolani_matebese Thanks Zolani, and good to see you here. Short answer, yes. That is exactly what it does.
Most tools tell you a campaign flopped. They stop at the what. Pebbles is built to find the why, and the why almost always lives in an assumption you made before a single email went out.
A campaign is really a stack of bets: this ICP, this segment, this offer, this channel, this message. When it underperforms, the Strategy Assistant works backwards through that stack to find which bet was wrong, rather than blaming the subject line and moving on.
In practice it separates the layers. Good opens and no replies points at the offer or the value proposition, not delivery.
Strong replies from the wrong titles points at the ICP or the segment
One channel flat while another sings points at potential channel fit
Same message landing in one segment and dying in another points at positioning
It isolates the variable instead of lumping it all into "that one did not work".
We use a science-based, domain-expertise led approach with all performance audit and data analysis requests.
@dima_antoniuk I've been using the app, and finally I can stay on one tab and be focused
Only one piece of improvement from my side is the navigation bar. It feels too massive, and at first I got lost in it. Over time, I got used to it and knew where to press without thinking, but for new users that many options at once may be overwhelming. It would be great to optimise it, so the navbar feels more concise, simpler, and clearer.
@dima_antoniuk Nice timing, man. Feels like a lot of teams are hitting this exact problem right now.
Pebbles Ai
@rchornovol appreciate it, thank you!
Lancepilot
Pebbles Ai
@raihanshezan Brand voice here isn't a static style guide you set once and hope it sticks. It's actively built through a neurosymbolic workflow inside Pebbles that structures how your voice gets defined – not assumed or inferred passively. Once constructed, it's saved to the centralised Library alongside your other company context, making it a living asset you can reuse and refine over time, not a passive setting buried in a panel.
From there, the Library acts as persistent memory. It holds your outputs and team knowledge so the AI draws on your real content, not a blank slate. To be precise about the mechanics: this is context-driven memory retention, not model fine-tuning in the ML sense. No weights are updated. The system gets sharper as usage grows because it's working from richer, more specific context.
On follow-ups, the logic is adaptive – not a fixed drip sequence. Assistants use stored context and reply signals to decide what comes next. It's not being retrained on the fly, but it's also not running a rigid script.
Your core question is the right one to ask: does this reduce downstream editing, or just move it somewhere else? Persistent context – including the brand voice you've explicitly constructed – is exactly how we're trying to solve that. One more thing worth knowing: tenant isolation is in place, so your data never crosses over with other clients. What you build stays yours.
Happy to go deeper on any part of this.
Pebbles Ai
@raihanshezan You clearly know this space at a veteran level, the kind of read that only comes from running real outbound and watching exactly where it breaks. Your sharp questions. One at a time. Here we go:
Trained on your own content, or a style guide you configure upfront (Brand Voice Creation Feature)?
Your own, and the way you build it is the fun part. It runs as a guided Q&A, about 2 hours, closer to a sharp interview than a setup form.
It pulls your voice out of your answers and your best existing writing, then hands you a full spec: a word arsenal you actually use, a forbidden list you never touch, your signature phrases, and the mechanical fingerprints like sentence rhythm and punctuation.
What comes out is a brand voice that is distinctly yours and, more to the point, one that actually performs.
Our Brand Voice Creation Feature was built on principles drawn from McKinsey strategy practice and Saatchi and Saatchi creative heuristics, then layered with persuasion science, communication sciences, and a full library of anti patterns.
So it does many things at once. It captures how you sound, sets you apart, and makes the voice ACTUALLY effective, not just a gimmick.
This is the first half of the puzzle.
How does it learn and hold the voice over time?
Two parts, depth and enforcement.
The depth is a layered stack sitting under every message in every feature (from Marketing Assistant, Auto SDR to Smartbox). It uses 7 layers as enforcement. All proprietary Pebbles IP, only 2 are general heuristics tuned to you.
From your foundation up to the surface:
Organisational intelligence, your company's source of truth
GTM knowledge base, proprietary best practices and business netiquette
Neurosymbolic logic, what to do in every case, even the edge cases
Applied persuasion sciences
Brand voice framework
Hyper-personalisation
Persona-centric writing
Cultural nuances
Every draft runs a final check against your spec before it leaves, so the voice stays put instead of drifting the way a fine tuned model does. This is around (a) precision, (b) accuracy, and (c) efficacy.
Think of the baby of a senior Saatchi and Saatchi copywriter and a marketing scientist. It has your style guide memorised, knows every persuasion principle, and never has an off day. As you approve and edit, the spec sharpens toward your style too (the cherry on top).
This enforcement is the other half of the puzzle.
Do follow-ups adapt to reply signals, or is it a fixed sequence?
They adapt. The logic reads the intent behind each reply, then acts on it:
An objection gets answered on its merits
A "not now" gets a gentle nurture
A no gets turned into a maybe
Not me gets the colleague in
Silence gets the auto follow up
The sequence bends to the symbolic signal instead of marching on like chatbot.
There is more IFTTT neurosymbolic logic built in, but I'll spare you the novel 😂
Pebbles Ai
@imtiaj_ahmad That gap is the whole reason why Pebbles Ai exists. Hearing it from someone who ran growth at 20 people is very interesting. You had the talent. You just did not have the war chest for the institutional playbook.
That's what we built. Enterprise firepower without the price tag.
Also, your skepticism is the correct default. "Thinks before it writes" is easy to print on a landing page and still be a templated-wrapper prompt chain.
A standard LLM predicts the most probable next words. That is the whole thing. It has no separate step that asks "is this true, and does it follow the rules." If the sentence looks right, it goes for it, even when it is confidently wrong.
With Pebbles Ai, the neurosymbolic layer adds a second system that reasons with explicit rules and a structured knowledge base.
In other words, Pebbles Ai is a neurosymbolic reasoning system, not a wrapper chatbot.
The neural half drafts. The symbolic half checks that draft against the rules of GTM and against grounded facts before it provides you the output.
If the draft breaks a rule or asserts something that is not in the data, it gets caught and corrected instead of sent. One half writes fluently, the other half checks the writing against logic and evidence. This massively simplified btw, the truth is much more complex.
A concrete example. Ask a normal LLM to personalise a cold opener using the prospect's recent funding. If it does not actually have that data, it will often invent a plausible one, "congrats on the recent Series B," because a confident guess reads better to the model than admitting it does not know.
That is how people end up congratulating a company on a round that never happened, which is a fast way to torch the first impression.
Our symbolic layer only uses a signal that exists in the verified lead data. No real funding event, no funding line. It reaches for a different, true angle instead. The model reaches for a nice sounding sentence. The symbolic layer reaches for a correct one.
The same logic covers strategy. If it drafts a plan that contradicts a constraint you set, or pitches an enterprise motion to a 30-person startup, the neurosymbolic rules catch the mismatch rather than letting a fluent paragraph paper over it.
In our own testing this cut errors to roughly a third of naive prompting, measured on HalluScore. Not zero, we would never claim that. A system that checks its work beats one that only sounds sure of itself.
Happy to run a live one. Give me a prospect and a claim you would want in the opener, and I will show you where it refuses to make something up.
Here are some actual stats:
Claude Opus (MAX) vs Pebbles Ai
Accuracy: 33% vs 87%
Precision: 57% vs 91%
Sales Efficacy, MQL to SQL: 15% vs 85%
Cost per usable reply: ~$5 to $7 vs $0.012
Hallucination on rule-bound queries: 31.4% vs under 2%
Interesting take on skipping open-rate tracking for deliverability reasons. I've seen the same pattern in my own outreach, decent opens, zero replies. How does the AI decide what actually counts as a good reply signal vs just politeness?
Pebbles Ai
@benjouss Yeah, open rates are basically noise at this point. Bots, preview panes, and Apple MPP have made the metric too unreliable to act on, so we made a deliberate call to drop it entirely. What Pebbles does instead is classify reply intent: positive, neutral, or negative – so a "thanks, not right now" gets tagged differently from a reply that actually moves the conversation forward. The system is specifically built to tell politeness from a real buying signal. Rather than opens, we track reply rate, positive reply rate, meeting booked rate, a message quality score the AI runs before anything sends, and deliverability and inbox placement.
The "decent opens, zero replies" pattern you described is exactly the problem we set out to fix – because opens are vanity and replies are the only signal worth optimising for.
@kuzmovych That distinction between "polite" and "moves the conversation forward" is the one that's hard to get right without over-engineering it. Makes sense to track reply intent rather than opens if bots and preview panes have made opens that unreliable. Does the model get better at telling the two apart over time per account, or is it more of a fixed set of rules across all your users?
Pebbles Ai
@benjouss Ha, "decent opens, zero replies" is something I hear often. An open tells you the subject line worked. It says nothing about whether anyone actually wants what is inside.
The silence tells you nothing worked once they did.
When a whole campaign gets opens and no positive replies, it is not bad luck. It is an equation you have not solved yet.
Picture outbound as a formula, not a sum:
Y (outbound success) = Strategy × Positioning × Targeting × Value Proposition × Offer × Netiquette × Communication Sciences × Persuasion Tactics × Lexical Semantics x Persona-centric Writing
One incorrect part of the formula (variable) drags the whole result toward zero, however strong the others are. Bad luck is just the name people give the equation when they have not solved for the variables.
So Pebbles treats a reply as a strategy signal, not just a lead status. A campaign of polite nothings gets read as a diagnostic. Wrong audience, weak hook, or a value prop that does not answer "so what".
The Strategy Assistant is built to pull the formula apart, so instead of running the same dud again next month with new subject lines, you fix the term that was actually near zero.
Or my personal favourite response to a dud campaign: MOREEE!!! More emails, more LinkedIn messages.
Doubling down on a formula that is still unsolved, which is a bit like flooring the accelerator when the handbrake is on.
On telling a real reply from mere manners, it reads for intent over tone. "Sounds interesting, will keep you in mind" is the corporate cousin of "we should grab coffee sometime". Warm words, empty calendar. A reply that asks about price or fit, raises a real objection, or floats a next step is the one that counts. The neural side reads the tone, the symbolic side checks whether anything was actually asked or committed to.
Does that make sense?
@emincanturan Good breakdown, thanks for the tip. The "one weak variable drags the whole thing to zero" framing is a useful way to think about it, most people just add more volume instead of finding which term is actually near zero.
Pebbles Ai
@tihomiropacic Totally fair point. We don’t expect teams to rip out everything at once. The better path is often to start with a single use case, get value fast, and then expand from there. Pebbles Ai is built to support that kind of step-by-step adoption.
Pebbles Ai
@tihomiropacic Appreciate it. This is exactly the concern we built around. Moving from N tools to a bundle is a real risk, so the sensible route is a foot in the door.
You can start with one module, Fresh Leads for pipeline or Smartbox for outreach, and run it beside your current stack.
No rip and replace, and no burning down the old setup before you have judged the new one. We don't want hostages as customers, we want our customers to be in love with us.
Customer that have Stockholm syndrome 🤣. Joke.
All kidding aside, the payoff grows as you expand. Because it is one operating system with a neurosymbolic brain underneath, each feature you switch on makes the others smarter.
The features within the OS share the same context but have different intelligence. That's why I personally call it your digital marketing-sales department.
Different people, different domain experts, different hard-skills, different expertise, different judgement, and different knowledge.
We built the platform in that image.
For the next 24 hours you can start free, no card, so the foot in the door costs you 3 minutes onboarding.
Congrats on the launch! Curious — what does the neurosymbolic part actually catch that a plain LLM would miss?
Pebbles Ai
@alex_tomilinThank you. This is my favourite question by a mile. Most people admire the Ferrari (the GTMOS). Don't really care about the engineering that went into the engine (neurosymbolic AI). Let me break it down in 3 levels.
Macro Level | What is it?
A base LLM is inherently a chatbot system. It predicts the most probable next word, brilliantly, but that is not enough for complex domains such as B2B GTM, Corporate Law, and Human Resources. Nothing in it stops to ask "is this true, and does it obey the domain rules".
Neurosymbolic AI is 3-step systems working together, with a check between them:
The neural half reads language and context, the way any strong LLM does
The symbolic half applies explicit logic, rules and a structured knowledge base
A verification step sits between that and the output, catching anything that breaks
So you get the fluency of an LLM with a reasoning and fact checking layer bolted underneath. Outputs are accurate, reproducible and explainable (even auditable) rather than a confident black box.
It is also rare: academic interest went from 112 papers in 2015 and 2016 to over 9,000 in 2025 and 2026 [Google Scholar], yet real production systems are almost nowhere, because building one needs machine learning, formal logic, knowledge engineering and domain science in the same room at once.
Meso Level | How did we build it?
The engine is not a weekend project. The numbers behind it:
18+ months of research before the first line of production code
70,000+ engineering hours in the proprietary neurosymbolic architecture
800tn parameter permutations in business communication alone
Over 3,000 rule-based IFTTT rules across all B2B use cases and workflows
7 specialised neurosymbolic cores under GTMOS™, each mimicking the top 1% domain experts
Every output reasons up through a layered stack. Seven layers are proprietary Pebbles IP, two are heuristics tuned to you: your organisational truth at the base, then the GTM knowledge base, neurosymbolic logic, applied persuasion sciences and brand voice, with persona and cultural nuance on top.
That knowledge layer encodes six GTM disciplines, persuasion science, neuromarketing, behavioural economics, competitive strategy, segmentation theory and sales methodology, codified from closely-guarded secrets.
Break a rule or assert something not in the data, and it gets caught and corrected before it is ever sent. It also doesn't agree with you. It cares more about your success, than your ego. It will not allow you to make mistakes.
Micro level | Why it matters?
Let's look at some numbers. First lets look Claude Opus on the Max tier with a single instruction versus the full Pebbles pipeline:
Accuracy: 33% vs 87%
Precision: 57% vs 91%
Sales Efficacy, MQL to SQL: 15% vs 85%
And at the architecture level:
82% lower error rate than LoRA fine tuning, because the architecture is structurally accurate rather than nudged
3x better gross margin than wrappers, because the reasoning is not rederived from scratch on every call
~2% hallucination on rule bound queries, versus 31.4% across real world use HalluScore benchmark
But don't take my word for it: Claude Opus sits around 33% factual hallucination on the public HalluScore benchmark, while neurosymbolic methods approach ~100% accuracy on rule based tasks [arXiv 2502.01657]. We are closing up at 98%.
Cost is where it gets almost silly. To rebuild one reply with a raw model:
Around 12 prompts per reply, each re sending 25,000 to 35,000 tokens of context
Roughly $5 to $7 per usable output, and that's not even top 1% reasoning.
About $5,000 to $7,000 a month at a outputs, before 500 hours of human prompting
Pebbles does the same job inside a subscription near $450 a month, all in. Same output, a fraction of the cost, none of the babysitting.
The neursoymbolic reasoning is what lets it carry complex, multi-faceted, and cross-functional B2B work all the way through. These are examples you can build and execute on, which is impossible with base-models or tools with wrappers:
A full go to market strategy, grounded in your ICP, positioning and live market signals, then turned into the campaigns that run it
A beachhead strategy, picking the wedge segment worth attacking first, sizing it, and sequencing the entry instead of guessing
Industry trend analysis read across macro, meso and micro signals, so you see the shift before it hits your pipeline
Investor decks and enterprise sales assets, two pagers, RFPs and proposals that hold up when a sharp reader pushes on them
An omnichannel outbound engine, from fresh leads to reasoned email and LinkedIn sequences to replies captured and qualified in Smartbox, built and run end to end
And this is only the current stage. We are makingthe first steps toward a true Jarvis for go to market: a system that can safely, securely and reliably run the work fully autonomously, with no human in the loop.
Try it yourself, break it, and see where it holds.
Pebbles Ai
@tehreem_fatima5 tool sprawl is genuinely one of the most exhausting parts of running a small team, so I'm really glad this resonates.
What we do make easy is bringing your contacts in via CSV, Google Contacts, or Microsoft Contacts, so you're not starting from scratch.
Deep integrations with products like HubSpot, Attio, Monday.com are already on our roadmap for Q3. And we are not going to stop there and gonna bring our own agent-first CRM solution to support the vision of replacing 10+ tools
Pebbles Ai
@tehreem_fatima5 That is the exact nightmare everyone is in. Whether you are an early-stage startup of two co-founders or a mid-market organisation with 1,000+ employees.
Our primary research with over 120 respondents (companies) that they use anywhere between 10-25 siloed, mechanical tools (commercial related) that don't talk to each other.
Their monthly OPEX is anywhere between 2k-20k USD. Insane.
Pebbles puts management, marketing and sales under one roof instead. It sounds cliché, but a sustainable GTM with compounding momentum requires strategy, marketing and sales to work harmoniously together.
Singing from the same hymn sheet as it may.
On migration, you do not need a clean data move to get value on day one. Fresh Leads brings its own enriched data through a three provider waterfall, so Pebbles is useful from the first login, not after an import project that eats your first fortnight.
For your existing pipeline, you do not have to burn the old place down before you move in. Run Pebbles alongside your current CRM and bring your data across at your own pace, so nothing breaks halfway through a quarter.
In fact, we built it in such a way that it actually compliments your existing CRM tech stack.
For the next 24 hours you can start free, no card, and see how it feels once you are settled in. Let me know if you want a demo.