Do Not Let Research Pretend to Be Buyer Evidence

Proof note: This template comes from real AI-assisted research and CRM handoff work. The recurring repair was to separate sourced facts from hypotheses and to keep lifecycle authority with the receiving system. All examples are fictional; no real targets, contacts, opportunity states, or outreach assumptions appear here.

AI-assisted research can produce a convincing account brief in minutes.

The company fits the market. A new executive joined. Hiring suggests growth. The product has an obvious integration surface. Public comments hint at a workflow problem. The agent assembles the clues and recommends an opportunity.

Then the brief enters the CRM and something subtle goes wrong.

“Public signals suggest” becomes “buyer pain.”

“Likely owner” becomes “decision-maker.”

“Possible timing” becomes “why now.”

“Worth researching” becomes “qualified.”

No single sentence looks reckless. Together they turn research into buyer evidence without a buyer.

The dangerous handoff is not missing data. It is an inference arriving in the next system dressed as fact.

Research can establish less than it appears to

Public and permissioned research can establish useful things:

• what a company sells;
• who appears to buy it;
• recent public changes;
• visible hiring or product signals;
• stated priorities;
• likely workflow surfaces;
• relevant owners by role;
• known technologies or integrations;
• evidence that makes further investigation worthwhile.

Research usually cannot establish:

• private buyer intent;
• active budget;
• authority to purchase;
• access to the relevant people;
• internal urgency;
• the real workflow bottleneck;
• approval for outreach;
• acceptance of an offer;
• a qualified lifecycle stage.

This does not make research weak. It makes research a different kind of record.

A good company brain lets research be research. It does not force every useful clue to cosplay as pipeline truth.

The receiving system keeps lifecycle authority

A research agent should be allowed to recommend. It should not be allowed to quietly promote its own work.

The CRM—or the accountable revenue owner operating through it—retains authority over lifecycle state:

• accepted for further research;
• queued for review;
• qualified;
• approved for contact;
• active opportunity;
• nurture;
• closed or disqualified.

The handoff should ask for a state change, not declare one.

That small design choice prevents a research system from grading its own homework and updating the official transcript.

Preserve three evidence classes

Every handoff should separate at least three classes.

1. Sourced facts

Claims that can be traced to a source:

• the company launched a named product;
• an executive joined on a public date;
• a role is listed on the careers page;
• the company documents a specific integration;
• a public interview contains a direct statement.

Store the source and date. A fact without provenance becomes an assertion on its second trip through the system.

2. Working hypotheses

Interpretations worth testing:

• the launch may create implementation-handoff pressure;
• the executive may own the affected workflow;
• the hiring pattern may signal a capacity gap;
• a useful first artifact may be a bottleneck map.

A hypothesis should name what would confirm or weaken it.

3. Missing buyer evidence

Facts the research cannot supply:

• whether the problem is active;
• whether the buyer cares now;
• whether budget exists;
• whether the proposed owner has authority;
• whether the account wants contact;
• whether the suggested artifact would help.

Missing evidence is not an empty field to be filled with confidence. It is part of the handoff.

Research-to-CRM Evidence Handoff

Use this template when research is ready for CRM review.

# Research-to-CRM Evidence Handoff

## Identity
- handoff_id:
- account_id or candidate account:
- research owner:
- receiving CRM owner:
- created_at:

## Requested result
- requested CRM action: accept for review / create candidate record / refresh / nurture / reject
- requested lifecycle state:
- why this result is requested:
- acceptance criteria:

## Sourced facts
- fact:
  - source:
  - source_date:
  - retrieved_at:
  - confidence:

## Working hypotheses
- hypothesis:
  - supporting facts:
  - what may be wrong:
  - evidence needed to confirm:
  - evidence needed to disconfirm:

## Missing buyer evidence
- intent known? yes / no
- budget known? yes / no
- authority known? yes / no
- access path verified? yes / no
- private workflow problem confirmed? yes / no
- timing confirmed? yes / no

## Proposed next move
- smallest useful internal action:
- proposed buyer-useful artifact:
- external action required? yes / no
- external action status: prohibited / not requested / draft only / awaiting approval / authorized
- human approval owner:

## CRM acceptance
- decision: accept / accept with changes / hold / reject
- accepted facts:
- hypotheses retained:
- claims rejected:
- lifecycle state assigned:
- next owner:
- next review date:
- decision evidence:

The requested result and CRM acceptance sections make the boundary visible. Research asks. CRM decides.

Match the next action to the evidence

Weak evidence should produce a small internal action.

If research shows only thematic fit, the next action may be to keep watching. If it shows a relevant workflow and likely owner, the next action may be to prepare a fictional sample artifact for internal review. If a human confirms relationship context, the next action may be to ask for guidance—not to launch a sequence.

The evidence should control the size of the move:

| Evidence state | Proportionate next move | |---|---| | Public fit only | Watch, research, or reject | | Sourced workflow signal | Draft an internal hypothesis or sample artifact | | Verified owner, no intent | Ask for human route/context or hold | | Buyer-confirmed problem | Update lifecycle state and define next artifact | | Explicit approval to contact | Prepare or execute only the approved action |

An agent that recommends the same action at every evidence level is not making decisions. It is filling a queue.

Keep external-action status explicit

Research and outreach are often placed too close together.

A brief is completed, a message is generated, and the interface presents Send as the natural next button. The workflow has smuggled authorization into layout.

Put external-action status in the handoff itself:

• prohibited: no external action is allowed;
• not requested: the handoff is for internal state only;
• draft only: an artifact or message may be prepared but not sent;
• awaiting approval: a named human must approve;
• authorized: the specific action, channel, and scope are approved.

A useful artifact is not permission to act. A researched account is not permission to contact.

Let rejection improve the research system

The CRM should not merely accept or reject the handoff. It should return structured feedback.

Useful rejection reasons include:

• insufficient evidence;
• wrong account fit;
• duplicate or stale record;
• lifecycle state overclaimed;
• owner not verified;
• no useful artifact;
• anti-ICP signal;
• timing unsupported;
• external-action request exceeds approval.

That feedback belongs in the research system's next review. Otherwise the research agent keeps producing the same plausible handoff and the CRM keeps cleaning up after it.

Typed handoffs should move learning in both directions without giving both systems authority over the same truth.

One action this week

Open one account brief or AI-generated research packet before it enters the CRM.

Mark every consequential claim as:

1. sourced fact;
2. working hypothesis;
3. missing buyer evidence.

Then replace any declared lifecycle change with a requested result and a named acceptance owner.

If the handoff cannot survive those labels, it is not ready to become operational memory.

For the account-judgment loop before a handoff, read A Useful GTM Brain Judges Account Stories, Not Raw Signals. For current-state evidence fields, use Evidence-Aware CRM Fields for AI Revenue Work. If research, CRM, proposals, and handoffs are sharing conclusions without clear authority, map your revenue bottleneck.