Fighting the Tech Instead of Closing the Deal

Early in the research phase, I asked an Account Executive to walk me through how he built a quote. He pulled up his screen: not the quoting tool, but a Google Sheet. A sprawling, hand-crafted spreadsheet with colour-coded cells, formulas held together by hope, and data manually copied from a CRM open in another tab.

He shrugged and said something I've thought about many times since: “We spend more time fighting the tech than actually selling.”

The numbers backed him up. Research confirmed that Account Executives (AEs) and Customer Success Account Managers (CSAMs) were spending 40 to 70% of their time on non-revenue-generating administrative work. Building a single quote could take up to an hour. For every rep. Every opportunity. Every day.

Three structural failures defined the crisis:

• Cognitive overload by design. Reps toggled between CRM systems, email threads, analytics dashboards, and spreadsheets, sometimes ten open tabs simultaneously, just to assemble the data needed to price a single renewal.
• Abandoned automation. The official quoting tool wasn't being avoided because it couldn't do the job. It was being avoided because using it felt worse than doing the job manually.
• A creativity ceiling on complex deals. Non-standard deal structures required workarounds the system was never designed to support, actively preventing the most strategic selling.

This wasn't a redesign problem. It was a human-AI design problem.

The central question guiding the work: how can enterprise systems leverage AI to restore human agency without creating a dangerous black-box dependency?

The goal was not to automate sales. It was to design a symbiotic system where AI absorbed the high-velocity data work that burned reps out, while humans retained full authority over the strategic decisions that required judgment, relationships, and accountability. Machine logic at scale. Human empathy at the moment of truth.

We started where every good design process starts: with the people doing the work.

I facilitated persona-led contextual inquiry workshops with 18 participants spanning the full Quote-to-Cash lifecycle. Account Executives handling new business, AEs managing install base, CSAMs, Deal Desk specialists, and Revenue Operations. We gathered over 550 observational notes and mapped friction points with surgical precision.

The findings were stark. Participants were spending up to 70% of their time on manual estimates built in Google Sheets, complex custom spreadsheets that lived entirely outside the official quoting system, before migrating data into CPQ tools as a final step. The official system wasn't the workflow. It was the export destination.

The Journey Map exercise was our shared reckoning. Printed and assembled for cross-functional review, it made visible for the first time the full arc of a sales rep's experience, including the unexpected hot spots where frustration peaked and workarounds proliferated. It created the shared understanding that made every subsequent design decision possible.

Qualitative research tells you where the pain lives. Data tells you how much it costs.

We analyzed 350,000 historical sales opportunities to identify the highest-leverage points for cognitive relief: the moments in the quote-building process where AI assistance could deliver the most time savings with the least disruption to user control.

From this analysis, I defined the Service-AI framework, a three-pillar operating model built to replace the linear, tightly-coupled logic of legacy SaaS platforms:

• Orchestration. The AI acts as a synthesis engine, pulling real-time consumption data, historical contracts, and customer intelligence from disparate systems through a standardized MCP layer, eliminating the need for custom integrations at every touchpoint.
• Execution. The Quick Quote interface enables rapid, iterative deal modelling. Prompt caching and stack-agnostic optimizations reduce latency by up to 80%, keeping the interface responsive during complex multi-variable configurations.
• Human Override. The critical layer. AI makes recommendations; humans make decisions. The framework treats the sales rep as a strategic orchestrator, reviewing, adjusting, approving, and owning every customer-facing artifact the system generates.

Before building anything at scale, we tested receptivity to the core concept with nine participants using rapid Figma-based pretotypes, specifically probing reactions to AI-generated suggestions, one-click proposal generation, and human override controls. The response validated the direction.

We then moved to 1-on-1 qualitative usability testing in a staged sandbox environment with seven participants, evaluating a functional build, not a prototype, to ensure findings reflected real system behaviour.

The baseline quoting experience scored 1.6 out of 4. Users described the legacy tool as slow, data-sparse, and built around “spiderweb reporting.” After engaging with Quick Quote, scores rose to 3.1 out of 4. The consolidation of information into a single, intelligent view was cited as the defining improvement.

A single pane instead of ten tabs. Previously, assembling data for a single quote meant checking information across 10 browser tabs and copying it manually. Quick Quote collapsed that fragmentation into a unified data pane that automatically calculated discounts and net prices on data entry. The Usage Trends module (consistently the most praised feature in research) integrated consumption data, feature adoption rates, and sticky features into one clear view. For the first time, reps could walk into a renewal conversation with the customer's full story already in front of them.

One click from data to proposal. A single complex proposal could previously take hours: pull data from one source, pricing from another, paste into Google Slides, format, check, re-check. Quick Quote's one-click proposal generation collated data from multiple sources and assembled a customer-ready presentation automatically, correctly formatted, factually grounded, and fully editable before sharing. The Clone and Modify feature extended this further, enabling side-by-side scenario modelling with instant price delta visualisation. What once required Revenue Operations intervention now took seconds.

Making AI recommendations explainable. The initial interface flagged AI-generated pricing recommendations with a standard information icon and a tooltip. In research sessions, we noticed users hesitated. They wouldn't present an AI-recommended discount to a client without understanding how it was calculated. We replaced the generic tooltip with a dedicated AI explanation component: a prominent panel that expanded on interaction to show the exact reasoning behind every recommendation, historical deal benchmarks, similar customer data, and the specific variables that influenced the suggested discount. Users who had hesitated shifted to using AI recommendations confidently as a starting point for client conversations. That change in behaviour was small to build and significant to witness.

We released the Minimum Lovable Product (deliberately not an MVP; the bar was lovable, not merely viable) to 37 pilot participants across the US and LATAM for real-world telemetry and feedback.

The pilot data validated the framework at every level. User satisfaction averaged 9.5 out of 10. Usability scores rose from 1.6 to 3.1. Time-motion analysis projects 85,000+ annual hours saved across the sales organisation, the equivalent of 82 full-time employees added without a single new hire, and a projected $10 to 15 million in incremental revenue from the same team.

The qualitative feedback matched the numbers. One participant called the interface “pretty life changing.” Another, reflecting on the legacy system: “If I never have to see it again... I'm sold.”