Uber stopped building a central AI team. Instead, they assigned their best engineers to business units with one job: find the manual work and automate it. No grand strategy deck. No transformation roadmap. Just people who understand the technology sitting next to people who understand the business, fixing the expensive stuff.

Consulting firms should do the same thing. Not because Uber did it, but because the work that kills your margin lives in the practice areas. Proposal writing. Client research. Knowledge management. Every senior person in your firm spends 20 hours writing a deck that pulls from three past proposals they can’t find. Every engagement starts with two weeks of secondary research someone else already did. Every project produces IP that dies in a folder.

You don’t need a centralized AI function. You need someone who knows what an agent can do, rotating through your teams, identifying the repeated work and building the automation. This article walks through what that looks like for a consulting firm doing $1M to $25M in revenue, where the cost of manual work compounds faster than headcount.

The Work No One Sees

A partner at a strategy firm told me his team spent 35 hours on a proposal last month. The client wanted a market entry study for a B2B SaaS company. The firm had done four similar projects in the past two years. They had the case studies. They had the methodology. They had the pricing model. But none of it was in one place, so the partner and two consultants rebuilt it from memory and old decks.

They won the work. The proposal was good. But the firm paid $12,000 in billable time to produce a document the client never paid for. That’s the cost-of-sale, and it’s invisible until you add it up across a year. For a firm doing 30 proposals annually, that’s $360,000 in senior time that could have been client work.

The research phase is worse. Every engagement starts with the same pattern: industry overview, competitive landscape, regulatory environment, customer segmentation. One consulting firm I work with has 18 people. They run six to eight active projects at any time. Each project kicks off with 15 to 20 hours of research. Most of that research overlaps with something they did six months ago, but there’s no system to pull it forward. So they do it again.

Knowledge management is the long tail. Every deck, every meeting transcript, every deliverable your firm produces contains something useful for the next client. But it sits in folders organized by project name, which means it’s organized for nobody. A consultant working on a healthcare pricing engagement can’t find the pricing model from the logistics project, even though the logic is identical. The firm pays for the same insight twice.

These three problems share a structure. They’re repeated work. They require senior judgment at the end, but the first 70% is mechanical. And they compound. The more projects you run, the more proposals you write, the more knowledge you generate, the worse the problem gets.

What Embedding Looks Like

Uber’s approach was simple. Take an engineer who understands machine learning and API design. Put them in the operations team for three months. Their job isn’t to build a product. It’s to watch the work, find the repetitive tasks, and automate them. Then move them to the next team.

For a consulting firm, that person isn’t an engineer. It’s someone on your team who understands what agents can do and how to connect them to your workflows. It might be a consultant who’s been experimenting with AI tools. It might be an ops person who’s automated parts of your CRM. It doesn’t matter. What matters is they have permission to spend a month in each practice area asking one question: what do we do every time?

In a strategy practice, the answer is proposals. The embedded person sits with the partners and maps the proposal process. They identify the repeated sections: firm overview, team bios, case studies, methodology, pricing. They find the past proposals that contain those sections. Then they build a Proposal Generation Agent that pulls the relevant pieces and assembles a first draft when a new opportunity comes in.

The agent doesn’t write the proposal. It gives the partner a 70% draft in 20 minutes instead of a blank page. The partner still tailors the positioning, adjusts the pricing, and writes the executive summary. But the mechanical work is done. The 35-hour proposal becomes a 10-hour proposal, and the firm gets 25 hours of partner time back.

In a research-heavy practice, the embedded person builds a Research Agent. It takes a client name and an industry, runs structured searches across public sources, pulls financial data, identifies competitors, and produces a one-page brief with citations. The consultant still interprets the data and builds the analysis, but the first two weeks of secondary research collapse to two hours. That’s 60 hours per project, across eight projects a year, which is 480 hours of consultant time returned to client work.

For knowledge management, the embedded person builds a Knowledge Agent that reads every document the firm produces and answers questions across the corpus. A consultant working on a pricing engagement types “show me past pricing models for B2B services” and gets three examples with context. The agent doesn’t replace judgment, it surfaces the work the firm already paid for so you don’t pay for it again.

This is what Omni for consulting firms does. It’s not a platform you adopt. It’s a set of agents you deploy in the places where repeated work lives, built by someone who understands your business and knows what the technology can do.

The Dollar Reality

A consulting firm doing $5M in revenue with 12 people typically runs 40 to 50 client engagements per year. If each engagement requires a proposal, and each proposal takes 25 hours of senior time at a $200 blended rate, that’s $250,000 in cost-of-sale annually. Cut that time in half with a Proposal Generation Agent, and you’ve returned $125,000 worth of billable capacity to the firm.

Research time is harder to quantify because it’s baked into project budgets, but the pattern holds. If six active projects each require 20 hours of secondary research at the start, and you can compress that to 5 hours with a Research Agent, you’ve freed 90 hours per cycle. Over a year, that’s 360 hours, or $72,000 in consultant time that can be redeployed to client work or used to take on an additional project without hiring.

Knowledge management doesn’t show up as a line item, but it shows up as repeated effort. One firm I work with estimated they rebuild the same analysis three times per year across different clients. That’s not because the consultants are inefficient. It’s because they can’t find the original work, so they start over. A Knowledge Agent doesn’t eliminate that work, but it cuts the duplication. If you’re paying for the same insight twice, you’re leaving $50,000 to $100,000 on the table depending on the complexity of your engagements.

These numbers aren’t projections. They’re the cost of manual work that already exists in your firm. The question isn’t whether you can afford to automate it. The question is whether you can afford not to.

If you want a structured way to identify where agents fit in your workflow, we built a worksheet that walks through the process. Download the Deploy Your First Business Agent guide and use it to map the repeated work in one practice area. It takes 30 minutes and gives you a clear picture of where to start.

Why Centralized AI Teams Don’t Work

Most consulting firms that try to adopt AI start with a centralized function. They hire someone to own the AI strategy, or they assign a partner to lead an innovation committee. That person spends three months researching tools, building a business case, and presenting a roadmap to the leadership team. Then nothing happens, because the centralized team doesn’t understand the work well enough to automate it, and the practice teams don’t have time to explain it.

Uber figured this out early. Centralized AI teams build tools that solve theoretical problems. Embedded engineers build tools that solve the problem in front of them. The difference is specificity. A centralized team builds a “knowledge management platform.” An embedded person builds an agent that answers the exact question a consultant asked yesterday.

For consulting firms, this means the person automating the work needs to sit in the practice area long enough to see the repeated tasks. They need to watch a partner write a proposal. They need to sit through the research phase of a project. They need to hear the consultant say “I know we did this before, but I can’t find it” three times in one week. That’s when the opportunity becomes obvious.

The rotation model works because it spreads the capability across the firm without requiring every practice to hire an AI specialist. One person spends a quarter in strategy, a quarter in operations, a quarter in financial advisory. They build the agents, train the team to use them, and move on. After a year, the firm has automated the repeated work in every practice area, and the embedded person has a full picture of where the next opportunities are.

This is the structure behind the AI audit for consulting firms. It’s a 60-minute session where we map the repeated work in your highest-cost practice area, identify the agent that fits, and give you the build plan. No deck. No roadmap. Just the next thing to automate and the person who should build it. Book a 60-min Omni Audit and we’ll walk through it together.

What You Build First

The first agent you build should target the work that costs the most and repeats the most. For most consulting firms, that’s proposals. Proposal Generation Agents are straightforward to deploy because the inputs are structured: opportunity details, client background, scope of work. The outputs are structured too: firm overview, team bios, methodology, pricing. The agent pulls from past proposals, assembles a draft, and hands it to the partner for final positioning.

The second agent depends on your practice mix. If you run research-heavy engagements, build a Research Agent. If you have a knowledge management problem, build a Knowledge Agent. If you spend too much time scheduling and coordinating across client teams, build a scheduling agent that handles the back-and-forth. The pattern is the same: identify the repeated task, map the inputs and outputs, build the agent, deploy it in the practice area, and measure the time saved.

You don’t need to build all of this at once. One firm I work with started with a Proposal Generation Agent in their strategy practice. It saved 15 hours per proposal. They used that time to take on two additional clients in the next quarter, which paid for the cost of building the agent four times over. Six months later, they built a Research Agent for their market entry practice. That saved another 60 hours per project cycle. Now they’re working on a Knowledge Agent because the research and proposal work surfaced how much duplicated insight they were producing.

This is the compounding effect of embedding AI capability in your practice areas. Each agent makes the next one easier to justify, because the time saved from the first agent funds the build for the second. After a year, the firm has automated the repeated work across every major practice, and the cost-of-sale drops by 30% to 50% without changing headcount.

If you want to see what this looks like for your firm, the Omni Audit walks through it in detail. We map your highest-cost repeated work, identify the agent that fits, and give you the build plan in 60 minutes. No pitch. No follow-up deck. Just the next thing to automate and the person who should own it. You can explore more about how we approach this across Omni Ops and the broader Omni platform, or dive into case examples and frameworks in our insights library.

The Rotation Schedule

The embedded person should spend one quarter in each practice area. That’s long enough to see the repeated work, build the agent, and train the team to use it. It’s short enough that they don’t get pulled into client delivery and lose focus on automation.

In the first month, they observe. They sit in on proposal meetings. They watch the research process. They ask questions about what gets repeated and what takes the most time. They don’t build anything yet. They just map the work.

In the second month, they build. They take the highest-cost repeated task and build the agent that automates it. They test it with one project, refine it based on feedback, and deploy it across the practice.

In the third month, they train. They run workshops with the practice team, document the workflow, and hand off ownership to someone in the practice who will maintain it. Then they move to the next practice and start again.

After four quarters, the embedded person has built agents in every major practice area. The firm has automated the repeated work that was costing $80,000 to $300,000 per year in senior time. And the embedded person has a full map of where the next opportunities are, because they’ve seen the work across the entire firm.

This is how you scale AI capability without hiring a team. One person, rotating through the business, building the automation where it matters. It’s not a transformation program. It’s a build schedule. And it works because the person building the agents understands the work well enough to automate the right thing.

What Happens Next

Most consulting firms know they should be doing something with AI. They’ve read the articles. They’ve sat through the demos. They’ve added “AI-powered” to a pitch deck. But they haven’t automated the work that costs them $200,000 per year in repeated effort, because they don’t know where to start.

Start with one practice area. Pick the one where senior people spend the most time on repeated tasks. Map the work. Build the agent. Deploy it. Measure the time saved. Then move to the next practice and do it again.

If you want help mapping the first one, book my Omni Audit. We’ll spend 60 minutes walking through your highest-cost repeated work, identify the agent that fits, and give you the build plan. No deck. No roadmap. Just the next thing to automate and the person who should build it.

The firms that win in the next three years won’t be the ones with the best AI strategy. They’ll be the ones that embedded the capability in every practice area and automated the work no one sees. That’s where the margin is. And that’s where you should start.