A 2-week manual audit, rebuilt as an AI product clients asked to buy

Every new client engagement at this enterprise digital strategy firm started the same way: 2 weeks, 5 to 8 people, a manual audit that produced no institutional memory and blocked strategy work. The team called it an efficiency problem. Meanwhile a repeatable intelligence product sat inside those project fees, given away for free. Ninety days after the reframe, 4+ enterprise pilots had signed across financial services and manufacturing, and one buyer requested pricing unprompted during a live C-suite demo.

The challenge

• Each audit consumed 2 weeks and 5 to 8 people of opportunity cost across the organization.
• Different teams used different methodologies with no institutional memory.
• Strategy work was bottlenecked by data collection.
• Zero baseline KPIs to measure whether engagements improved client performance.
• Enterprise clients were asking for ongoing intelligence, not one-off reports, but the agency had no recurring revenue model.

The solution

An AI-powered platform with 8 specialized ML agents that automates Day 1 client intelligence delivery.

1. Crawler analysis. Technical infrastructure, site architecture, SEO fundamentals.
2. Keyword intelligence. Third-party integration for competitive keyword positioning.
3. Analytics benchmarking. Proxy metrics for competitive landscape.
4. GA4 deep dive. Optional authenticated client analytics exploration.
5. Content gap analysis. Topic coverage mapping versus the competitive set.
6. UX heuristics evaluation. 140+ criteria assessment with industry-vertical variants.
7. User journey analysis. Flow-based friction point identification.
8. Brand perception. Sentiment and perception intelligence from public signals.

My role

Sole Product Manager owning strategy through execution. The team grew to 8 to 12 contributors spanning AI/ML engineering, front-end development, platform architecture, design, and business leadership.

Strategy and approach

Leadership signed off on treating the platform as a revenue product with its own go-to-market. Two tracks let it earn early while the SaaS infrastructure matured.

Track 1, implementation model (Q1 2026). Sell platform-powered audits as a service through the existing agency model. Revenue starts inside the engagement frame the firm already runs.

Track 2, SaaS model (Q2 2026 and beyond). Open the platform for direct client access with subscriptions, once the underlying security and product infrastructure was ready.

Planning

The Q1 2026 critical path was structured across four priority streams.

Priority 1, security and platform readiness. Auth0 with Okta SSO, user management dashboard, audit logging, cookie consent, magic link deprecation.

Priority 2, multi-run comparison. Historical benchmarks, timeline visualization, before/during/after analysis. This became the number one feature because stakeholders agreed: without comparison, the product would have missed the mark.

Priority 3, agent portfolio completion. Code conversion from Airtable prototypes to production architecture, UX heuristics quality enhancement, structured output standardization.

Priority 4, data architecture. Multi-run data structure, brand-run relationship modelling, export capabilities, BigQuery and GCS storage strategy.

The "not now" backlog kept scope honest: learning agenda standardization, custom agent development, vertical-specific configurations, and client self-service all moved to Q2 and later.

Cross-functional alignment

The team had been working in ad-hoc Kanban with documentation scattered across Jira, FigJam, Airtable, and Confluence. Consolidating to Jira plus Confluence as a single source of truth, with weekly status reporting and a decision log for async alignment, cleaned the operating model.

• Two major strategy sessions with 6 to 8 stakeholders resolved the commercialization approach.
• Weekly coordination with the Product Lead on strategic direction.
• Regular alignment with the Tech Director on architecture feasibility.
• Design team collaboration on component prioritization and design system migration (shadcn/ui foundation).
• Business leader engagement on pricing framework and sales enablement.

Stakeholder trust

Trust came from sequence. The first two weeks went to a full project review across 20+ transcripts and artifacts; the roadmap framework proposal waited until week four, after the context work had earned it a hearing.

The mid-cycle CMS migration was the defining moment. The original prototype stack (no live updates, manual redeployments before every demo) was a liability during a live sales cycle. The team migrated to a production CMS in parallel and phased over. The decision proved right when a client needed last-minute data corrections before a board presentation. The new architecture made it possible in minutes instead of hours.

Before and after

Tooling and reporting

• Product management. Jira (backlog, sprints), Confluence (PRDs, decision logs, security docs).
• Design. Figma (component specs, dashboard UX), FigJam (roadmap visualization).
• Data and analytics. Google Analytics, Tableau (measurement framework).
• AI/ML stack. BigQuery plus GCS (data architecture), third-party SEO and analytics integrations.
• Platform. Production CMS, Next.js plus shadcn/ui (frontend), Auth0 plus Okta (planned auth).
• Communication. Weekly status reports, pre-meeting alignment protocols, decision log in Confluence.

Results

• 4+ enterprise pilots secured across financial services and manufacturing in 90 days.
• One buyer requested pricing unprompted during a live C-suite demo.
• The 2-week manual process now runs as Day 1 automated delivery.
• Dual-track go-to-market defined and aligned across 8 to 12 contributors.
• Platform migrated from prototype to production-grade architecture during an active sales cycle.
• AI agent quality framework with confidence scoring validated by enterprise clients.
• First measurement infrastructure created for a team with zero prior baseline KPIs.
• Multi-run comparison elevated to the number one Q1 2026 priority based on measurement framework data.
• The team moved from ad-hoc Kanban to a structured sprint cadence.

AI agent quality

The ML team and design team were deadlocked for weeks. ML wanted 90%+ confidence thresholds: only surface results the system was near-certain about. Design wanted to show more results at lower confidence: give users a broader view even if some data points were less reliable.

A transparency-first approach resolved the deadlock: show the confidence score alongside every result, let the user decide what threshold matters to them. This avoided the false choice between accuracy and utility. Enterprise clients confirmed it met their trust requirements because the methodology sat in plain view alongside every result.

That decision unblocked the entire agent portfolio and became the template for how quality trade-offs were handled across all 8 agents.

Why this worked

The product insight came before the technical solution. The default response to operational inefficiency is internal tooling. Treating the audit as a revenue product instead set the architecture, the go-to-market, and the timeline on a different course from day one.

Revenue started inside the existing model. Selling platform-powered audits through the agency frame let the product earn its keep before the SaaS infrastructure was mature. Waiting for perfection would have meant zero revenue for two quarters.

Measurement killed the wrong priorities. Before structured KPIs, the roadmap was opinion-driven. Once baselines existed, multi-run comparison rose to number one because clients couldn't justify recurring spend without before/after analysis. The framework also killed a feature with zero client pull, freeing cycles for what mattered.

Deciding what the product is turned out to be harder than building it.