Fortune ran a piece recently on why every company suddenly wants an AI model router. If you run a consulting or advisory firm and you’ve watched your AI bill creep past $500 a month without anyone quite knowing why, this is the article that explains it. It’s not a hype story. It’s a plumbing story, and plumbing is where the money leaks out.

The problem is not the AI, it’s the routing

Most firms adopt AI the same way, one tool at a time. Someone starts using a premium model for everything, from drafting a one-line email to synthesizing forty pages of industry research. The premium model is good at both. It’s also priced for the hard problem, not the easy one. When every task gets routed to your most expensive model by default, you’re paying top-tier rates for work that a cheaper model handles just as well.

This is the exact mechanism the model router story is about. A router sits between your team and the models, looks at what the task actually requires, and sends it to the right tool for the job. Simple classification, formatting, or first-pass drafting goes to a fast, cheap model. Complex synthesis, nuanced client writing, or anything with real judgment calls goes to the expensive one. The firms Fortune is describing are seeing 40-60% reductions in their AI spend just from this one change, with no drop in output quality because the routing logic is built around task complexity, not convenience.

For a consulting firm doing $1M-$25M in revenue, this isn’t a rounding error. It’s the difference between AI being a line item your partners grumble about and AI being a tool that pays for itself several times over every month.

Where the manual work is actually hiding

Model routing solves a cost problem, but the reason most firms are even spending $500+ a month in the first place is that they’ve been trying to patch three specific gaps with brute-force AI use. Worth naming them plainly.

Proposal and pitch time. Senior people are still writing decks and proposals close to scratch for every new opportunity. Win rates might be fine, but the cost-of-sale is brutal, often 20-40 hours of partner and manager time per major proposal. When that work gets pushed through a general-purpose AI tool with no memory of past proposals, the firm pays premium-model prices to reinvent something it already built six months ago.

Research and synthesis. Every engagement starts with weeks of secondary research, most of it repeating work the firm has already done for a different client in a similar sector. This is the pain that compounds. Each analyst runs their own research pass, feeds it into whatever AI tool is open, and the firm never builds a reusable base of industry knowledge. It just pays for the same synthesis over and over.

Knowledge management debt. Every project generates real intellectual property, decks, memos, transcripts, frameworks. Almost none of it is reusable across the firm. Ask a consultant in Q3 whether the firm has done work on a topic before, and the honest answer is usually “I think so, ask around.” That’s not a knowledge problem so much as a retrieval problem, and retrieval is exactly the kind of task that should never touch your most expensive model.

If you want a full breakdown of how these gaps show up in your specific numbers, our guides section walks through the diagnostic in more detail. But the short version is that the AI tools your team is already using were never wrong to adopt. They’re just being deployed without any cost discipline behind them.

What this looks like end-to-end with the right agents

The fix isn’t “use AI less.” It’s building the routing logic into the actual workflows your team runs every week, so the cost discipline is automatic instead of a policy nobody follows. At Omni, we build this as named agents that live inside your existing tools, not a new platform to learn.

Proposal Generation Agent. This one pulls from your firm’s past proposals, case studies, and pricing history to produce a tailored first draft the moment a new opportunity comes in. The heavy lifting, structuring the narrative, matching case studies to the client’s sector, pricing consistency, gets handled by a model sized for that complexity. The formatting and boilerplate sections route to something cheaper. Your partner still reviews and refines, but they’re editing a strong draft instead of staring at a blank deck at 9pm.

Research Agent. At the start of every engagement, this agent runs a structured research pass, industry trends, competitive landscape, relevant regulatory or market shifts, and delivers a one-page brief with sources attached. It’s not replacing your analysts’ judgment. It’s replacing the first three days of Googling and PDF-skimming that every engagement starts with. Simple lookups and summarization get routed cheap. The synthesis step, where the agent has to weigh conflicting sources and produce a coherent point of view, gets routed to a model that can actually handle it.

Knowledge Agent. This is the one that pays off the knowledge management debt. It reads every deck, document, and meeting transcript your firm produces and lets anyone on the team ask it a question across the whole corpus. “Have we done work on supply chain resilience in mid-market manufacturing?” gets answered in seconds with the actual source documents attached, instead of a Slack message that goes unanswered for two days.

None of these three agents require your team to change how they work day to day. They require someone to design the routing and the workflow correctly once, and that’s the part most firms skip. You can see how this fits together specifically for advisory work on the AI audit for consulting firms page, which walks through how these agents map onto a typical engagement lifecycle.

The dollar math for a firm your size

Let’s put real numbers against this instead of leaving it abstract. Say your firm has six consultants using AI tools regularly, and the average spend per seat has crept to $60-$120 a month across various subscriptions and API costs. That’s $4,300-$8,600 a year on tools alone, before you count the hours lost to research that gets redone and proposals that get rebuilt from zero.

Now factor in the time cost. If your senior team spends even 15 hours a month collectively on proposal drafting that a well-built Proposal Generation Agent could cut to 4-5 hours, and your blended senior rate is anywhere near $150-$250 an hour, you’re looking at $1,500-$2,500 a month in recovered capacity. Multiply that across a year and you’re well into five figures, and that’s before the Research Agent and Knowledge Agent start compounding savings on their own fronts.

This is where the model router conversation and the agent-design conversation meet. Routing alone cuts your raw AI spend by 40-60%. Building the right agents on top of that routing layer is what turns “cheaper AI” into “less manual work.” Firms that only fix the routing keep paying for the manual labor underneath it. Firms that only add agents without routing discipline end up with a bigger, badly-optimized AI bill instead of a smaller one.

If you want a structured way to think through which workflow to fix first, we put together a practical worksheet called Deploy Your First Business Agent. It’s built for exactly this decision, picking the first agent that pays for itself fastest rather than the one that sounds most impressive in a partner meeting. You can grab the direct download here if you’d rather skip the landing page.

What an Omni Audit actually gives you

We don’t open with a platform pitch because most firms don’t need one yet. What they need is a clear picture of where the money and hours are actually going, and that’s what the audit is built to produce.

It runs 60 minutes, on a call, and there’s no deck. You walk away with three things: a breakdown of where your current AI spend is going and how much of it is unrouted premium-model cost, a shortlist of the two or three workflows in your firm bleeding the most hours (usually proposal work, research, or internal knowledge retrieval), and a rough dollar estimate of what fixing each one is worth annually. No sales deck, no 40-slide framework walkthrough. Just the numbers, from your business, with a recommendation attached.

For a firm doing $1M-$25M, this usually surfaces something in the $80,000-$300,000 range in recoverable cost and capacity, though every firm’s mix is different and yours might land above or below that. The point isn’t the exact number. It’s having a real one instead of a guess.

You can book my Omni Audit directly, no pre-call form to fill out first. We’ll go through your current AI stack, look at where your team’s time actually goes on a typical engagement, and tell you plainly whether routing and agents are worth building for your specific situation. Sometimes the honest answer is “not yet, fix this operational thing first.” We’d rather tell you that than sell you something you don’t need.

Start with one agent, not a platform overhaul

If there’s one thing we’ve learned building these systems for advisory firms, it’s that trying to fix everything at once is how these projects stall. The firms that get real value pick one workflow, usually the one costing the most partner hours, and fix it completely before moving to the next.

Model routing is the mechanism that makes AI spend rational again. The agents are what turn that savings into actual freed-up hours for your team. Neither one works as well without the other, and neither one requires ripping out what you already use.

Our insights page has more breakdowns like this one if you want to see how other verticals are approaching the same routing problem, and our blog covers the build side in more technical depth if you’re curious how the agents actually get wired into tools like Slack, Notion, or your CRM. But if you run a consulting firm and you’re spending real money on AI every month without a clear sense of where it’s going, the fastest next step is still seeing Omni for consulting firms and getting the actual numbers in front of you.

Sixty minutes. Three outputs. No deck. That’s the whole ask.