OpenAI just launched a $200-per-month ChatGPT Pro subscription. Visa announced they’re processing AI-prompted transactions for them. The headlines make it sound like enterprise AI is finally here.

Here’s what they don’t tell you: if a user actually utilizes the full capacity of that subscription, OpenAI’s compute cost hits $14,000. That’s a 70x loss on every power user. They’re betting most people won’t push the limits.

Meanwhile, Lindy — a competitor in the AI agent space — moved 100% of their infrastructure to open-source models like DeepSeek. They cut their compute costs by more than 90% without sacrificing capability.

For consulting firms spending $2,000 to $20,000 a month on AI subscriptions across partners and senior staff, this matters. You’re paying for brand and convenience, not performance. The gap between what you’re charged and what the work actually costs is widening every quarter.

If your firm is serious about embedding AI into proposal generation, research synthesis, or knowledge management, you need to know where the money’s going and whether you’re getting value or just renting access to someone else’s margin.

The Real Cost of Renting AI by the Seat

Most consulting firms I work with started their AI journey the same way: a few ChatGPT Plus subscriptions at $20 a month, then Team at $25 per user, now Pro at $200 for the partners who need unlimited o1 access.

It feels incremental. But when you’re running a 12-person firm with six people on Pro and six on Team, you’re at $1,350 a month before you’ve built a single custom agent or integrated anything into your workflow.

That’s $16,200 a year for what amounts to a very good chatbot. You’re not automating proposals. You’re not synthesizing research across engagements. You’re not building institutional memory. You’re renting access to a model that forgets the conversation the moment you close the tab.

The value ceiling is low because the tool wasn’t designed for your business. It was designed for consumer scale with enterprise pricing bolted on.

Here’s the uncomfortable part: if you actually used ChatGPT Pro the way OpenAI’s pricing implies you should — running deep research queries, generating long-form content, iterating on complex analyses — their cost to serve you would be $14,000 a month. They’re subsidizing your usage and hoping you don’t notice.

But you’re not getting $14,000 worth of work out of it because the tool doesn’t know your clients, your past proposals, or your firm’s methodology. It’s generic by design.

What Open Models Change About the Economics

DeepSeek, Llama, and other open-source models now match GPT-4 on most business tasks. They’re not better at creative writing or edge-case reasoning, but for structured work — research summaries, proposal drafts, meeting transcripts, knowledge retrieval — they’re indistinguishable in output quality.

The difference is cost. Running DeepSeek on your own infrastructure or through a lightweight API costs about one-tenth what OpenAI charges. For a consulting firm processing 500 queries a month across research, proposals, and knowledge lookup, that’s the difference between $1,500 and $150.

Lindy’s move to open models wasn’t a publicity stunt. It was a recognition that the margin in AI tooling is about to collapse. The firms that figure this out early will have a structural cost advantage over competitors still paying SaaS rent.

But here’s the catch: you can’t just swap ChatGPT for DeepSeek and call it done. The value isn’t in the model. It’s in what you build around it.

A consulting firm doesn’t need a chatbot. It needs a Proposal Generation Agent that pulls past case studies, pricing history, and win themes into a tailored draft for the next RFP. It needs a Research Agent that runs structured industry and competitor analysis at the start of every engagement and outputs a one-page brief with sources. It needs a Knowledge Agent that reads every deck, document, and transcript the firm produces and answers questions across the entire corpus.

None of those agents require a $200-per-month subscription. They require a model, a data layer, and a workflow that fits how your firm actually works.

Where Consulting Firms Leak Money on Manual Work

Let’s be specific about what this looks like in a typical advisory practice.

Proposal time is the most visible cost. A partner or principal spends 20 to 40 hours writing a major proposal from scratch. They pull pieces from past decks, rewrite the methodology section, adjust pricing, and tailor the case studies. Win rate is fine, but the cost-of-sale is brutal. If you’re billing that partner at $300 an hour and they’re spending 30 hours on a proposal, that’s $9,000 in opportunity cost per pitch.

Multiply that across 15 to 20 major proposals a year and you’re at $135,000 to $180,000 in senior time that could have been spent on delivery or business development.

Research and synthesis is the hidden cost. Every engagement starts with secondary research: industry reports, competitor analysis, market sizing, regulatory context. A senior associate spends two weeks pulling this together. The output is good, but it’s not reusable. The next engagement in a similar space starts from scratch again.

Across a firm doing 30 engagements a year, that’s 60 weeks of research time — more than one full-time equivalent — doing work that compounds across clients but never gets captured as institutional knowledge.

Knowledge management debt is the long-term cost. Every project produces IP: frameworks, models, data sets, insights. Almost none of it is searchable or reusable. It lives in SharePoint folders, email threads, and individual hard drives. When a new engagement needs similar work, the firm pays for the same insight twice because no one can find what was already done.

For a firm doing $5M in revenue, this debt typically represents $80,000 to $300,000 in duplicated effort every year. It’s not a line item on the P&L, but it shows up as lower margins, slower delivery, and partners doing work that junior staff should be handling.

These are the costs that AI agents can actually address — but only if the agent is built for your workflow, trained on your data, and designed to produce outputs your team can use without heavy editing.

What an Agent-First Workflow Looks Like

Here’s what it looks like when a consulting firm moves from renting AI to building it.

A new RFP comes in. Instead of starting with a blank deck, the partner opens the Proposal Generation Agent. They input the client name, the scope, and a few key requirements. The agent pulls past proposals in the same sector, extracts relevant case studies, adapts the methodology section, and generates a first draft in 20 minutes.

The partner reviews it, makes strategic edits, and sends it to the team for refinement. Total time: four hours instead of 30. The agent didn’t write the final proposal, but it eliminated the blank-page problem and gave the partner a structure to react to instead of creating from scratch.

At the start of the engagement, the Research Agent runs a structured analysis: industry trends, competitor positioning, regulatory landscape, market sizing. It pulls from public sources, synthesizes the findings, and outputs a one-page brief with citations. A senior associate reviews it, adds firm-specific context, and delivers it to the client in two days instead of two weeks.

Throughout the engagement, every meeting transcript, deck, and document gets ingested by the Knowledge Agent. When a partner on a different project asks, “Have we done work in this space before?” the agent surfaces the relevant frameworks, models, and insights in seconds. The firm stops paying for the same analysis twice.

None of this requires a $200-per-month subscription per user. It requires a model that costs $15 to $50 a month to run, a data layer that connects to the firm’s existing files, and a set of workflows that match how the firm actually operates.

The difference in cost is 10x. The difference in value is higher because the agent is trained on your work, not the internet.

If you want a practical starting point for scoping this kind of system, we’ve put together a worksheet that walks through the decision tree: Deploy Your First Business Agent. It’s not a sales pitch. It’s a checklist for figuring out which process to automate first and what infrastructure you actually need.

The Build vs. Buy Decision for Consulting Firms

Most firms assume that building custom AI means hiring a data science team and spending six months on infrastructure. That was true in 2022. It’s not true now.

The tools for deploying agents on open models are mature. You don’t need to train a model from scratch. You need to fine-tune an existing one on your data, connect it to your file storage, and wrap it in a simple interface your team will actually use.

For a consulting firm, the build cost is typically $15,000 to $40,000 for the first agent, depending on complexity and data volume. Ongoing hosting and maintenance runs $200 to $800 a month. Compare that to $16,200 a year for ChatGPT subscriptions that don’t know your business and can’t access your files.

The payback period is usually three to six months. After that, you’re running AI at cost instead of renting it at margin.

The bigger question isn’t cost. It’s control. When you build on open models, you own the data layer. You decide what gets logged, where it’s stored, and who has access. When you rent SaaS AI, you’re trusting that the vendor’s privacy policy won’t change and that your client data won’t end up in someone else’s training set.

For consulting firms handling confidential client information, that’s not a theoretical risk. It’s a liability you’re accepting every time you paste a client brief into a third-party chatbot.

What the Omni Audit Finds in 60 Minutes

When I run an Omni Audit for consulting firms, I’m not trying to sell you a platform. I’m trying to map where your firm is leaking time and money on work that an agent could handle.

We spend 60 minutes on a call. I ask about proposal volume, research workflows, and how your team currently shares knowledge across engagements. I look at where senior people are doing work that doesn’t require their judgment — the stuff that’s necessary but repetitive.

By the end of the call, you get three things: a process map that shows where agents fit into your workflow, a cost model that estimates annual leakage and payback period, and a build path that prioritizes which agent to deploy first.

There’s no deck. No follow-up meeting. No pressure to sign anything. If the numbers don’t make sense for your firm, I’ll tell you. If they do, you’ll know exactly what to build and what it’ll cost.

Most firms I work with are losing $80,000 to $300,000 a year on duplicated research, proposal time, and knowledge management debt. The audit quantifies it and shows you where to start.

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

Why This Window Won’t Stay Open

The gap between what OpenAI charges and what open models cost is temporary. Either OpenAI will drop prices to compete, or more firms will figure out that they don’t need to rent AI at SaaS margins.

The firms that move early will have a structural cost advantage. They’ll be running AI at $500 a month while competitors are paying $5,000. They’ll own their data layer and control their workflows. They’ll be able to deploy new agents in weeks instead of waiting for a vendor roadmap.

The firms that wait will eventually make the switch, but they’ll do it under pressure — when a competitor is winning deals faster or delivering engagements at lower cost. By then, the advantage will be defensive, not offensive.

If you’re a consulting firm doing $1M to $25M in revenue and you’re spending more than $1,000 a month on AI subscriptions, the math is worth running. The cost to build is lower than you think. The payback is faster than you expect. And the control you gain over your data and workflows is worth more than the dollar savings.

We’ve built agents for research, proposals, and knowledge management across dozens of advisory practices. The pattern is consistent: firms that move from renting AI to building it see payback in three to six months and cost reductions of 70% to 90% after that.

You can keep paying $200 a month per user for a chatbot that forgets your work, or you can build an agent that knows your clients, your methodology, and your past engagements. The infrastructure is ready. The models are proven. The only question is whether you want to own the tooling or rent it.

If you want to see what this looks like for your firm specifically, book your Omni Audit and we’ll map it in 60 minutes. No deck, no pitch, just the numbers and the build path.

The firms that figure this out in 2026 will have a margin advantage their competitors won’t catch for years. The window’s open. Don’t wait for it to close.