Improving Decision Quality in AI-Enabled Product Teams

How a leading global strategy consulting firm used reusable AI workflows to keep evolving decisions consistent across business, product, and delivery teams.

01The situation: Advanced AI was already part of everyday work

A leading global strategy consulting firm had equipped a roughly 50-person organization, working across approximately 10–15 teams, with advanced AI tools to develop and launch products for clients. Individual contributors could take on more work. Engineering and QA teams, in particular, had already developed effective AI-assisted execution practices.

AI adoption was not the problem. The organization had access to capable models and substantial technology resources. Its goals were to improve product quality, meet client delivery commitments more predictably, and launch products capable of succeeding in their markets.

Those outcomes depended on the decisions connecting business leaders, product managers, engineering teams, QA, and client-facing deployment teams. As the work progressed, product direction changed through client demonstrations, market feedback, and senior-partner input. The organization needed each new deliverable to reflect the decisions that still applied—not simply everything that had ever been documented.

02The challenge: AI could read the history, but not reliably identify the current decision

Over roughly 12 months, the project had accumulated requirements, discussions, changes, and approvals across Coda, Jira, spreadsheets, meeting notes, and other working materials. AI tools could access this information. The systems were already connected; adding another connection would not resolve the difficulty.

A decision made earlier in the project could remain in a Coda page or Jira ticket after client feedback or a senior leader had changed the direction. Jira remained effective for engineering execution, but a ticket did not necessarily explain the business reasoning or later decisions affecting that work. The latest direction might be known to a manager or recorded in a recent discussion, while older and potentially conflicting information remained available to AI.

A senior project leader illustrated the operational cost. She maintained an extensive spreadsheet to track workstreams, decisions, and progress. The spreadsheet served a purpose, but aligning it with changing business decisions, Jira activities, and team documentation required repeated manual reconciliation. Other leaders encountered similar issues, and useful AI workflows created by one manager were not necessarily reusable by another.

When different teams worked from different interpretations of what had been decided, even faster individual execution could create additional review, clarification, and rework. The need was especially visible in product management, product launch, and forward-deployed engineering, where business changes had to translate into coherent client-facing work.

03Our approach: Put decisions and human accountability at the center of AI workflows

We did not replace Jira, Coda, or the leadership spreadsheet, nor did we ask people to document every detail more exhaustively. The firm already had capable models and access to its working information. We changed the task the AI was being asked to perform: from searching project history to helping teams establish which decisions applied, why they had changed, who was accountable, and what needed attention before the next deliverable moved forward.

We implemented transferable AI-assisted workflows built around four steps.

1. Establish the current working decision

The workflows examined available project records to reconstruct how requirements and decisions had evolved and distinguish recent direction from earlier assumptions. They helped identify what appeared to be current and what might have been superseded.

Crucially, a newer document did not automatically become an approved decision. When records conflicted or did not establish authority, the workflow could surface that uncertainty for human review rather than silently treating one source as correct.

2. Understand what changed, why, who owned it, and what it affected

A changed requirement could affect product design, downstream engineering tasks, launch plans, or client deployment. The workflows helped reconstruct the decision trail: what changed, when, the rationale available in records, the responsible decision owner or approver where identifiable, and the activities potentially affected.

This made ownership explicit. The person responsible for making or approving a business decision was not necessarily the same person responsible for implementing it. Where the available evidence could not establish responsibility or approval, the gap needed to be visible rather than inferred away.

The purpose of this history was practical: to help people assess the implications of a change and make better subsequent decisions—not to produce an audit log for its own sake.

3. Bring unresolved questions to the right people

Some inconsistencies could be reconciled from existing evidence. Others represented competing priorities, unclear authority, or genuine open decisions. AI helped assemble the relevant history, identify apparent conflicts, and show what downstream work might depend on resolution.

The accountable product or business leader still made the decision. Once clarified, the decision and its rationale could be reflected in later AI-assisted work. AI shortened the work of finding and comparing evidence; it did not replace human judgment or approval.

4. Make effective workflows transferable

Previously, a manager could develop a useful AI process for analyzing Jira activity or reconciling feedback, yet another manager might have to recreate the logic and context. We made the workflows reusable by bringing together the working decisions, supporting assumptions, rationale, identifiable owners, and a consistent process for review and escalation.

That gave participating team members a more consistent starting point. The value was not merely shared prompts: it was a transferable way of using AI to support human decisions without repeatedly rebuilding months of context.

The workflows were communicated across product teams and business leadership. Their use and impact varied by workstream; the engagement did not establish an organization-wide adoption rate.

04What changed: More consistent decisions, less repeated review

During the approximately three-month engagement, the most visible improvements among the involved product teams and senior business leaders were in decision consistency, leadership visibility, and review effectiveness.

Business leaders could better reconcile decisions recorded in their spreadsheets with evolving requirements and downstream work in Jira. Product launch and forward-deployed engineering teams benefited from more consistent interpretation of the decisions shaping their activities. Managers spent less time repeatedly tracing what had changed across documents and discussions, and could focus more attention on the exceptions that required judgment.

These improvements supported higher-quality product work and greater confidence in delivery schedules. Product quality also depended on substantial human review and direct managerial effort; the gains cannot be attributed exclusively to the AI workflows. Nor did the engagement demonstrate a reduction in end-to-end delivery timelines.

The practical improvement was that existing AI investments became more useful to the business and product leaders responsible for maintaining the quality of decisions—not only to engineers and QA specialists accelerating their individual tasks.

Evidence note: outcomes are practitioner observations from the teams directly involved over a roughly three-month engagement. No controlled, organization-wide measurement was conducted.

05What remained unresolved

The new workflows could reconstruct decision histories, identify conflicts, support accountable resolution, and make effective practices easier to reuse. They did not fully automate the ongoing work of maintaining a current, approved project context. Managers still needed to recognize changes, confirm authority, resolve exceptions, and carry decisions into subsequent work.

As AI increased the pace of individual output, that managerial burden remained consequential. The engagement improved the process without eliminating the need for human effort to keep the working context current.

06The executive lesson

For leaders investing in AI, the critical question is not how many documents a model can read. It is whether a team can answer, when work is being produced: What decision governs this work, why does it apply, and who owns it?

This engagement demonstrated that reusable AI workflows could help people answer those questions more consistently without replacing their established tools or their judgment.