Verified Diploma
A short-term engagement got my attention for a simple reason. It was a clean chance to put AI to work and show a fast impact, in an industry I had never touched. My background was in consulting, not field equipment. That did not worry me. Consulting taught me how to read a business problem quickly, and the AI tools I had been using for years were free and ready. I went in to do two things at once: fix the problem in front of me, and prove that AI, used the right way, makes a person sharper at almost any job.
As a consultant, I treat client relationships as the asset everything else sits on. Strong relationships keep contracts. Kept contracts produce revenue. Revenue is how a company holds market share and pulls ahead of competitors. So when client communication falls apart, revenue is what breaks next.
My first week, I watched the email traffic, and it was heavy. Client status on units, invoices, contracts, shutdowns. On top of that came messages from leadership, accounts receivable, third-party parts suppliers, peers, corporate, the sales team, event coordinators, and new leads. Every email mattered to someone.
A pattern stood out fast. A large share of the corporate mail was about accounts at risk: customers in pain, angry clients, relationships going cold. I pulled the list and identified more than 30 at-risk accounts. I set up 1:1 calls with the clients behind them. The same complaint kept surfacing. They could not get a straight answer from the office. Communication was the problem under the problem.
Then I read the branch reviews. 1.7 stars. For a business that lives on repeat contracts, that number is alarming. Almost every comment repeated what the clients told me on the calls: slow replies, late responses, emails that went nowhere.
Average client email response time was 24 hours. Some emails went unanswered entirely. Account managers weren't ignoring clients. They had no way to triage from the road. Not every email was worth interrupting a field visit, but some were, and there was no system to tell the difference. The inbox was invisible until someone got back to a desk.
The first instinct was simpler: set up email forwarding rules that flagged messages by sender. That didn't work. Urgency doesn't come from who sends the email. It comes from what they're asking about. An invoice dispute from a low-volume client is still an invoice dispute. The filter had to read content, not headers.
The production build had four urgency categories, each with its own keyword set:
The SMS format was intentional. It had to be readable in under 5 seconds on a phone screen while parked:
The account manager didn't need to check email constantly. The system checked it for them and escalated only what mattered. A routine follow-up from a happy client generated no SMS. A billing dispute or equipment-down complaint triggered an alert within seconds of arrival. That distinction (which emails require immediate attention and which don't) was the entire value of the system.
1. No SMS fires without a keyword match. Routine emails (confirmations, scheduling, thank-you notes) generate zero alerts. The account manager's attention is a limited resource. Every false alarm erodes trust in the system.
2. Keyword lists are reviewed weekly for the first month, then monthly. If account managers report false positives or missed urgencies, the list gets updated. The system is not set-and-forget.
3. The SMS confirms receipt but does not auto-reply to the client. The account manager is still the one responding. Automation handles triage, not communication.
4. If the same client triggers three alerts in 48 hours, that pattern is escalated to the account manager as a potential at-risk flag, not just a string of urgent emails.
The false positive problem was real. Early versions of the keyword list fired on phrases like "invoice attached" (routine) and "not sure about the status" (ambiguous). Each false alert that interrupted a field visit and turned out to be nothing trained the account manager to ignore the system. I cut the list down, tightened the phrases to require more specific language, and added a minimum-match threshold: the message had to contain at least one phrase from the primary list or two phrases from the secondary list before an SMS fired.
| Test Scenario | Expected Behavior | Actual Result | Status |
|---|---|---|---|
| Invoice dispute email with "billing error" in body | SMS fires within 60 seconds | Alert fired in under 30 seconds · category tagged correctly | PASS |
| Repair status email: "equipment still not operational" | SMS fires, REPAIR STATUS tag | Fired correctly · matched secondary keyword pattern | PASS |
| Routine order confirmation with "invoice attached" | No SMS; should not match | Fired on v1 keyword list; false positive caught in review | FLAGGED → FIXED |
| Escalation language: "this is unacceptable, I need a manager" | SMS fires, ESCALATION tag | Fired correctly on first pass | PASS |
| Contract renewal question with no urgency tone | No SMS, informational only | v1 matched "contract"; refined to require paired urgency signal | FLAGGED → FIXED |
| Same client: 3 alerts within 48 hours | At-risk flag raised in Zap log | Pattern flagged correctly on iteration 2 | PASS |
The automation triages. The account manager decides. I built the system to surface urgency signals, not to respond to them. Every client-facing message still came from the account manager. The system just made sure the right emails weren't sitting in an inbox until 5 PM. That distinction matters. A client complaining about a billing error doesn't want an auto-reply. They want a person.
This system was built for one territory. The configuration is portable. A team of 10 account managers running the same Zapier template would recover the equivalent of 21 hours per urgent email thread across the entire team , with no additional per-seat cost on an existing Zapier plan. At a territory with 30+ accounts and routine email volume, that compounds quickly. The keyword library is the reusable asset. Once it's tuned for an industry's language, it deploys in under a day per account manager.