The Tool Your Team Needs Isn't Search — It's Memory

By Milo Team · April 14, 2026 · 5 min read

Every six months, a new enterprise knowledge management tool gets funded and announced. Glean. Notion AI. Guru. Confluence AI. They all share the same premise: make it easier to search across all the places knowledge lives. The retrieval problem is real. But it's not your actual problem.

The retrieval trap

To retrieve knowledge, it first has to exist somewhere retrievable. That's a constraint that sounds obvious but has enormous implications. Every search tool — no matter how sophisticated — is only as good as what was written down. You can't search for a conversation that never got documented. You can't surface the insight that lived in someone's head.

Organizational psychologists call this the tacit knowledge problem. Studies consistently find that 70–80% of organizational knowledge is tacit — it exists in people's minds, not in documents. This figure traces back to Nonaka and Takeuchi's foundational knowledge management research and has been replicated across industries. It's the stuff that experienced employees just know: how a client really makes decisions, what a particular vendor's contract terms actually mean in practice, which internal processes work despite what the handbook says.

Enterprise search tools — however smart — operate entirely on the 20–30% that was written down. The rest is invisible to them by design.

What tacit knowledge actually looks like

In an agency or consultancy, tacit knowledge is everywhere. It looks like this:

None of this is in any document. It was never going to be. People don't sit down after a client call and update a wiki. They move on to the next thing. That's not a discipline failure — it's just how work works.

Why search can't solve this

When a search tool indexes your Slack, email, Drive, and Notion, it can help you find things that were explicitly written. That is valuable. But the information that matters most — the judgment-shaped knowledge that comes from experience — was never written in the first place. You cannot index what does not exist.

There's also a freshness problem. Even when knowledge is written down, it goes stale. The wiki page about a client's preferences from 2023 might have changed completely since then. Search tools surface what exists; they have no mechanism for ensuring it's still true. A confident result from a two-year-old Confluence page can be worse than no result at all.

These are fundamentally different problems with fundamentally different solutions. Competing on search quality — Glean vs. Notion AI vs. Guru — is fighting over the same constrained surface. The floor, not the ceiling.

The memory-first alternative

The question is whether there's a fundamentally different approach. Instead of indexing what already exists, what if you captured knowledge as it's created?

This is the distinction between retrieval and memory. Retrieval finds what was stored. Memory captures what's happening — continuously — and keeps it current.

When an employee is exchanging emails with a client, those emails contain knowledge: preferences, decisions, relationship dynamics, open issues. When they're on a call and someone mentions that the timeline slipped, that's knowledge. When they update a doc and note why a strategy changed, that's knowledge. All of it is signal. None of it requires anyone to decide to document it.

The capture-first approach treats all of this as continuous input to a knowledge base — not as material that needs to be manually curated or structured.

What this means practically

For individuals: Milo accumulates context from Gmail, Calendar, and your actual work. It knows what you know about your contacts because it has observed the relationship develop over time. When you prepare for a call with a client you haven't spoken to in three months, Milo surfaces what has changed, what was promised, and what that person cares about — without you having to search for it.

For teams: every employee's Milo instance is feeding a shared intelligence. The orchestrator can answer “what does our team know about this client?” by synthesizing across all employee contexts — not by searching documents, but by querying accumulated memory. The answer reflects what's actually known, not just what was written down.

The result is a knowledge base that builds itself. No documentation overhead. No discipline required. The knowledge exists because people did their jobs.

The implication for your team

If your agency uses Glean or a similar enterprise search alternative, you're solving half the problem. You're making the knowledge that was written down more findable. That's useful.

But the knowledge that walks out the door when your account manager quits — the three years of client context she accumulated — that was never written down. Search can't surface it because it doesn't exist in a form search can reach.

Memory-first tools build the knowledge base continuously, so it exists when you need it, regardless of whether anyone thought to write it down. That's not an incremental improvement on search. It's a different category.