AI consulting for professional services: how an 800-person firm built a governed AI roadmap in 14 weeks

Overview

An international professional services firm with roughly 800 employees across 14 countries. When Phos AI Labs came in, the firm was not short on AI activity. People across the organization had started using it on their own. Some were running real work through personal ChatGPT and Claude accounts. One engineer had already shipped an internal financial tool without any security or architecture review. A few departments had published their own apps on platforms nobody at the company officially managed.

The COO could see what was happening. The energy was real. The problem was that everything being built lived in a different place, followed different rules, and depended on whoever had built it. There was no shared foundation underneath any of it.

There was a second problem too, and this one was harder to fix. The COO and her Head of AI were the only two people in the organization who understood both the business and the technology well enough to make good calls on what to build. Every project touched them. Every decision waited for them. They had become the bottleneck inside their own AI strategy.

She came to Phos AI Labs with one question: what do we build first, and in what order, before this gets away from us?

That is the question AI consulting is designed to answer.

The problem

The firm had AI activity everywhere. What it did not have was a foundation underneath any of it.

The energy was genuine. People were solving real problems with AI every day. But three things were quietly building underneath all of that activity.

Nobody owned what was being built.

Tools lived in personal accounts, personal environments, personal workspaces. There were no shared standards for how something got built, who was responsible for it, or what happened to it when the person who created it moved on.

When someone left, the work left with them. The context behind it; why it was designed that way, what data it touched, who had approved it; gone too. The organization had no way to see everything in flight, no way to know what it actually owned, and no way to protect what had already been built.

Requirements were verbal. Builds grew beyond anyone’s original intention.

When a business user had an idea, they went directly to an engineer. The engineer built what they heard. What they heard was sometimes incomplete, sometimes conflicting, and sometimes based on a process that had already changed by the time the build was finished.

There was no written requirement before engineering started. No stage gate between idea and execution. No one asked whether the process being automated was still the right one.

The two people who could make good decisions were already at capacity.

The COO and her Head of AI were the only two people in the organization who understood both the business and the technology well enough to make sound calls. Every build touched them. Every governance question landed on their desks.

The organization had no way to move faster than two people could review. And those two people were already running at full capacity.

The activity was real. The architecture to hold it was not.

The objective

The firm did not need more AI ideas. They had 33 of them already.

What the COO needed was a way to look at all of that activity and make honest calls: which builds were ready to move forward, which needed more groundwork before they were worth engineering time, and which were solving problems that had already changed shape after the restructuring.

The goal was to build the right things, in the right order, with a foundation that would hold.

What we did

Phos AI Labs started where every engagement starts: understanding the operation before touching anything.

The first two weeks were not about building. They were about looking. We ran discovery sessions across five departments, used voice-based AI interviews to map how work actually moved through each team, and reviewed every active build and every proposed one.

What we found shaped everything that followed.

A readiness scorecard for every department.

The audit produced one answer per department. Not a score. A call.

One team was ready to build. Their process was documented, their data was reliable, and their stack was clean. The conversation moved immediately to what to cut. Two manual reporting tools they had been paying to maintain for years. Both replaceable with a single AI agent. One line removed from the budget, five figures back in the business, and a workflow that now runs without anyone touching it.

One team needed to slow down before they could speed up. Real goals, real pressure, and no documented process underneath any of it. They were using three different tools to produce the same weekly output depending on who was available. Building an AI layer on top of that would have automated the inconsistency, not fixed it. The call: document the workflow first. Then we build.

One team was the most expensive conversation in the room. Four systems in daily use, none of them connected, with staff manually moving data between them every morning. One of those systems carried a six-figure annual contract for functionality the team was using at half capacity. The call: stop renewing. A custom AI agent connecting their actual data sources, built once, owned permanently, would cost less in year one than the contract they were already paying.

Every department left the audit with a specific answer and a defined next step.

A governance framework the team would actually use.

This was the piece the organization needed most, and the piece most AI initiatives skip entirely.

Every new build request went through a structured intake that took under 15 minutes to complete. Business problem, data sources, proposed workflow, expected users. A two-person triage committee reviewed every intake before any engineering time was allocated. Requirements had to be written down before anyone started building.

Every tool going into production had to meet four criteria before going live: single sign-on authentication through Microsoft Entra ID, role-based access tied to the HR system, data residency inside the firm’s own infrastructure, and a named owner responsible for keeping it current as the business changed.

A roadmap the team could execute in order.

The firm came in with 33 proposed builds. Every one of them had an internal champion pushing for it.

The roadmap delivered eight. Sequenced in a specific order, for a specific reason. Each build had to clear four gates before it made the list: a mapped process underneath it, a named owner inside the department, clear success criteria, and a defined integration point with the systems the team already used, including Copilot, ServiceNow, and the firm’s financial planning platform.

The first build for each department was chosen to produce a visible result within 30 days. Organizational trust is a prerequisite for the bigger builds; it has to be earned before it can be assumed.

The roadmap’s value was not the list of builds. It was the honest sequencing of them.

The outcome

Fourteen weeks after the engagement started, the firm had something most organizations never get from an AI initiative: a foundation they actually owned.

Phos AI Labs turned those fourteen weeks into three compounding assets: a governance framework the organization will run on for years, a sequenced build order the team owns and executes without external approval, and complete visibility into every active build across every department.

Before Phos AI Labs came in

• Multiple builds in flight with no intake process and no written requirements before engineering time was allocated
• Tools built by individuals lived in personal accounts; when someone left, the work left with them
• 33 proposed builds competing for resources with no agreed criteria for what moved forward and what waited
• No single place where leadership could see everything in flight, who owned it, and what it connected to
• The COO and her Head of AI were the approval layer for every single decision in the AI program

After the engagement

• Eight builds sequenced in a clear roadmap, each with a mapped process underneath it, a named owner, and a defined integration point with existing systems including Copilot, ServiceNow, and the firm’s financial planning platform
• Every active build tracked in a single use case registry: status, owner, integration dependencies, and adoption metrics
• A deployment architecture standard in place across every tool going into production
• The COO’s review load went from a constant bottleneck to a 30-minute weekly check
• Four systems deployed within 90 days of the roadmap being approved

The build that almost went first

The financial planning platform replacement sat at position four on the roadmap for one reason: the team needed to trust the process before they handed it their most critical financial system.

By the time it entered active design, four working systems were already live. The CFO came to the requirements sessions already aligned on scope. A conversation that would have taken weeks took one meeting.

That sequencing call, more than any single build, was what made the whole program work.

The COO reflected on the engagement at the six-month mark.

What’s next

The roadmap Phos AI Labs delivered was designed to compound. Each build creates the conditions for the next one. Each department that moves through the process becomes a reference point for the ones still waiting to start.

For this client, the work continues.

The communications cluster is live and expanding into new regional markets. The ServiceNow replacement is in active design. The financial planning platform replacement, the build that almost went first, is now in full discovery with a finance team that already trusts the process underneath it.

Phos AI Labs built the architecture that makes sure that everything is built right: in the right tech stack, in the right order, in the right place, with the proper guidelines, security, roles, and permissions.

That is the shift AI consulting is designed to create: a foundation the organization owns, a sequence it can execute, and a team that knows how to keep building without two people at the top approving every decision.

If your organization has AI activity but no foundation underneath it, that is exactly where this starts.

The AI Readiness Audit maps where you are. The AI Foundation gives you the sequence and the governance to move. Every build after that has solid ground beneath it.