Gartner just put a number on what we’ve been watching unfold for the past eighteen months: 40% of agentic AI projects will fail by 2027. Not because the models aren’t capable. Not because the vendors overpromised. They’ll fail because firms don’t have the process discipline to make automation work.

If you run a consulting firm, that stat should stop you cold. Because right now, every conversation in our world is about agents. Claude Fable 5 just dropped with new safeguards. Cursor Bugbot completes code reviews in 90 seconds. Perplexity Computer routes tasks across 20 models to produce work-ready reports. The technology is real, it’s fast, and it’s getting cheaper every quarter.

But here’s what nobody’s saying out loud: if you can’t describe your workflow in a document, you can’t hand it to an agent. And most consulting firms can’t describe their workflows. Not the real ones. Not the way proposals actually get written, or how research actually gets done, or where the knowledge from last year’s engagement actually lives.

That gap between “we should automate this” and “we have a process worth automating” is where the 40% failure rate lives. It’s also where $80K to $300K in annual leakage sits in firms your size, waiting to be recovered or wasted depending on what you do next.

The Process Debt Most Firms Don’t See

Walk into any consulting firm doing $1M to $25M and ask how proposals get written. You’ll hear some version of “the partner pulls together past work and writes it.” Ask where past work lives. You’ll get Dropbox, SharePoint, someone’s hard drive, and a few decks in email threads.

That’s not a process. That’s institutional memory held hostage by whoever’s been there longest.

Here’s what the real workflow looks like for a major proposal:

The partner gets the RFP or the inbound lead. They block 8 to 12 hours over the next week to write the response. They pull three or four past proposals that feel relevant. They rewrite the case studies. They rebuild the pricing table. They write the methodology section from scratch because it’s faster than finding the last good version. They send it to a junior person for formatting. The junior person rebuilds the template because the last one was in the old brand. Total cost of sale: 20 to 40 hours of senior time, most of it spent recreating work the firm already paid for once.

Now multiply that across every proposal, every pitch, every new engagement. You’re not paying for sales. You’re paying a recurring tax on the fact that your firm doesn’t have a system.

The same pattern shows up in research. Every engagement starts with 10 to 20 hours of secondary research. Industry trends, competitive landscape, regulatory environment, company background. It’s good work. It’s also repeated work, because the research from the last three engagements in the same sector is scattered across folders and nobody’s sure what’s still current.

And then there’s knowledge management. Every project produces documents, decks, models, frameworks, insights. Most of it never gets used again. Not because it’s not valuable. Because nobody can find it, and even if they could, they don’t trust that it’s the latest version or that it applies to the new context.

This is the process debt that makes agentic AI projects fail. You can’t automate chaos. You can’t hand an agent a pile of unstructured files and say “figure it out.” Well, you can. But that’s how you end up in the 40%.

What Process Discipline Actually Looks Like

Before you build an agent, you need to document the workflow you want it to execute. Not the aspirational workflow. The one that actually happens.

For proposal generation, that means writing down where past proposals live, what sections are reusable, how pricing gets structured, what case studies apply to which verticals, and what the approval chain looks like. It means deciding whether the agent drafts the full document or just pulls the building blocks. It means knowing what good looks like so you can evaluate the output.

For research, it means defining what questions you ask at the start of every engagement, what sources you trust, how you want the synthesis structured, and what format the deliverable takes. It means deciding whether the agent runs the research autonomously or whether it’s a co-pilot that accelerates a human-led process.

For knowledge management, it means tagging your existing corpus so the agent knows what’s a final deliverable versus a working draft, what’s client-specific versus reusable IP, and what’s current versus archived. It means building a taxonomy that reflects how your people actually search for information.

None of this is glamorous. It’s not the part of AI that gets covered in the press. But it’s the part that determines whether your implementation works or joins the 40% that don’t.

The firms that are getting this right aren’t the ones with the biggest AI budgets. They’re the ones that treated the AI project as a forcing function to finally document their processes. They used the implementation as an excuse to clean up the mess they’ve been meaning to fix for years.

One mid-sized strategy firm we work with spent the first month of their Omni implementation just mapping their proposal workflow. No code. No agents. Just a Miro board and a series of interviews with every partner. What they found was that each partner had their own version of the process, and none of them matched what the junior staff thought was happening. The AI project didn’t start until they had one agreed workflow. When it did start, it worked.

What an Agent Actually Does in This Workflow

Let’s get specific. Here’s what a Proposal Generation Agent looks like when it’s built on top of a documented process.

You get an RFP. You open the agent interface and paste the RFP text. The agent reads it, identifies the sector, the scope, and the key requirements. It searches your past proposals for relevant case studies, pulls your standard methodology for this type of engagement, generates a pricing table based on your rate card and the estimated hours, and drafts a proposal document with all the sections populated.

You’re not starting from a blank page. You’re starting from a 70% draft that includes real content from real past work. You spend your time refining the positioning, tailoring the case studies, and adjusting the pricing. Instead of 20 hours, you spend 4. Instead of recreating the same work, you’re adding the judgment and client-specific insight that only you can add.

The agent isn’t replacing you. It’s doing the part of the job that shouldn’t require a partner’s time in the first place.

A Research Agent works the same way. You kick off a new engagement. You tell the agent the client name, the sector, and the three questions you need answered. It runs structured searches across your trusted sources, pulls relevant data, synthesizes it into a one-page brief with citations, and flags anything that conflicts with past research you’ve done in the same space.

The research still needs your interpretation. But the 15 hours of secondary research just became 2 hours of review and synthesis. And because the agent is working from a defined process, you know what you’re getting. No surprises. No gaps. No wondering if it missed something.

A Knowledge Agent sits on top of your entire document corpus. Every deck, every model, every meeting transcript. You ask it a question and it searches across everything the firm has ever produced, surfaces the relevant sections, and gives you an answer with links to the source documents.

It doesn’t replace institutional memory. It makes institutional memory accessible to people who don’t have it yet. The junior consultant who just joined can ask the same question the 10-year partner would ask and get a real answer, not a “check with Sarah, she might remember.”

These agents work because they’re built on top of processes that were documented before the agents existed. The firms that skip that step are the ones that end up with agents that hallucinate, pull the wrong content, or produce output that nobody trusts. That’s the 40%.

The Omni Audit: Where Process Discipline Starts

If you’re reading this and thinking “we don’t have documented processes,” you’re not alone. Most firms don’t. That’s not a criticism. It’s the reality of how consulting businesses grow. You hire smart people, they figure it out, and the firm scales on talent rather than systems.

But agentic AI doesn’t work that way. It needs the system first.

That’s why we built the Omni Audit for consulting firms. It’s a 60-minute working session where we map one high-cost workflow in your business, identify where the manual work is happening, and show you what an agent doing that work would look like.

You walk away with three things: a process map of the workflow you chose, a list of the manual tasks that are automatable, and a build spec for the agent that would handle it. No deck. No discovery phase. No six-week scoping exercise. Just the information you need to decide whether this is worth doing.

The audit isn’t a sales call. It’s a forcing function. Because the act of mapping the workflow forces you to confront the process debt. And once you see it, you can’t unsee it.

Most firms pick proposal generation or research as their first workflow to audit. Those are the two that have the clearest ROI and the most obvious pain. But we’ve run audits on knowledge management, client reporting, market sizing, and even internal hiring workflows. The pattern is always the same: the firm thinks they have a process, the audit reveals they have five different processes, and the conversation shifts from “should we automate this” to “we need to fix this whether we automate it or not.”

Book a 60-min Omni Audit and we’ll map it together.

The Dollar Reality of Doing Nothing

Let’s talk about what this costs if you don’t fix it.

A partner at a $5M consulting firm bills at $300 to $500 an hour. If that partner is spending 20 hours a month on proposal work that could be handled by an agent, that’s $6K to $10K a month in opportunity cost. Over a year, that’s $72K to $120K. And that’s just one partner.

Add in the research work that gets repeated across engagements. Add in the knowledge management debt where the firm pays for the same insight twice because nobody can find the first version. Add in the junior staff time spent reformatting decks and rebuilding templates. You’re looking at $80K to $300K in annual leakage for a firm doing $1M to $25M. That’s the range we see when we run the audit.

Now compare that to the cost of fixing it. An agent implementation for one workflow typically runs $15K to $40K depending on complexity. If it saves 15 hours a week of senior time, it pays for itself in three to six months. Every month after that is pure margin recovery.

But here’s the part that doesn’t show up in an ROI calc: the strategic capacity you get back. When your partners aren’t spending 20 hours a month writing proposals, they can spend that time on client development, thought leadership, or just thinking. When your senior consultants aren’t doing secondary research, they can focus on the synthesis and insight that clients actually pay for.

The firms that are winning right now aren’t the ones with the most sophisticated AI stack. They’re the ones that figured out how to get their senior people out of the weeds. Agentic AI is just the tool that makes that possible at scale.

Why This Matters Right Now

The window for getting this right is shorter than most firms think. The technology is moving fast. Claude Fable 5, Cursor Bugbot, Perplexity Computer, these aren’t future products. They’re live today. Google Gemini 3.5 Pro is launching this month with a 2 million token context window. That’s enough to fit your entire proposal library in a single prompt.

The firms that document their processes now and build agents on top of them will have a 12 to 18 month lead on the firms that wait. And in a business where competitive advantage is measured in expertise and efficiency, 18 months is a long time.

The firms that skip the process discipline step and try to automate chaos will end up in the 40%. They’ll spend the budget, get output they don’t trust, and conclude that agentic AI doesn’t work for consulting. They’ll be wrong, but they won’t know it until their competitors are running at half the cost of sale and twice the throughput.

You don’t need to become an AI company. You need to become a consulting firm with documented processes and agents that execute them. That’s the gap between the 60% that succeed and the 40% that don’t.

If you want to see what that looks like for your firm, book your Omni Audit here. We’ll map one workflow, show you where the leakage is, and give you the spec for the agent that fixes it. Sixty minutes. Three outputs. No deck.

The process discipline starts with seeing the process. Let’s map it.

For more on how consulting firms are using AI to recover margin and scale expertise, visit our insights library or explore the full Omni platform we’ve built for professional services firms.