Build a Customer Intelligence System (Even If You Have to Start With a Spreadsheet)

A free how-to guide for support and ops leaders who are tired of their agents knowing nothing about the human in the queue

If you read "Your Silos Are Killing Your Market Share (https://buttondown.com/betts/archive/you-silos-are-killing-your-market-share/)," you already know the problem: your customer has read the docs, searched the knowledge base, asked their LLM, and finally reached out to a human — and the first thing you do is make them start over.

That's more of an information problem than a support problem. And information problems are fixable as a team.

Here's the plan. It's written for support and ops leaders at SaaS companies, but the bones will hold no matter what you sell. Some steps will take an afternoon. Some will take a quarter. Start wherever your org's stomach can handle, and don't let "we can't do all of it" stop you from doing any of it.

Phase 1: Make the Money Case First

Nobody greenlights a cross-functional data project because it sounds nice. They greenlight it because someone did math.

• Pull your churn rate, customer acquisition cost, lifetime customer value, and cost per support interaction (human-handled vs. deflected).

• Go find your finance team. Not to ask for a budget yet — just to ask for numbers and context. Finance sees the whole business in a way almost nobody else does, and they are chronically under-invited to these conversations. Fix that.

• Stay besties with finance. You'll need them again at the tracking phase, and a standing relationship beats a cold ask every time. Besides, finance people are a gosh darn delight.

You don't need a perfect ROI model. You need a "here's what unexplained churn is costing us" number you can say out loud in a room with executives in it. It also helps to have these figures in hand as you go through the project. There will be costs you didn't know about and you'll find places to save or make money you'd never considered before.

Phase 2: Start Where Your Customer Actually Is

Not where your org chart says support lives. Not where you wish people would go. Where they actually show up.

• Pull your real contact-channel data: email, chat widget, community forum, social DMs, that one Slack Connect channel someone set up two years ago and never told you about.

• Rank channels by volume and by frustration signal — a channel with lower volume but higher escalation rate is telling you something.

• Resist the urge to redesign the intake before you understand it. Map first, fix second.

Phase 3: Audit What You Have vs. What You Wish You Had

This is where you talk to two very different groups and write down everything they say.

Ask your agents: What do you wish you knew the second a conversation starts? No idea is too small or too "pie in the sky." Write down all of it, including the stuff that feels impossible.

Ask every other team what they're already sitting on:

• Marketing — engagement data, campaign source, content they've already consumed

• Sales — discovery call notes, original use case, why they signed

• Product — usage data, feature adoption, error logs, rage-click signals

• CS/CSMs — check-in notes, QBR history, renewal risk flags

• Support itself — prior ticket history, sentiment trends

Common gold to dig for, regardless of team:

• How long have they been a customer, and why did they stay?

• What was their original reason for signing up?

• What have they been working on lately — and where have they gotten stuck?

• What error messages or friction points have they hit recently?

• Who do they work for, or what's their use case, if that context matters for your product?

If something you want doesn't exist yet, take it to engineering and ask what it would take to start collecting it. Use your social capital here. Go into debt on it if you have to — this is the kind of thing worth spending capital on.

Phase 4: Get It All in One Place (a Spreadsheet Is Fine)

Seriously. No new tooling purchase required at this stage.

• Build one shared spreadsheet that categorizes every data point you found: what it is, where it lives, who owns it, how hard it'd be to access.

• Pinky swear: no fancy apps until you've proven the concept works.

Phase 5: Sort by Lift, Not by Excitement

Get a cross-functional group in a room (or a doc, if your org runs async) and work through each data point with four questions:

1. What's legally or contractually off the table entirely? Push back on these. Sometimes a no can be a maybe if the data is stored or collected differently.

2. What could come back on the table, and what would that take?

3. What's low-lift vs. high-lift to implement?

4. What could ship now, even in a rough form, to get momentum going?

Remember: partial information beats no information. You're not trying to build the perfect system on day one. You're trying to stop your agents from typing "hello" into a void.

Phase 6: Track It Like You Mean It

Build a dashboard — again, a spreadsheet is genuinely fine — using the numbers from Phase 1 plus whatever new ones surfaced along the way. Over the next few months, watch for movement in:

• Customer sentiment

• Renewal conversations and outcomes

• Expansion or tier upgrades among customers who've interacted with support

• Support volume itself — don't panic if it goes up. Rising support volume can mean rising engagement, not rising failure. Zero volume is the number that should actually worry you.

Visibility beats polish at this stage. You're still experimenting.

Phase 7: Check In, Redistribute, and Only Then Automate

Schedule regular check-ins with your cross-functional crew and ask the unglamorous questions:

• What's working?

• What's painful?

• Who's getting buried under this project, and how do we redistribute that load?

• Now that this has been running a while, what's actually ready to hand off to a well-trained, well-monitored AI agent?

That last question only belongs at the end of this process, not the beginning. AI trained on fragmented, siloed, half-collected data just automates the guessing. AI trained on a system you've actually built on purpose can start closing the gap between "we deflected a ticket" and "we understood a customer."

Remember: your customer doesn't experience your org chart. They experience one company, one relationship, one conversation at a time. This plan is just a way to get your internal systems to catch up to that truth — one un-siloed data point at a time.