Aartha logoAartha

The category

The Customer Memory Graph.

Your CRM stores what someone typed. The memory graph records what is true, when it changed, and where the evidence came from — so risk arrives with a receipt instead of a score.

Definition

What is a Customer Memory Graph?

A Customer Memory Graph is a cited, time-aware model of what is true about a customer account — the stakeholders, goals, commitments, risks, and outcomes — recording when each fact became true and what evidence supports it. Unlike a CRM record, which stores current field values, a memory graph preserves change over time so it can explain what moved and why.

Why it exists

The system of record stopped being enough.

The truth lives in conversations

A CRM captures what someone remembered to type after the call. The decision-relevant information — who actually holds budget, which commitment slipped, what objection was raised and never resolved, who has gone quiet — is spoken in meetings and written in email, and no system of record understands it.

State overwrites history

When a contact field is updated, the previous value is usually gone. But the change was the signal. A record that shows only the current sponsor cannot tell you the sponsor was replaced six weeks ago and has not attended a call since — which is precisely the fact that would have predicted the loss.

AI without memory produces disposable output

Summarisation is close to commoditised. What it does not do is resolve contradictions: deciding that the champion named in March is no longer the champion, or that a stated goal has changed. Summaries accumulate. Memory reconciles.

How it works

Four properties that make memory useful.

01

Facts, not documents

Signals are reconciled into discrete facts — this person is the economic buyer, this commitment was made, this risk is open — rather than stored as text to search later.

02

Bi-temporal by design

Each fact records both when it was true in the world and when the system learned it. That is what lets Aartha answer "what did we know, and when did we know it".

03

Provenance on everything

Facts keep a link to the meeting utterance, email, or CRM change they came from. Any claim can be traced to its source.

04

Change is recorded, not overwritten

When new information contradicts an existing fact, the transition is preserved. History becomes queryable instead of being destroyed by the update.

05

Diffing produces the signal

New information is compared against what is already known. A contradiction of a prior fact is far higher-signal than a fresh observation, and it needs no rule configured in advance.

06

One memory, every workflow

Health, risk, renewal forecasting, meeting prep, and drafted actions all read the same account memory, so they cannot disagree with each other.

FAQ

Questions, answered.

What is a Customer Memory Graph?+

A Customer Memory Graph is a cited, time-aware model of what is true about a customer account — the stakeholders, goals, commitments, risks, and outcomes — recording when each fact became true and what evidence supports it. Unlike a CRM record, which stores current field values, a memory graph preserves change over time so it can explain what moved and why.

How is a Customer Memory Graph different from a CRM?+

A CRM is a system of record: it stores the fields people type in, organised around deals and contacts, and shows current state. A Customer Memory Graph is a system of understanding: it reconciles meetings, email, and CRM activity into facts that carry both a timestamp and a source, so it can tell you the champion changed in March, a commitment made in Q2 was never delivered, and where each of those claims came from. The CRM remains your system of record; the memory graph sits above it.

Why does time-awareness matter?+

Because the most valuable signal in a customer relationship is change, not state. Knowing an account has one executive sponsor is mildly useful; knowing the sponsor was replaced six weeks ago and the new one has not attended a call is what predicts churn. A system that overwrites facts as they update destroys exactly the information that would have warned you.

What does "cited" mean in practice?+

Every important fact retains a link to the source it came from — the specific meeting utterance, email, or CRM change. So when Aartha reports that an account is at risk, you can click through to the sentence that caused it. This is what separates evidence a CSM will act on from a score they learn to ignore.

Is this just RAG over my customer data?+

No. Retrieval fetches relevant text at query time and reasons over it, which produces fluent answers that can quietly contradict what the same system said last month. A memory graph resolves contradictions into a maintained current view: when new information conflicts with an existing fact, the change is recorded rather than the old fact being silently replaced. Retrieval is stateless; memory accumulates and reconciles.

Why is the memory the moat rather than the model?+

Models are replaceable and improving rapidly for everyone. The durable asset is the accumulated, reconciled understanding of your specific customer relationships, which compounds with every interaction and cannot be reconstructed by a competitor from scratch. The longer it runs, the more accurate it becomes and the harder it is to replace.

Your next account move is already in the signals

Watch the memory change on your own accounts.

Connect your CRM, email, and calendar. In about a week you will see facts you did not know were in your own conversations.