ru/vu

Meet Alder Works.

Alder Works makes industrial equipment that nobody at Alder Works can quite describe at parties. Its VP of Sales last opened the CRM during a fire drill. Its renewal forecast runs on vibes. Its most important spreadsheet is called FINAL_v7_REAL.

None of it is real. Every problem on this page is. Here's how we'd fix four of them.

Talk about your workflow

Alder Works and all its data are invented. These are illustrations, not client case studies.

The account brief.

A renewal call at 2pm, and the answers are in six places. The account owner needs a brief. The facts are spread across CRM, product usage, support, and old call notes.

The information used.

The workflow matches records to the right account, checks what the seller is allowed to see, timestamps every source, and flags gaps instead of filling them with guesses.

Alder Works
Meeting
Renewal planning with the account owner and customer lead.
Account record
Renewal listed for November 30. Contract terms not yet reviewed.
Product activity
Active users down across the last two weekly reports. Cause unknown. Usage extract is three days old.
Support
One unresolved setup ticket. Unclear whether it explains the usage drop.
Suggested questions
Has the rollout plan changed? Who owns the setup issue? What needs fixing before renewal planning?
Missing information
No recent note confirming the customer's renewal priorities.

What stays with the seller.

They check the brief, choose the questions, and own the conversation. The workflow never sends outreach, changes terms, or edits the CRM on its own.

What we'd test.

Right account every time. Every statement traceable to an allowed source. Stale and missing facts visible. Sellers actually find it useful.

Account prioritization.

Which accounts first, with rules anyone can read. Leadership agreed target segments. The account data disagrees with itself in places.

The information used

The workflow applies a versioned set of criteria, scores fit and engagement separately, and sends incomplete records to review instead of inventing a score.

Alder Works
Fit
Matches target industry and size. Region missing, so territory isn't confirmed.
Engagement
Recent product enquiry. Worth a look, not proof of buying intent.
Recommendation
Review after confirming region and owner.
Action
Create a review task. No ownership change, no outreach.

What stays with the team.

Sales and GTM planning own the criteria, routing, and exceptions. Rule changes get reviewed before the next batch runs.

The agent approval pack.

The document that gets your risk team to yes. Alder Works' ops team wants an agent to draft renewal briefs and propose CRM updates. Risk has to approve first. This is what lands on their desk.

Approval pack: renewal brief agent at Alder Works
What it does
Drafts renewal briefs and proposes next-step and close-date updates. The owner approves or rejects every update.
Data it reads
Accounts, usage, tickets, and meeting notes the owner can already see. No contracts or billing.
Allowed actions
Draft a brief and propose two CRM field updates. Not allowed: send email, change pricing or terms, edit contracts, update records without approval.
Evaluation results (fictional)
120 questions with checked answers. 112 answered correctly from an allowed source, 6 correctly declined for missing or stale data, 2 wrong. Both traced to an outdated usage extract, fixed with a freshness rule, and retested.
Guardrails
PII masked in drafts. Policy checks block commitments on price or terms. Every tool call checked against the allowed-action list.
Escalation
A cancellation signal stops drafting and routes the account to its owner with context.
Audit trail
Every input, tool call, output, and approval logged with model, prompt, and data versions, so any past brief can be reproduced.

What stays with the reviewers.

Approve, approve with conditions, or send it back. Autonomy widens task by task as results support it.

What we'd test

Are refusals correct when data is missing? Do guardrails hold against adversarial inputs? Can a reviewer reproduce any past output from the log? Does performance hold after launch on live samples?

The renewal risk review.

The health score nobody trusted, replaced. Alder Works' CSMs ignore their weighted-average health score, and they're right to. A model trained on the company's own renewal history flags accounts for a weekly review.

Alder Works
Renewal
In 60 days.
Model output (fictional)
0.42 churn probability, up from 0.18 a month ago. Calibrated, so accounts near 0.4 have churned about 40% of the time in this fictional dataset.
What moved it
Weekly active users down six weeks running. Champion changed roles. Two tickets open more than ten days.
Suggested action
CSM calls this week to meet the new stakeholder and confirm the rollout plan. Escalate the open tickets.
What the agent won't do
Contact the customer, offer discounts, or change the forecast.

What stays with the CSM.

Whether to act, what to say, and whether to escalate. The decision gets logged so the model and the review keep improving.

What we'd test

How many days of warning does it give? Is it calibrated on accounts it hasn't seen? What share of flags do CSMs act on? Does retention improve against a control group where practical?

Got your own Alder Works?

Tell us which decision needs to get better and what data you have. We promise not to put your company on this page.

Talk with us