Two companies in the same industry send AI-drafted proposals to prospects in the same week.
Company A’s proposals are competent, professional, and could have been sent by any firm in the category.
Company B’s proposals reference the specific operational pressure the prospect is under, use the language the prospect used in the discovery call, and make a recommendation calibrated to the prospect’s timeline and team constraints.
The AI that produced Company B’s proposals did not have more intelligence than Company A’s. It had better archetypes.
Customer archetypes in an AI strategy are not the same as buyer personas in a marketing plan. They serve a different purpose and require different content.
A buyer persona describes who to target.
An AI customer archetype describes how the AI should behave when serving someone in that category; what to emphasise, what to avoid, how to frame the offering, and what the person behind the business card actually cares about.
Why buyer personas don’t work for AI: the specific gap
What a buyer persona is built for
A buyer persona is a marketing and sales tool.
It describes who the ideal client is; their role, their industry, their challenges, their goals; in enough detail to help the marketing team create targeted content and the sales team tailor their pitch.
A buyer persona answers: “Who is this person and why do they buy from us?”
An AI archetype answers a different question: “Given that I am now producing an output for this person, how should I behave?”
Why buyer personas fall short
A buyer persona might say: “COO at a manufacturing company; focused on operational efficiency; skeptical of technology for its own sake; values results over process.”
This is accurate and useful for sales positioning. It is not enough for an AI drafting a project status update for that COO.
The AI does not know:
- Does this COO want a brief summary or detailed data?
- Does this COO want options or a recommendation?
- Does this COO react badly to language that implies the project has problems; or do they explicitly want problems surfaced early?
- Does this COO consider a good update one they can act on in under two minutes?
The archetype that answers these questions produces a different status update from the one that does not. The buyer persona does not answer them.
The seven components of an AI customer archetype
Component 1: Role and context (50–75 words)
What it contains: the role, industry, company size range, and operating context of this archetype. Not a demographic profile; an operational snapshot of the person at their desk doing their job.
The specificity test: would this description be accurate for only 20–30% of the company’s clients (an archetype), or for 80% (too generic to be useful)?
Example:
“COO or VP of Operations at a $15M–$25M specialty manufacturer or engineering firm. Responsible for all non-sales operations: production, logistics, finance, HR, and administration. Usually the second-most-senior person in the company, reporting to the founder/CEO. No dedicated CTO or IT function; all technology decisions come through them.”
Component 2: Operational situation (75–100 words)
What it contains: where this archetype currently stands on AI adoption, what they have already tried, and what the operational reality looks like from their desk.
Why this component is critical: the AI output that references what the client has already tried and why it has not worked produces a completely different first impression than one that treats them as a beginner.
Example:
“Has been using Claude or ChatGPT personally for 12–18 months; mostly for drafting and summarisation. Has tried giving the team access once; bought licenses, ran one demo, saw no lasting adoption. Suspects the problem is the team, not the tool. Has not yet considered that the problem might be the absence of context and workflow documentation. Currently in a phase of quiet frustration: AI clearly works; but only for them personally.”
Component 3: Trigger to the conversation (50–75 words)
What it contains: the specific event, pressure, or realisation that caused this archetype to seek help right now; not their general motivation, but the specific catalyst.
Why triggers matter: the proposal that addresses the specific trigger lands. The one that addresses the general motivation feels generic. Two clients with the same role and situation may have arrived via completely different triggers.
Example:
“One of three triggers usually brings this archetype: a competitor win attributed to AI capability; a direct question from the owner or board (‘what’s our AI strategy?’); or a specific painful episode where the manual process visibly failed (a delayed report, a missed follow-up, a proposal that took four days and lost).”
Component 4: Primary concerns (75–100 words)
What it contains: what this archetype is worried about going into the conversation; not their aspirations, but their anxieties.
Why concerns outperform aspirations for AI calibration: outputs that address the concern produce the “they understand my situation” reaction. Outputs that only address aspirations produce the “this sounds good but…” response.
Example:
“Three primary concerns: (1) investing time and money in AI that the team still does not use after three months; (2) buying a system that creates more work than it saves because it requires constant maintenance; (3) having a conversation with the founder about the AI investment and not being able to demonstrate ROI. Secondary concern: being seen as naive about AI by someone who knows more; they want to be treated as someone who has done their homework.”
Component 5: Communication preferences (50–75 words)
What it contains: how this archetype prefers to receive information; format, tone, level of detail, and the specific patterns that make them trust the communicator.
Example:
“Direct and concise. Bullet points for operational detail; prose for recommendations. Does not want options; wants a recommendation with the rationale. Reacts badly to jargon and to any communication that implies more complexity than necessary. Respects bluntness. Has low tolerance for ‘we look forward to discussing’; prefers a specific next step they can confirm or modify.”
Component 6: Success definition (50–75 words)
What it contains: what this archetype would consider an excellent outcome from this specific type of communication; not a general success metric, but the specific thing they would point to when describing a good experience.
For a proposal:
“The proposal makes them feel that the firm actually listened to the discovery call, understood their specific situation, and produced a recommendation that fits their constraints. They want to be able to forward it to the founder and say ‘this is what I’d suggest we do’ without extensive editing.”
For a status update:
“The update tells them exactly where the project is, what (if anything) is behind, and what they need to do; without requiring them to read more than two paragraphs to get the picture.”
Component 7: Relationship sensitivities (50–75 words)
What it contains: the specific things that would make a normally appropriate communication wrong for this archetype; sensitivities about tone, topic, framing, or timing.
Example:
“Sensitive to any implication that they are behind their competitors; this archetype knows the competitive AI gap is real and does not need it emphasised. Sensitive to being talked down to about AI; they know more than most of their peers and have already dismissed three vendors who treated them as beginners. Do not use the word ‘transformation.’”
Building archetypes from the actual client base: the sourcing process
The sourcing principle
Archetypes built from imagination are marketing assets. Archetypes built from the actual client base are AI intelligence.
Imagination-sourced archetypes produce slightly better generic outputs. Experience-sourced archetypes produce outputs that specific clients recognise as having been written for them.
The three-source method
Source 1: The founder or sales lead’s mental model (the starting point)
The founder who has closed twenty clients in this archetype category has a strong intuitive model of what that type of client is like.
The archetype build starts by externalising that model; the founder describes the archetype in response to the seven component questions, as if describing a specific client who is the most representative member of that type.
Time: 30–45 minutes of structured interview, one archetype at a time.
Source 2: CRM and engagement history (the validation layer)
After the initial archetype is drafted, it is checked against the CRM and engagement records of five to ten actual clients in that category.
The check questions:
- Does the trigger description match how these clients actually came to the company?
- Do the primary concerns match what these clients asked about in discovery calls?
- Do the communication preferences match what the most successful client interactions look like?
Discrepancies between the initial archetype and the actual history are the adjustments that make the archetype accurate.
Source 3: Team input (the calibration layer)
The account managers and project managers who interact with clients in this archetype category every week know things about how they behave that the founder may not see.
A brief conversation with two or three team members who work with this archetype produces the specific sensitivities and communication preferences that the founder’s view does not capture.
How archetypes work in practice: the before and after for three output types
Output type 1: Proposal opening section
Without archetypes:
“Thank you for the opportunity to present our proposal for [Company Name]. At [Firm Name], we specialise in AI strategy and implementation for mid-market companies. We believe our approach aligns well with your needs.”
With archetype loaded (COO at manufacturer; trigger: owner pressure; concern: team non-adoption):
“The conversation we had on Tuesday made one thing clear: you’ve seen what AI can do personally; but the team’s adoption attempt last year didn’t produce what you hoped. Before we get to the approach, we want to address that directly; because the reason that attempt didn’t stick is exactly what Phase 1 of our engagement is designed to fix. Here’s how.”
The second version is possible because the archetype told the AI: this client has tried before and it failed; the primary concern is team adoption; and the communication preference is direct and specific.
Output type 2: Client status update
Without archetypes:
“Dear [Client], I wanted to provide you with a brief update on the current status of Project X. We have been making good progress on the deliverables and are tracking well against our planned timeline. Please let me know if you have any questions.”
With archetype loaded (COO; communication preference: concise and direct; success definition: tells them exactly where things stand and what they need to do):
“Week 3 update; Project X. On track. Phase 1 documentation is 90% complete; the two outstanding items are the approval workflow you flagged on Tuesday and the pricing convention document. We need both from you by Thursday to stay on schedule. No issues requiring your attention beyond those two. Next check-in Friday at 2pm.”
The second version is possible because the archetype told the AI: this client wants bullet-point operational detail, wants a specific next step, and considers a good status update one they can act on in under two minutes.
Output type 3: Discovery call follow-up email
Without archetypes:
“It was a pleasure speaking with you today. We enjoyed learning about your business and look forward to exploring how we can support your AI journey. We will follow up with a proposal shortly.”
With archetype loaded (COO; sensitivity: do not imply they are behind; primary concern: being talked down to):
“Good call this morning. A few things we want to confirm before we put together the proposal: (1) your primary constraint is team adoption, not leadership buy-in; (2) the timeline is real; Q4 results matter; (3) you’ve evaluated two vendors already and both were too generic for your industry. We’ll come back with something specific to how a company your size in your industry actually runs, not a framework built for someone else. One question before we do: the ERP situation; are we working with Dynamics as the primary data source or is there something else the AI needs to connect to?”
The third version is possible because the archetype told the AI: this client has already evaluated vendors; resists generic approaches; and communicates best when the next communication demonstrates that the previous conversation was actually heard.
Common questions on customer archetypes for AI
”How is a customer archetype different from an ICP?”
An ideal customer profile (ICP) defines who to target; the firmographic and demographic criteria that make a company a good prospect.
An AI customer archetype describes how to serve a person in that target group; the operational and relational intelligence the AI needs to produce calibrated outputs.
The ICP tells the salesperson who to call. The archetype tells the AI how to behave once the call happens.
”Can I use an existing buyer persona as the starting point for an archetype?”
Yes; as raw material. A buyer persona typically covers components 1, 2, and part of 4 (role, situation, and general concerns).
Components 5, 6, and 7 (communication preferences, success definition, and relationship sensitivities) are almost never in a buyer persona; and they are the components that most directly determine AI output quality.
Start with the buyer persona; then conduct the structured interview to fill in the missing components.
”What if my clients are too diverse to fit into two to four archetypes?”
Every client base with three or more archetypes contains one archetype that represents 50–60% of clients. Start there.
Build that archetype first; deploy it; observe what it handles well and what it misses. The gaps reveal the second archetype. Build two to four archetypes over three to six months rather than trying to build six at once.
More archetypes reduce quality because the AI cannot hold too many simultaneously. The right number is the minimum that covers the distinct communication and situation differences in the actual client base.
”How detailed should each archetype be: one page or multiple?”
One page; 500–700 words covering all seven components. Longer archetypes tend to contain lower-signal content that the AI uses less accurately than the high-signal core.
The quality of each component matters more than the length.
”Should archetypes be updated when a client’s situation changes?”
The archetype is updated when the type changes; which happens slowly. The individual client record is updated when the specific client’s situation changes; which happens frequently.
The archetype is the type template. The individual client record (in the CRM or context pack) is the override. When a specific client differs materially from the archetype, the individual record takes precedence.
”How many archetypes do I need before my AI outputs are noticeably better?”
One. A single well-built archetype for the most common client type produces immediate, visible improvement in outputs for that type. The jump from no archetypes to one archetype is larger than the jump from one to four.
Build the first archetype this week. Deploy it. Observe the improvement. Build the second archetype next month.
Want archetypes built from your actual client base: specific enough to produce the second version of every example in this article?
Customer archetypes are the AI’s intelligence about who the company serves. Without them, every AI output is calibrated to a generic buyer.
With them, every output is calibrated to the specific type of person receiving it; their trigger, their concerns, their communication preferences, and the sensitivities that determine whether the communication builds or undermines trust.
Two to four well-built archetypes, sourced from the actual client base and loaded into the shared AI workspace, produce the output quality difference that founders often attribute to the model when the true cause is the context.
Path one: build your first archetype today. Interview yourself using the seven components above. Write one 500–700 word archetype for the most common client type the company serves. Load it into your Claude Project. Run the proposal opening test. The difference from the first example to the second tells you what the archetype is doing.
Path two: bring in a partner. If you want archetypes built from structured founder interviews, CRM validation, and team calibration; at the depth that produces the second version of every output in this article; that is the Phos AI Labs Phase 1 Foundations work. We have run 400+ AI engagements. Clients include Zapier, Coca-Cola, Medtronic, Dataiku, and American Express. Thirty minutes, no deck. Start here.