For nine years at Upstack Studio, every new deal closed through me.
The enquiries came to me. The proposals came to me. I wrote them, I chased them, and I did all of it while running the rest of the company.
Some enquiries got a reply the same day. Some sat for a week. A few I never answered at all. Not because I didn't care. Because one person can only do so much, and I was the bottleneck for everything sales.
I knew. I just never made the time to look at it properly.
That's the part most owners recognise. You already suspect your follow-up is leaking. What you don't have is proof, and a way to decide what to do next.
Here's what you'll do this week. Pull four numbers from your last 90 days of enquiries: how fast a human replies, what share get only one follow-up attempt, how many quotes get chased instead of forgotten, and how many open enquiries have no owner. Then read what they're telling you, and fix the system before you buy any tool.
And yes, AI helps here. Just not the way it gets sold to you. AI is very good at reading a pile of enquiries and telling you what's in it. It's terrible at deciding what your business should do next. Count those four numbers first. Then choose what to fix.
First, check that this is actually your problem
Don't fix this because it sounds true. Suspecting isn't the same as knowing, and which of the four numbers is worst decides what you fix. It costs you a week to find out.
Start with one list: every enquiry that arrived in the last 90 days. Email, web form, phone, WhatsApp, referral, the lot.
If that list doesn't exist anywhere, stop. That's your finding. You can't manage demand you can't see.
Once you have the list, pull four numbers.
Number 1: how long until a human replied. Not the auto-reply — anyone on your team who wrote back as a person, not just you. Write down the middle value and the worst one.
Number 2: what share got one attempt and nothing after. Count them, then divide by your total for the period. One email, no reply, silence.
Number 3: quotes sent, quotes chased, quotes closed. That's three counts, not one. Most companies know the first and the third. Almost nobody has counted the middle one.
Number 4: how many open enquiries have no owner and no next date. Pick ten at random. For each one, can you name the person responsible and the thing that happens next? Count how many times you can't.
Two of these numbers came up before, in where a B2B company should use AI first: reply times and quote close rates, there as one leak among several. This piece stays on that single leak and digs in: the chased count in the middle, your worst case next to your middle one, and the owner test. Same method, one target.
How to read the four numbers
| What you counted | What a bad answer looks like | What it's telling you |
|---|---|---|
| Time to first human reply | Your worst case is many times your middle case | Speed depends on who happens to be free, not on a system |
| Enquiries with one attempt only | A big share of the list | The second attempt is nobody's job |
| Sent, chased, closed | Sent is far bigger than chased | You pay to produce quotes, then let them go quiet |
| No owner, no next date | You can't answer for most of the ten | There's no system here, just memory |
Notice what's missing. There's no industry benchmark in that table, because you don't need one. Your own numbers, compared against each other, tell you enough.
A worked example
The figures below are invented to show the shape of it. They're not from a real company.
A company gets 120 enquiries in 90 days.
- Middle time to first reply: 6 hours. Worst: 9 days.
- 71 of the 120 (59%) got one attempt and nothing after.
- 64 quotes sent. 18 chased. 11 closed.
- Of 10 open enquiries, 7 had no owner and no next date.
Read it back. The 6-hour middle says the team can be fast when it wants to be. The 9-day worst case says speed is luck. And 18 chased quotes out of 64 says the follow-up isn't slow, it's missing.
That company doesn't have a lead problem. It paid for 120 chances, sent 64 quotes, and then chased only 18 of them.
Paid demand
71got one attempt and nothing after
64 quotes sent. Only 18 chased. The demand arrived. The follow-up did not.
The one-person test
Now the question underneath all four numbers.
Look at your list and ask who owns the follow-up. If the honest answer is one name, you've found your ceiling.
Do the sums on that person, still using the invented company above. Say a real follow-up takes ten minutes: read the history, write something worth reading, set the next date. Say three touches each. At 120 enquiries that's 360 follow-ups a quarter, or about sixty hours. Those two numbers are my assumptions, not measurements. Put your own in and the shape holds. Now add the proposals that person writes, the calls they take, and the job they already had.
That isn't a discipline problem. It's a capacity problem, and trying harder doesn't fix it. For nine years, that name was mine.
Where AI helps, and where it doesn't
AI is genuinely good at three jobs here.
- Reading the pile. Point it at 90 days of enquiries from every channel — email, web form exports, WhatsApp chat exports, call notes, referral records, whatever you've got in writing. What people asked for, which ones went quiet, which ones nobody answered. If your records are a mess, this is the fastest way to get numbers 1, 2 and 3, as long as you feed it every channel. Read email only and you get an email-only answer, not the real one. Number 4 you still have to do by hand, because a next date that was never set isn't written down anywhere.
- Writing the first draft. It reads the history and gives you something to edit in seconds. You still decide what goes out.
- Watching for silence. A machine never forgets that a quote went out eleven days ago and nobody replied.
It's bad at three others.
- Holding the relationship. The customer bought from a person.
- Judging price and scope. That's your risk, not the model's.
- Deciding what the business does next. Still your job.
So give the machine the counting, the watching and the first draft. Keep your people on the judgment.
One warning. If you buy an AI follow-up tool before you've counted, you'll automate a process nobody owns. Faster mess is still mess.
Build the system in four parts
None of these start with buying anything.
- One list. Every enquiry lands in one place, whichever way it arrived. Two lists means one of them is wrong.
- One owner per enquiry. A name, not a team. Teams don't follow up. People do.
- A clock. Every open enquiry has a next date. When that date passes, somebody sees it.
- A memory. The history sits where the next person can read it, not in one person's head.
Get those four right on paper first. A whiteboard and a spreadsheet will do. Then ask which tool does them for you.
What to buy, and in what order
Three questions before you pay for anything.
- Does it hold the one list, or does it quietly become a second one?
- Does it set the next date by itself, or does it wait for someone to remember?
- Can the person doing the work use it on a phone, on a bad day, without training?
Then buy in this order: fix the process, use the shared inbox or CRM you already pay for, automate the reminders, and put AI on top of that. Most companies I met were shopping at the last step with nothing in place at the first three. If you're tempted to build your own tool instead, read how to decide between building and buying software first.
Buy in this order
- FirstFix the process
- ThenUse the shared inbox or CRMyou already pay for
- ThenAutomate the reminders
- LastPut AI on top of that
Most companies start here
The first three steps come before AI.
Which tool does which job
Nobody paid to be named here and there are no affiliate links on this page. Match the tool to the job you just wrote down, not to the brand with the best home page.
For the one list. If enquiries reach you by email and web form, you want a shared inbox before you want a CRM. Help Scout, Front and Missive all do the same core job: one mailbox that several people can work, with a name attached to every conversation. Already paying for Google Workspace or Microsoft 365? Try a shared mailbox there first. Free and dull beats paid and clever while you're still learning what you need.
If a lot of your enquiries arrive on WhatsApp, that's a different tool. The free WhatsApp Business app covers more than one person: labels, saved replies, and one account linked to up to four more devices, so a colleague can answer the same number from their own handset. What you get is one shared view. The free app runs out when you need a named owner on each enquiry, work you can hand to that person, and handling you can check. That's when you move to the WhatsApp Business Platform through a provider. Wati sits at the simpler, cheaper end. Respond.io does more routing and reporting.
For the owner and the clock. This is what a CRM is really for. Pipedrive is built around it. Its whole model is activity-based selling, and it will show you every open deal that has no next step booked. That's Number 4, done by the software instead of by you. HubSpot has a free tier worth starting on, but check today's limits on users and contacts first, because they've changed. Zoho Bigin is the cheapest paid option. If your follow-up is mostly phone calls, look at Close.
For the memory. Same tool, one extra test. Does it save the email thread by itself, or does somebody have to remember to paste it in? If it's the second one, that's not a memory. That's a second list.
For the reminders. Only automate what you already do by hand. Zapier is the easiest to start on. Make costs less once the volume is real. Use n8n if you want to run it on your own server.
For the AI, last. A 90-day enquiry list usually has names, pricing and client detail in it, so this isn't a personal-account job. Use a business workspace with proper data-processing terms, pull in the minimum you need, strip what you don't, and check your retention settings and your right to process the records before anything leaves your systems. From there, you have two ways to read the pile with Claude or ChatGPT: export the 90 days into that business workspace, or connect your CRM or inbox directly through a native connector or MCP so the model queries the records instead of you handing over a raw file. MCP isn't automatically safe or seamless. It still needs the same permission checks as an export, just narrower, since you can usually limit what it's allowed to see. For drafts and for watching the clock, use the AI already inside your CRM before you buy a separate AI sales agent. A bolt-on that can't see your history writes confident nonsense.
Your next 30 days
- Week 1. Pull the four numbers.
- Week 2. Give every open enquiry an owner and a next date. Do it by hand. It's meant to be uncomfortable.
- Week 3. Watch what breaks. Whatever you forget twice is what you should automate.
- Week 4. Fix that one thing, and only that one thing.
Write your stop rule now, while you're calm. Compare one 90-day period against another at similar enquiry volume, not a bigger period against a smaller one. If the middle reply time and the share of enquiries getting only one attempt haven't moved, this is a capacity problem. You need another person, not another tool.
Start here this week
Pull number 2. Count the enquiries that got one attempt and nothing after, then divide by your total for the period. It takes an afternoon, and it's usually the number that stings.
When you have it, join the newsletter for what I'm learning building Second Team. Reply to the welcome email you get when you join, and tell me which of the four was worst.
Want somebody from outside to run the count with you? That's the AI Profit Opportunity Audit. It ranks where the money leaks, names the three fixes worth doing first, and puts them on a 90-day roadmap. It stops at the diagnosis. Nobody builds anything.
You probably don't need more leads this quarter. You need to know what happened to the last hundred.
