
Pinned Tweet
Replying to @typesafeai
@typesafeai Jev is INSANE.
I gave it a rent roll.
60 tenants, 22 columns, 1,320 cells. Every cell checked against the lease clause that actually governs it. Not the latest amendment, the right one.
Jev doesn't read the lease. It just gets the two numbers and decides.
7 typed questions per cell, a probability for each answer. Match, mismatch, conditional, or "you haven't shown me enough."
9,240 decisions. ~710ms each, 16 at a time. About a minute. $0.09.
It found $1.416M of rent missing from the roll. A tax base year an amendment reset three years ago. A letter of credit that expires 18 months before the lease does.
And the cells it couldn't prove? It greyed them out instead of guessing.
That last part is the whole thing.
If you want this running on your portfolio, that's what we do at @fifthdimensionai.
Comment "RENT ROLL" and I'll run yours.
4
446
the best distribution channel most small b2b teams have is already on payroll.
company pages broadcast. people get trusted. here's how i'd set up team posting without it turning cringe:
1. one lane per person. the engineer posts what they shipped and why. whoever runs demos posts the questions buyers keep asking (never names). the founder posts the bets
2. write from the work, not about the company. a screenshot of a messy spreadsheet beats a launch graphic every time
3. don't hand people copy. everyone can smell a marketing draft. give them a list of prompts and let them write it in their own words, typos included
4. one company mention every 4 or so posts, max. the rest is just useful
5. 15 min every friday, everyone drops one thing they learned from a customer or a build into a shared doc. that doc is next week's posts
6. track replies and dms from people who look like buyers, not likes
we sell into real estate and the pattern holds. buyers there trust a person they've been reading for months way more than a logo.
start with step 5. the posts write themselves once the doc fills up
1
3
222
benchmarks say astra and fable are basically tied. my workday says otherwise
I'm staying on claude. not because it's smarter on a chart, because it does what I asked the first time. when you run agents all day, following instructions beats a few extra points on an index
cheaper is great. reliable is what I pay for
2
3
251
AI made building the easy part. the cofounder most technical founders are missing now is distribution.
if that's you, this is the first 30 days I'd run before hiring anyone for marketing:
1. write who it's for in one sentence, in their words. pull the exact phrases from your first 10 call transcripts (granola is great for this), not from your landing page
2. pick one channel and ignore the rest for 30 days. B2B: linkedin + direct outreach. consumer: short video or X. one channel done properly beats two half done
3. build a list of 50 people by hand. not 5,000 from a scraper. you should be able to say why each one is on it
4. send 10 personal messages a day. one line on why them, one line on what you built, ask for 15 mins of feedback, not a sale
5. post what you shipped every day you ship. screenshot or video, two lines. building in public is the cheapest top of funnel there is
6. turn every call into content. the objection you heard on tuesday is wednesday's post
7. keep one sheet: name, where they came from, replied, booked. after 30 days the sheet tells you which channel to double down on
it's the same loop we run selling AI into real estate, just smaller.
building gives you feedback in seconds. distribution takes weeks. that's why it starts on day one, not after launch
3
4
326
the playbook that works in the US breaks the moment you sell into Asia.
I run gtm for apac and most of what I learned from us growth content needed rewriting:
1. one region is really six markets. singapore, hong kong, japan, australia, india and korea each buy differently. a single "apac" campaign is usually aimed at nobody.
2. intros beat volume. a warm intro from someone they respect does more than a month of sequences. so I spend less time on lists and more on who knows who.
3. the deal is never one person. there's the person you demo, the person who signs and the person in head office who can say no. you need a story for all three.
4. in person still matters. one coffee or one event conversation moves things faster than ten emails. the trip is part of the funnel.
5. patience is the strategy. cycles run longer, but once a group trusts you they stay and they refer you to their peers.
the tactics travel. the timing and the trust don't.
3
292
buyers are already asking chatgpt who to shortlist.
this is how I actually check whether we show up:
1. write 20 questions a real buyer would ask, in their words
2. run the same set in chatgpt and claude, plus perplexity. save the answers
3. note if you get named, who else does, and what they get wrong
4. run it again every week. only keep the questions where the answer actually moved
the useful part isn't a score. it's catching a wrong description before a buyer believes it.
1
254
most event follow-up dies in the attendee spreadsheet.
now I hand the list to one bot after an event and get a ranked follow-up plan back. works the same for a trade show or a webinar registrant list. the setup:
1. score every name against the ICP I pick for that event. I tell it which one, it doesn't guess.
2. clean before you rank. tag everyone as buyer, vendor or spam, internal or test. a lot of any list is people you shouldn't email at all.
3. check the CRM. pull the owner and deal stage for each person, mark them existing pipeline or net-new, and flag owners who've gone inactive.
4. check the contact ledger, one record of everyone we've contacted, so nobody gets reached twice by two people or two tools.
5. add intent. anyone who watched our demo videos moves up.
6. rank into 3 tiers. tier 1 gets a drafted follow-up in my gmail drafts within a day of the event.
7. route it. existing pipeline goes to whoever owns the deal, never into a new sequence. net-new goes to the pipeline queue.
it's read only in hubspot and sends nothing. I read every draft first.
the step I care about most is 3. the worst follow-up is a cold email to someone a colleague is already talking to.
3
3
269
chatgpt has been talking about your business for months.
now you can see exactly how often it mentions you, how many people view your profile, how many click through to your site, and what topics people find you through.
it's called business hub. to claim yours:
go to chatgpt.com/business-hub
search for your business
select your listing
verify ownership by email, phone, DNS or shopify
complete identity verification
who should do it? anyone with an online presence. local businesses, ecommerce, saas, agencies, consultants, publishers.
and in real estate: brokerages, property managers, funds. your tenants and investors are already asking chatgpt about you.
the part I care about most is the topics data. for the first time you can see what people actually ask before chatgpt brings you up.
still in beta, but a pretty big deal for SEO. I'd claim yours asap.
1
1
285
I run APAC go-to-market for an AI startup selling into real estate.
in the last few months we got demos with some of the biggest real estate groups in the region. it wasn't a spray and pray email blast. here's what actually worked:
1. show up at the events those teams already attend
I registered us for a local industry conference, talked to people, and followed up the same week.
2. follow up on the exact thing they told you, not a generic pitch
one conversation was about AI costs rising as more people used the tools. I wrote back about that, and asked for an intro to the team that picks them.
3. if a meeting feels like too much, invite them to a live webinar instead
lower commitment than a demo. then send no-shows the recording plus one use case they can actually try.
4. write to investment, asset management and leasing people about the documents that eat their week
leases, rent rolls, investment papers. ask for 15 or 20 minutes, follow up twice, then ask clearly if it's a yes, later, or no.
5. host small gatherings too, not only go to other people's conferences
drinks, a webinar. people from those firms can show up without committing to a sales call first.
no paid lists. just be where they already are, follow up on what they said, and give them an easy next step.
3
7
567
you know @FiachraRM is winning when his videos end up on my FYP 🔥
taking a bet that @thsottiaux 28 days of shipping will give at least 10 resets. upgraded myself to the $200 plan while everyone downgrades.
268






