Vaibhav Verma is the founder of MachGTM. He was CTO at Salesken, where his team built AI that guided salespeople during live customer calls. He started RevQ, now Mach, on 23 July 2026. He builds it in small steps, each with its test. His rule for a bug: first a test that fails because of it, then the fix. Lead search rules are locked, and each tuned source must pass its lock before it ships. He writes the Building Mach posts from the product's own history, mistakes included.
The best sales leader can build a strong strategy. But the team often fails to act on it, day after day, account after account. We built Mach to fix that.
Mach, measured from its own history on 20 October 2026: 1,225 commits since 23 July, 5,156 automated tests, and 82% of product changes carrying a test or check in the same change.
Posts by Vaibhav
Conversations · AI made the personal first line free
AI made the personal first line free, so buyers now read it as a template. 69% of US decision makers say AI-written email bothers them.
Sending · Gmail's AI now sorts your buyer's inbox
Gmail's AI Inbox is reaching paid work accounts in the US. What it ranks, what it sums up, and how to write a first email that still lands.
Sending · LinkedIn outreach in 2026: what gets restricted
LinkedIn bans tools that automate your account and blocks invitations that get ignored. What gets restricted, for how long, and how to write safely.
AI sales agents · Outbound agents arrive inside your CRM
An outbound sales agent is due inside a major CRM in November. Check the data, the approvals and the stop rules before you pilot it.
Conversations · Phones now ask cold callers why they call
iPhone and Pixel can now ask unknown callers why they're calling. What the buyer sees, the US rule on AI voices, and what to say.
Sales engagement · Sales engagement vs CRM: what belongs in each?
What belongs in your CRM and what belongs in a sales engagement tool. Map records, outreach, replies and owners before you add a tool.
Sales automation · Which sales tasks should you automate first?
Pick the first sales task to automate: one your team repeats and can check. A checklist, a made-up cost example and clear stop rules.
Building Mach · Research an account without adding more people
Research a company without adding new contacts. Keep that choice through retries, and count the work each choice creates.
Building Mach · A reply rule belongs beside the next step
Put reply rules beside the sales plan. Say who acts on each reply, and test silence apart from an interested answer.
Building Mach · Your sales playbook should show its work
A clear sales playbook shows how each of your instructions became a step. It flags gaps and keeps your words when drawing fails.
Building Mach · Saving a playbook is not permission to send
Saving, drawing, approving and running a sales playbook are four separate decisions. How to check each one, with a made-up cost example.
Account research · Why we hide leads that might be selling the same thing
16 leads on one firm's list might have sold the same help it sells. How Mach's seller check works, and why it now sets them aside.
Sales operations · The second day: what a daily lead search finds
A daily lead search ran a second day. It found 2,279 posts, 958 already known, and kept 87 leads. What repeated, what held leads back, and a review to copy.
Conversations · Why Mach refuses to write some first messages
Of 167 first messages for one firm, Mach kept 62 and refused 105. Why, the rules each draft must pass, and a check to copy before you send.
AI sales agents · How to run an AI SDR pilot you can score
A one-page scorecard for an AI SDR pilot, with our own numbers in it. Agree what counts, score each part, count the full cost and set stop rules.
Building Mach · How we build Mach: small steps and failing tests first
How Mach is built, measured. A typical change touches 6 files, 82% of product changes carry a test, and each bug left a rule.
Sending · 10,569 emails Mach refused to guess about
When Mach cannot match a sender to a known person, it holds the email and checks again later. One firm: 49,079 seen, 10,569 held, none dropped.
Account research · Researching one company: 10 minutes, 173 calls, 43 cents
We measured 457 research runs. 88% finished, taking 10.1 minutes and 43 cents on average. Here are the 173 calls by kind, and the run that lost $1.09.
Sales operations · The plan had 4,085 phrases. The budget ran 226
One real lead search planned 4,085 phrases on a $20 budget. It ran 226 queries and spent $5.23. Here is what each source kept.
Account research · We read 7 posts per person before writing one line
Before Mach writes to a person, it reads their recent posts: 6.9 on average across 294 runs. It found an honest opener for 64%.
Buying software · What one real lead search cost, line by line
One real lead search cost $5.23. Here is that bill split five ways, from search to contact-ready lead. And a check that cut one plan by a quarter.
Account research · We judged 1,635 public posts. 70 were real buying signals
One lead search judged 1,635 posts and kept 70 real buying signals. Here are the drop reasons, how each source did, and the funnel in one picture.
Building Mach · Lead search shipped in five days and seventeen rule versions
Lead search went from first commit to locked rules in five days and seventeen rule versions. Each was backed by a blind check. Here are the checks and the live lock.
Building Mach · Approval rails: every step from lead to meeting has a switch
Every step from lead to meeting has its own switch, review or automatic. Some rails never switch off. The rule for flipping a switch is the part to copy.
Buying software · AI sales agent pricing: calculate the cost of useful work
Price AI sales work by the results you keep. Pick the unit, count visible and hidden costs, and make every seller show their sums.
Conversations · Automated sales follow-up: when to stop and hand over
Know when automated follow-up should stop and pass the conversation to a person. The reply rule, the silence rule, and when to bring in someone senior.
Building Mach · Building Mach since 23 July, in numbers
Building Mach since 23 July, in numbers. By 17 October: 861 commits, 263 database changes, one repair, one new name, and tests rising from 116 to 3,026.
Sales operations · Which AI sales actions should need approval?
A map of which sales automation steps need a person and which can run alone. The controls that matter live in code, not habits.
Buyer research · Useful sales content for AI-assisted buyer research
Make pages that AI assistants can quote. Answer first, give every claim a source, use plain page structure, and stay honest when quoted out of context.
Building Mach · The Brain: check sources and correct what Mach learns
Read what Mach learns from your website and files. Check the sources, fix mistakes, and know what research can use.
Account research · Signal-based selling: from a buying signal to a useful message
Turn a buying signal into an honest first message. Four checks: what changed, why it matters, what you can prove, what to ask.
Account research · An account research brief for software services teams
What a services team should ask of an account research brief. A verdict that can say no, findings with sources, the people who decide, and open questions.
Building Mach · We rebuilt account research while it was already working
Our account research worked, until an answer key written by hand showed it found only 28% of the facts. Here is the rebuild, and the two ways a brief fails.
Sending · Email deliverability: a preflight check before automation
Check email deliverability before automation sends more. Prove sender checks on a delivered message, connect the stop list, warm up slowly and count every message not sent.
Sales operations · CRM data quality: what to fix before sales automation
Fix four things in your CRM before sales automation: company identity, contact owners, the stop list and allowed changes. We learned this when a real sync broke in seven places.
Sales automation · Sales automation in 2026: five changes to act on
Five sales automation changes for 2026, each checked against our own build. Start with one task you can measure.