The client
A global B2B software company with about 150 products and 175 solution consultants supporting its sellers. Deals, quotes, and approvals run in Salesforce, and the teams work in Slack.
The engagement
The chief revenue officer sponsored the engagement, and the chief technology officer joined before the pilot. A non-standard deal pulled in people from both teams: security engineers answered questionnaires, and product managers were asked to commit to features. Four workflows were in scope:
- Demos: from request to the demo itself, across about 50 demo environments kept up by 1 person.
- Call preparation: researching the customer before a call, and following up after it.
- Deal desk: discount and terms requests, reviewed for price, terms, revenue, and security.
- Quote build: setting up and pricing each quote, the largest block of effort the pilot measured.
What the pilot found
Two forward-deployed engineers ran a 4-week pilot. They studied the approval records, the sales process, and the approval rules. They also interviewed everyone from sellers to the security team.
- The sales plan assumed sellers spent 5 days a week with customers. They spent 2. The 4 workflows in scope took 46% of the week.
- A deal desk request held about 20 minutes of work, spread over 3 days, 8 owners, and 7 rework loops. Only 12% were cleared by the desk's existing policy.
- Security questionnaires were answered from scratch, though most questions had been answered on earlier deals. Feature requests reached product managers as Slack messages with no deal attached, so nobody could rank them.
Counting every person who touched it, a request took about 4.5 hours, not 20 minutes. The delay cost more: 41% of deals slipped a quarter, and a slipped deal's win rate fell from 18% to 8%.
The redesign
The deal desk was the focus. It now runs on 4 agents with 2 decisions left to people, under the same policy and approvers. The agents gather what a reviewer needs and check that a request is complete before anyone looks at it. Security questionnaires arrive with answers drafted from past approved ones. Feature requests reach the product backlog with the deal attached.
Before20 steps, 8 owners, 7 loops
3 days
After6 steps, 2 human decisions, 1 loop
8 hours
- Manual step
- Agent step
- Person decides
- Rework loop
Process steps
- Before
- 20
- After
- 6
- Change
- 70% fewer
Cycle time
- Before
- 3 days
- After
- 8 hours
- Change
- 89% shorter
Owners per request
- Before
- 8
- After
- 4
- Change
- 50% fewer
Rework loops
- Before
- 7
- After
- 1
- Change
- 86% fewer
Requests cleared by policy
- Before
- 12%
- After
- 68%
- Change
- +56 pts
| Before | After | Change | |
|---|---|---|---|
| Process steps | 20 | 6 | 70% fewer |
| Cycle time | 3 days | 8 hours | 89% shorter |
| Owners per request | 8 | 4 | 50% fewer |
| Rework loops | 7 | 1 | 86% fewer |
| Requests cleared by policy | 12% | 68% | +56 pts |
4 weeks of discovery, then 4 weeks of deployment
- Weeks 1 to 4, discovery: the pilot above maps the process, interviews the team, and sets the baseline for every figure on this page.
- Week 5, build: the agents are built on the redesigned process and the pilot's prototypes.
- Week 6, shadow: the agents work live requests with read-only access. Nothing is sent, and their results are compared with the desk's decisions.
- Week 7, validate: the agents are tested on 3 months of past requests, and the desk and the security team sign off.
- Week 8, deploy: live in Salesforce, Slack, and Jira, with every action logged.
- From week 8, train: the engineers stay on site with the sellers, consultants, desk analysts, and product managers whose work changed.
What the agent does
- When a seller asks for a discount or terms outside policy, the request lands in Salesforce.
- The agent checks the discount, the payment terms, and the contract against the pricing policy.
- If the deal is outside policy, it sends the approver the case in Slack, with similar past deals attached.
- The approver decides. The agent updates the quote to match and records the reason.
Results
Value lost to discounts and concessions fell from 8.6% to 6.1% of contract value, because the approver sees the policy floor and past deals first. Deals slipping a quarter fell from 41% to 29%, and non-standard deals close 11 days sooner. Quotes need 1.4 rounds instead of 3.0. Demos happen in 6 days instead of 18, and call preparation takes 6 minutes instead of 25. Sellers now spend 2.5 days a week with customers, up from 2.
| Source | Annual value | How |
|---|---|---|
| Margin recovered | $11.0M | 2.5 points of value erosion on $440M of contract value, from floor and precedent shown at the point of concession |
| Revenue held in-quarter | $8.4M | 12 points off a 41% slippage rate on $700M of pipeline a year, at the 10 points of win rate a deal loses once it slips a quarter |
| Capacity released | $4.6M | 104,000 hours across the 4 workflows, at the function's $44 fully loaded rate |
| Rework removed | $1.9M | Quote iterations, re-qualification round trips, and re-approvals on cosmetic changes |
| Total | $25.9M | Against the $310M a year it costs to run the go-to-market operation in scope |
What stays with people
An agent works out the discount band, and a named person grants it. Every discount, every contract change, every security answer, and any feature commitment stays with the person the approval rules name. Every action records the policy it applied, and a wrong quote is reversed the same way a person's would be.
