Chapter 5 — Memory types & prospect profiles

The agent will talk to visitors who arrive knowing nothing about your business and leave with you knowing — ideally — quite a lot about them. Memory types are the schema for what the agent remembers about each visitor. Prospect profiles are named patterns over that memory state — labels like "MQL" or "Hot Lead" that classify a prospect for routing and triage.

This chapter is sequenced deliberately: skills (Chapter 6) and routing rules (Chapter 7) both depend on what you set up here. Get this right first.

By the end of this chapter you'll have:

• Reviewed the seeded memory types catalogue and toggled one off and on
• Added a custom memory type
• Configured one of the seeded prospect profiles (MQL) with criteria built from memory types

Note: the qualification page has been removed. Earlier versions of the product had a dedicated Qualification page for MQL/SQL thresholds and lead scoring. That has been retired — prospect profiles, covered later in this chapter, now do the work it used to do.

---

Opening the memory types screen

Click Prospects in the left-hand rail, then select Memory Types from the toolbar at the top of the page.

The page is organised into seven tabs:

• All — the default, showing every memory type grouped by category
• Discovery — what the prospect is trying to do (pain points, buying timeframe, target outcomes)
• Personal — who the prospect is (name, email, job title, contact details)
• Situation — context about their business (industry, org size, tech stack, geography)
• Solution — what they need from a solution (capabilities, integrations, success criteria)
• Value — the commercial picture (budget, ROI goal, decision criteria, approval process)
• Custom — memory types you add yourself

Each category shows how many memory types are enabled vs. available — e.g. Discovery: 5 enabled of 7. Toggles light up the ones that are on.

Enabled vs. required. Toggling a memory type on doesn't force the agent to ask for it on every conversation — it just enables the agent to capture and remember that value when it surfaces in conversation. Skills and routing decide when the agent should actively pursue specific memories. Toggle on liberally; the agent will not become pushy just because you enabled something.

Inspecting a built-in memory type

Click Open on any seeded memory type — for example, Pain Points — to see its definition in the right-hand panel.

Built-in memory types show their name, description, data type (Text, in this case), and settings (when to update, retention policy). They're marked Built-in · Read-only — you can't edit them but you can enable, disable, and use them in profiles.

Toggling a memory type off and back on

Find a memory type you want to disable and click its toggle. The "X enabled of Y" counter on that category updates immediately, and the agent will stop trying to capture that value on future conversations. Click the toggle again to re-enable.

The seeded set covers the standard B2B discovery axis well. Most tenants will leave it largely as-is, then add one or two custom memory types for concepts specific to their business.

---

Adding a custom memory type

Click Custom in the tab row to filter to custom memory types only. A fresh team has none.

Click Create Memory to open the create form in the right-hand panel.

The form takes:

• Name — up to 50 characters. What you (and the agent prompts) will refer to this memory by.
• Description — up to 200 characters. A short explanation shown to your team.
• Type — Text by default. Other types may be available depending on your plan and use case.
• When to update — When the value changes by default. Controls how aggressively the agent overwrites existing values when new information surfaces.
• Keep data for — 12 months by default. Retention period after which the memory expires.

The screenshot shows a custom memory type called Current Challenge — a free-text capture of what's prompting the visitor to engage. Pick custom memory types that map directly to the questions you need answered before a prospect is worth following up on. Once defined, they're available to skills and prospect profiles in the same way as the seeded ones.

Click Save. The new memory appears in the Custom section, defaults to enabled, and is immediately available to skills, routing, and prospect profiles.

Each custom memory has its own Configure and Delete buttons, plus the standard enable/disable switch.

---

Prospect profiles

Click Profiles in the toolbar at the top of the Prospects page.

The page lists six seeded profiles — Hot Lead, MQL, Small Operation, SQL, Tire Kicker, Warm Lead — each one a recognisable shape from B2B sales playbooks. All six start with 0 criteria: they're empty templates waiting for you to define what they mean for your business.

A prospect profile is, mechanically:

• A name and description for the segment
• A set of criteria, each one a condition on a memory type
• A match mode — either "all criteria" (AND) or "any criterion" (OR)

When a visitor's accumulated memory state matches a profile's criteria, the prospect picks up that profile. One prospect can match multiple profiles simultaneously (e.g. Warm Lead and Small Operation). Profiles are evaluated continuously during the conversation and continue to update after the chat ends as background processing surfaces new information.

The page also has two batch operations:

• Re-evaluate All — runs the profile evaluator over every existing prospect. Useful after you change a profile's criteria and want to refresh classifications.
• New Profile — create a new named profile alongside the seeded ones.

Editing the MQL profile

Click Edit profile on MQL.

The editor opens inline. You can edit the name and description, and below them is the Criteria builder. With zero criteria the profile is a no-op — it'll never match any prospect.

Click Add Criterion to add the first one.

A criterion has two parts:

• Memory Type — which memory you're testing
• Condition — Present, Not Present, Equals, Not Equals, or Contains. For text memory types like our Current Challenge, Present is usually the right choice — you care that the prospect has told you what their challenge is, not what the specific text says.

The screenshot shows two criteria configured for the MQL profile — a custom Current Challenge memory and the built-in Organization Size — both required to be Present. The principle: pick the small set of memories that together indicate the prospect is past the casual-curiosity stage and worth following up on.

After adding the second criterion, the Match: radio appears at the bottom of the criteria section. We leave it on the default All criteria — meaning both criteria must be satisfied for a prospect to be classified as MQL.

Click Save. The editor closes and the profile card now shows 2 criteria.

Cardinality and combinations. You can add as many criteria as the segment requires, and toggle between All (AND) and Any (OR) match modes. For more complex shapes — say, "MQL = articulated challenge AND (mid-market OR enterprise)" — you'd build two separate profiles (one for the mid-market path, one for the enterprise path) and have routing rules consume either. The criteria builder deliberately keeps each profile flat rather than supporting nested boolean logic; it's easier to reason about that way.

Configuring the rest of the seeded profiles

For brevity this chapter only configures MQL. In a real tenant you'd configure several:

• Hot Lead — high intent, e.g. Buying Timeframe present AND Current Challenge present
• SQL — sales-ready, e.g. Email present AND Job Title present AND Current Challenge present
• Tire Kicker — low intent, e.g. Current Challenge not present after multiple turns (use this carefully — it's a hard signal to get right)
• Small Operation — segment-fit indicator, e.g. Organization Size equals a specific small-team value
• Warm Lead — interested but unqualified, e.g. matches Current Challenge present but not Hot Lead's full criteria

The pattern is always the same: name the segment, decide which memory types support that segment, and pick the conditions that distinguish it.

---

How memory and profiles feed downstream

Memory and profiles by themselves don't change agent behaviour — they're the substrate that other configuration consumes. Specifically:

• Skills (Chapter 6) read memory types in two ways. First, they decide which memories to actively pursue during their flow (e.g. the qualify_prospect skill explicitly asks for the memories you've enabled and toggles on). Second, they use the current memory state to decide what to do next on each turn.
• Routing rules (Chapter 7) read prospect profiles. A rule like "suggest book_meeting when prospect matches MQL and has not yet booked a meeting" is exactly the loop we'll close in the next chapter.

This is why this chapter sits where it does in the manual order. Skills and routing both reference what you set up here by name. If you change a memory type or profile later, you may need to revisit dependent skill or routing configuration; the dashboard doesn't currently cascade renames.

---

What's next

The substrate is set. Time to give the agent its repertoire.

Continue to Chapter 6 — Configuring skills.