By KYN AI Advisory Team — AI implementation specialists, Singapore
Marketing agents miss legal and procurement blockers because their inputs never include those people. Scoring models read engagement, title, company size and intent. None of those fields shows that a contract is stuck in legal review or that a vendor form is waiting in a procurement portal. The fix is not a better threshold. It is new signal types, new CRM fields and a rule that flags a deal instead of advancing it when a required stakeholder is missing.
Why does lead scoring stop at the champion?
Most scoring, whether it sits in a spreadsheet or inside an AI agent, answers one question: does this person want the product? It checks who opened the email, who booked the demo, how senior the title is and how large the company is.
It does not ask whether that person can get a deal through legal, past a security questionnaire or through a procurement approval. Those steps happen in inboxes, portals and shared drives. They rarely reach a default CRM lead or opportunity field.
So a deal can score "hot" until the day it dies in a legal queue. The model never had a variable for legal. This is a gap in what the model can see, not a miscalibrated cutoff.
Assumption: this pattern is common in B2B deals with regulated or enterprise buyers. It is not a measured rate. Test it against your own closed-lost history before you act on it.
Who sits in the approval chain?
A buying decision is usually ratified by several people. A marketing tool typically sees only the first.
| Role | What they do | Visible to a scoring model? | |---|---|---| | Champion | Engages with content, books the demo | Yes | | Economic buyer | Approves budget | Sometimes, through title | | Legal or counsel | Reviews the contract and data processing terms | No | | Procurement | Runs vendor onboarding and RFP rules | No | | Security or IT | Issues the security questionnaire | No |
In Singapore the data-protection side of legal review often references the PDPA, which sets obligations for organisations that handle personal data. A buyer's counsel will ask how a vendor treats that data. A vendor who cannot answer stalls.
Which signals show a blocker is forming?
Each of these leaves a trace. None sits in a default lead field.
- A buyer sends a data processing agreement or a redline.
- A security questionnaire arrives from the buyer's IT team.
- The buyer asks you to register in a vendor portal.
- An RFP or tender names your category.
- A legal or procurement contact appears on the account for the first time.
Treat each as a stage signal, not as noise. A first legal contact on a deal at "demo done" is a fact about the next four weeks. A model that ignores it will keep predicting a close date that no longer holds.
How should the scoring logic change?
Borrow the rule used in reconciliation work: a false match costs more than a missed match. In bank reconciliation, an item that does not match inside a tight tolerance goes to a person with the reason attached. It is not auto-resolved.
Apply that to qualification:
- Treat "champion engaged" and "deal will close" as two separate claims.
- When a required stakeholder signal is absent, flag the deal. Do not auto-advance it to SQL.
- Show what could not be resolved and why. "No legal contact on a deal over the contract-review threshold" is useful. A silent score of 82 is not.
- Build the rule from your own history. Pull the last 30 closed-won and closed-lost deals and mark which ones had a legal or procurement step, and when it started. Write the rule from that pattern, not from a general idea of how enterprise buying works.
The same discipline applies to the data underneath. Before any score runs, the account record must hold the buying structure: who the economic buyer is, whether legal review is a standing step for vendors of this type, and whether a questionnaire is a known gate in this segment. If those fields are blank, every score built on them looks precise and is not accurate.
What does the build sequence look like?
- Define the inputs. For each segment, decide which stakeholder signals matter. Write them down before writing any scoring rule.
- Add CRM fields. "Legal contact", "procurement stage", "security review status" and "DPA requested" each need a defined field with a fixed format. An agent cannot score against a free-text note.
- Log in real time. Replies and outreach steps should write back to the CRM when they happen. If a legal request sits in an inbox until someone re-reads the thread, the agent sees it too late.
- Add the handoff rule. Marketing hands to sales only when required stakeholder fields are filled or explicitly marked "not applicable" by a named person.
- Test on messy history. Run the logic on deals that looked hot, sat in "legal review" for weeks and died. Those deals show whether the model sees a blocker or only describes engagement that was never going to convert.
For the sales side of the same problem, see how sales agents lose deal context in handoffs and why discount limits belong in the CRM. Marketing agents are covered at /marketing.
What can an agent not do here?
- It cannot read a buyer's internal procurement system unless the buyer sends you something.
- It cannot judge whether a redline is reasonable. A person in your business does that.
- It cannot guess a legal contact. It can only record one that someone has named.
Keep the agent to detecting, recording and flagging. A named person decides what to do about the flag.
What does a blocker flag look like in practice?
Illustrative example. The account, dates and fields are made up.
A mid-sized company has a deal at "Demo done". Marketing's score for the contact is 84. The CRM holds:
| Field | Value | |---|---| | Champion | Head of operations, engaged, booked the demo | | Economic buyer | Finance director, named, not yet contacted | | Legal contact | Blank | | Procurement stage | Blank | | Security review | Questionnaire received 4 days ago | | DPA requested | Yes |
Without the stakeholder fields, the score says 84 and the deal advances to sales-qualified. With them, the rule reads: security questionnaire received and DPA requested, legal contact blank. The agent flags the deal instead of advancing it, with the message "Security questionnaire and DPA request received. No legal or procurement contact on record. Deal not advanced." A named person in sales operations assigns the questionnaire, asks the champion who in legal will review, and records the answer.
The deal is no less hot. It now has a visible set of next steps, and the close date moves to match.
How do you build the stakeholder fields without burdening reps?
Reps resist extra fields. Reduce the load in three ways.
- Capture from the source. Detect a questionnaire, a DPA or a vendor-portal email in the mailbox, and fill the field with a link to the message. The rep confirms.
- Use short pick lists. "Legal review: not required, requested, in progress, complete". One click.
- Ask only where it matters. Require the fields for deals above a size or in a segment where legal review is standard. Smaller deals skip them.
Each field has an owner, a definition and a rule for who may change it. Free-text notes do not count, because a rule cannot read them reliably.
What do you do with deals already in the pipeline?
Run a one-off review rather than waiting for new data.
- Export open deals above your size threshold.
- Ask each rep to fill the four fields from what they know. Mark unknown as unknown.
- Flag any deal with a received questionnaire or DPA and no named legal contact.
- Sort the flagged deals by value and date.
- Review the top ten with sales, and agree a next step for each.
Unknown is a valid answer. A deal with unknowns in three fields tells you where to look. A deal with confident but wrong fields hides the same risk.
Why not just ask the champion?
You should. The champion usually knows the process. But the champion may not know the timeline, and may not own it. The field work above gives the conversation a structure: "Who in legal will review the contract, and how long does that usually take?" The answer goes into the record, and the next person to open the account sees it.
Do not rely on one person's memory. When that person is on leave, the deal stalls with nothing to show where it stands.
How do you test the rule on history?
Use your own closed deals. Take the last 30 won and 30 lost deals. For each, mark whether legal, procurement or security steps appeared, and the date each appeared. Then answer:
- In how many lost deals did a blocker appear before the loss?
- How early could the rule have flagged it?
- How many won deals would the rule have flagged unnecessarily?
The third question protects the sales team from noise. A rule that flags every deal is ignored. Adjust the segments or thresholds until the flags are useful, and keep the sample so you can test future changes.
What does marketing do differently after this?
Marketing's role shifts from handing over a hot name to handing over a mapped account. Content can support the blockers too: a one-page security overview, a data protection summary and a vendor onboarding checklist answer the questions before they stall a deal. The agent can note which of these were sent and when.
Outbound messages wait for a named person. The agent suggests which document fits the flag. A person sends it.
What are the limits?
- The rule sees only what is recorded. A legal review that happens by phone leaves no trace unless someone logs it.
- Small buyers rarely have a legal step, so the rule should not apply to them.
- Different regions have different data rules. Confirm the requirements for each market with a qualified adviser.
- The first version will have gaps. Review the flags monthly and adjust.
What should you do this month?
Pick one segment where legal or security review is common. Add the four fields, run the history test on its last 30 deals and write one flag rule. Review the flags after four weeks with the person who owns the pipeline, and extend to the next segment only when the flags are being used.
Where KYN fits
KYN builds marketing and sales agents inside a system that runs in your own cloud account. For an insurance brokerage, we built outreach with one checkable first line, a reply desk that sorts each reply, and a log of every step to the CRM. Every action is logged with the actor. Outbound messages wait for a named person.
The first 14 days are a review of your stack and interviews with your team. A full build ships in under 90 days. Rights in the custom code and configuration pass to you on full payment. See security for how accounts and agents are restricted.
Frequently asked questions
Can an MQL score include legal and procurement risk?
Yes, if the signals exist as fields. Add DPA requests, security questionnaires and vendor-portal registrations as inputs, then set a rule that flags a deal when a required stakeholder is missing. Without those fields, the score cannot reflect them.
Which team should own the blocker flag?
Sales operations, in most SMEs. Marketing sets the signal definitions, but the flag changes how a deal is handled, so the person who owns the pipeline owns the flag. Name that person in the rule.
How do I check whether this is a problem for my business?
Take your last 30 lost deals. Count how many stalled at contract, security review or vendor onboarding. If more than a few did, the gap is real for you and the fields above are worth adding.
Checked 30 October 2026.
Curious whether Marketing agents fits your business? Talk to KYN on WhatsApp — no forms, just a conversation.
