Benish J Gabriel

Low-quality deals drove a quarter of all customer contacts and millions in refunds, but Sales had no ranked, evidence-backed list of which deals to pause or fix, and four teams measured complaints four different ways.

• Built a 97,778-deal database across 9 sources and had AI read the customer conversations behind 13,887 deals to explain each risk.
• Designed the Revenue-per-Defect risk score and took the report to the CEO in 7 days through ~230 change requests.
• Ran three pause and derank waves covering 2,224 deals, 2,221 of them backed by a customer complaint narrative.
• Measured results weekly with an impact dashboard, a Save & Fix audit and an unpause audit, and unified the complaint-rate metric (CDPU) into one governed company definition.

Team Leaders manually audited 2 conversations per agent per week, while each agent handles ~80 conversations a day (~400 a week), so 99.5% of conversations went unchecked and errors and policy breaches were missed.

• Created 7 SaltedCX AI models that track specific agent errors on every conversation (rudeness, miscommunication, idle procedure, tone, repetition, over-apologising, macro placeholders), plus 7 placeholder models after finding 187 placeholders in 1,000+ macros.
• Created the Auto QA prompts that score every conversation on agents' soft skills, and tested each version on real chats with a Prompt Analyzer and a 9-tab Prompt Impact report.
• Built the Soft Skills dashboard where every agent sees chats audited by AI, the share scoring well, weekly progress, scores by pillar and comments on exactly what to improve.
• Launched a Resolution First QA framework (4 categories, 34 metrics) that has audited 5,800+ conversations.

Groupon could not see where support agents spent their time inside Salesforce, what each step cost per year, or which work should be automated.

• Ran the Salesforce time-and-motion study two years in a row, comparing activities, cost drivers and new or retired steps year on year.
• Watched 44 screen recordings and time-stamped 2,450 agent actions across 60 cases, condensed into 30 core activities.
• Built the annual dollar cost and SPH model for every activity and presented the case for automation to leadership.
• Built a demo chatbot that handled ~50% of common support conversations, proving what a bot could take over.

Some merchants cheat customers by hiding fees, pushing payment off Groupon or refusing vouchers. They hurt customers and Groupon's trust, drive complaints far out of proportion to their sales, and nothing caught them consistently; paused deals even kept selling.

• Built a detection system that scores every merchant on 90 days of orders, refunds and customer chats to find the ones cheating customers.
• Had AI read 61,900+ customer chats and 6,750 deals' fine print, cleared 5,277 false flags, and ranked merchants into Pause Now and Pause Review.
• Turned every finding into Salesforce cases in English, French, German and Spanish (11,000+ built and released in waves), with translated evidence so reps act in their own language.
• Closed the gap that let paused deals keep selling, and added a profitability model across 42,558 merchants in 14 countries to decide fix or pause.

Sales reps each own hundreds of merchants and were managed on refund rate, which shows money lost but not why, so they could not tell which merchants were hurting customers and their own incentive, or what to ask them to fix.

• Built HawkEye, a connected application that gives every rep, every Monday, the merchants to call that week, the issue customers report, real customer quotes and the right time to call.
• Scans 93,000+ deals and 50,000+ merchant accounts every week, flagging every deal running worse than its category.
• Reps log call notes on a Salesforce case, and any change needed is routed to the right team: Content Ops, Merchant Operations or the deal pause team.
• Automated the full Monday run end to end: data pull, AI analysis, rebuild, publishing and failure alerts, and opened the same view to Customer Support.

Internal Groupon terms such as "Fine Print", "Trade In", "Cooling Off Period" and "Security Code" had spread into everything customers saw (agent replies, macros, deal pages, internal tools and the customer FAQ), confusing customers, and nobody was measuring it.

• Found the most-used jargon across agent chats, macros, deal pages, internal tools and the customer FAQ, and set a plain-language replacement for each term.
• Audited 786 macros and rewrote 169 into jargon-free templates that agents now use.
• Built 10 SaltedCX AI models to track jargon, with dashboards for Service Delivery, Team Leaders and Area Managers.
• Trained Team Leaders and Area Managers on what to say and what to avoid with a Google Slides deck, FAQ and video walkthrough, after reviewing 500+ new-hire slides.

Customers were abandoning purchases and leaving low ratings over pending orders, fraud-screening rejections and broken gifting flows, while agents had no clear process or working macros to help them.

• Audited 655 July dissatisfied reviews against Salesforce, transcripts and Cyclops, tracing 20% to pending orders and mapping the INTL and NAM pending-order flow.
• Redesigned the Signifyd fraud-rejection process with a charter, new queues, pathways, macros and 3 Salesforce reports.
• Built the first customer-facing Zingtree path for Signifyd rejections so customers could self-serve before contacting support.
• Ran a gifting root-cause analysis and refreshed gifting snippets, macros and Zingtree nodes.

Misleading or incomplete deal pages caused avoidable complaints, but nobody could connect what a page said to what customers complained about, across every line of business.

• Scanned every live page across Local (108 categories, 70.6K deals), Health, Beauty & Wellness (35, 64.5K), Goods (105, 21.6K), Getaways (3, 7.3K) and Live (15, 1.3K).
• Graded each page on 3 issues (hidden costs, hidden restrictions, redemption friction) from prominent to silent, plus 7 category rules such as spa pages with no tip policy.
• Classified 345K customer chats, NPS responses and reviews against 24 complaint patterns with a free local text model at ~85% precision, with no paid AI.
• Rolled risk up by line of business, category and PDS, with customer quotes and a fix for each deal type.

Team Leaders had no consistent way to choose which agents to coach each week or what to coach them on, and the Monday report behind it took 6 manual steps.

• Built a fully automated pipeline that pulls the weekly QA export from Gmail, scores every agent on Empathy and Conciseness, and ranks the bottom 5 per Team Leader.
• Detects each agent's strongest communication problem out of 6 patterns (stalling, information splitting, delayed resolution, passive language, repeated apologies, excessive reassurance), with real quotes from their chats.
• Writes coaching guidance on why it matters and what to say instead, grounded in the Resolution-First principle.
• Creates every Team Leader's Asana coaching task with links to each agent's 10 lowest-scoring conversations, with duplicate checks built in.

Customer Support and Merchant Operations reporting in SaltedCX was fragmented, partly miscalculated, undocumented and unowned, and quality checks ran by hand, so leaders could not trust QA, CSAT, adherence or productivity numbers.

• Built 30+ dashboards at three levels across CS and MO: each agent sees their own performance, each Team Leader sees their team, and Area Managers see every team.
• Covered CSAT, QA, soft skills, schedule adherence, SPH, escalations, contact reasons and bot vs agent, with 100+ custom MAQL metrics including WFM-aligned Solved per Hour, reopen rate and a weighted agent stack ranking.
• Took QA from manual to automated: Auto QA prompts, QA forms with disqualifier logic, and a two-level dispute process with SOPs.
• Audited permissions for 2,220 users and trained Team Leaders, Area Managers and 20 Fraud team members to build their own views.

Customers faced payment and checkout failures, while inconsistent agent guidance and fragmented reporting limited visibility into customer pain points and payment related contact drivers.

• Analyzed 655 dissatisfied customer ratings, identifying checkout as the driver of 46% of dissatisfaction.
• Built the first consolidated checkout dashboard and introduced 9 SaltedCX tracking tags.
• Developed payment training adopted by the global training team, reaching 400+ agents.
• Coordinated the Germany PayPal rollout across 6 customer and agent facing resources in one week.

CEO and executive escalations were resolved one case at a time, so the agent, policy, tool and product gaps behind them kept creating new escalations.

• Root-caused 17+ escalated cases and grouped them into recurring issues: pricing, promo codes, merchant-charged tax, guest checkout and refund loopholes.
• Built SaltedCX detection models for pricing discrepancies, guest-checkout failures and merchant-charged tax, so each issue is tracked across every conversation.
• Took each root cause to its owner: Product, Merchant Operations, policy and agent training.
• Checked 1099-K escalation policy against IRS guidance, correcting an outdated $20K threshold to $5K.

Promo-code contacts were tracked by label only, so the Promo Code team did not know which codes were failing, why customers struggled, or what to fix first.

• Built a SaltedCX detection model that finds promo-code contacts the labels miss.
• Built an automation that reads every flagged conversation each week and groups it by real cause and by code.
• Delivered a weekly report to the Promo Code team: which codes fail, why, and the fix for each.
• Packaged it as a reusable skill that anyone can run for any date range.

A small Indian clothing business needed an online store and a way to run products, stock and orders from a phone, without paying for software.

• Designed and built a mobile-first storefront with a phone-friendly admin panel for products, stock and orders.
• Moved pricing and stock updates to the server so prices and stock cannot be changed from the browser.
• Integrated Razorpay online payments and cash on delivery, with verified payment confirmations.
• Set up instant Telegram order alerts for the owner and hosted everything on free tiers.