Corporate AI budgets have moved ahead of proven returns

A recent report on US corporate AI spending, based on Goldman Sachs analysis, landed on a point consulting partners should pay attention to. Corporate AI budgets are accelerating, while the direct earnings impact is still limited.

That gap isn’t a reason to wait. It’s the commercial condition your firm needs to work within.

Your clients aren’t buying AI only because they can point to an immediate 15 percent productivity lift in a quarterly report. Many are spending because they believe AI will reshape their operating model, their customer experience, their cost base, or their position in the market. They don’t want to be the company that discovers too late that a competitor built better capabilities while they were waiting for a perfect business case.

For consulting and advisory firms, this changes the proposal conversation.

A weak AI proposal promises instant savings, broad automation, and a big return that nobody can yet defend. A strong proposal frames the work as a managed strategic bet. It identifies where uncertainty sits, what can be tested in 30, 60, or 90 days, what data and process foundations are required, and how leadership will decide whether to scale.

That approach is more honest with clients. It also gives your firm a better path to valuable work.

The issue is that most firms trying to advise clients on AI still run their own business through manual research, scattered decks, and senior people rebuilding proposals from scratch. That mismatch isn’t just awkward. It is expensive.

For a consulting firm doing $1 million to $25 million in revenue, we usually see annual operational leakage in the $80K to $300K range. The source is rarely one bad process. It is the accumulated cost of repeated research, delayed proposal drafts, inaccessible past work, and partners spending their best hours looking for material their firm already owns.

Why immediate ROI is the wrong proposal promise

Clients have heard enough AI claims to be cautious. They may have already paid for a pilot that produced a polished demo and very little change in day-to-day work. They may have bought licences before their teams knew which processes could use them. They may be under pressure from a board that wants an AI plan and a finance leader who wants numbers that stand up.

That creates a real tension.

The client wants to act. They also don’t want to sponsor a vague technology experiment. Your proposal needs to help them hold both ideas at once.

A useful strategic AI proposal does four things.

First, it separates capability investment from near-term efficiency. Some work will have a clear operational case, such as reducing time spent preparing recurring management reports or triaging internal requests. Other work is a capability bet, such as building a proprietary research workflow or improving how account teams prepare for complex client meetings.

Second, it starts with a narrow decision. Instead of proposing an enterprise-wide transformation, define one workflow, one user group, and one measurable operating problem. You can then establish a baseline and learn quickly.

Third, it makes governance part of the scope. Clients need clear ownership, source controls, review steps, security boundaries, and criteria for moving from pilot to scaled deployment.

Fourth, it shows the client what they will own at the end. A strategic project cannot end with your team handing over a slide deck. It should leave behind a working workflow, a documented operating process, an adoption plan, and a decision on the next investment.

That is the right message for a market where spend is rising faster than reported earnings impact. You aren’t selling certainty that doesn’t exist. You are selling a disciplined way to place a smart bet.

The catch is that you need to produce this kind of proposal quickly, with evidence drawn from your own experience. That is where most firms get stuck.

The proposal cost your P&L doesn’t show clearly

A major proposal can consume 20 to 40 hours of senior time before it is even presented. A partner shapes the commercial angle. A director searches old decks for relevant case studies. A manager asks three people for the latest credentials. An analyst rebuilds market context and client background. Someone tries to locate prior pricing, often through an old email thread.

The pitch may still be good. The firm may even win.

But the cost of sale is brutal when this happens on every substantial opportunity.

The problem isn’t that your people don’t know how to write. It is that the firm has no reliable system for bringing prior knowledge into the next commercial conversation. A strong case study sits in a PDF on one person’s laptop. A pricing model is in a spreadsheet with unclear assumptions. The most useful language from a past engagement lives in a meeting transcript that nobody will search again.

Senior people then become the retrieval system.

This is exactly the sort of work a Proposal Generation Agent in Omni ops can take on. It doesn’t replace the partner’s judgement about the client, the deal, or the commercial risk. It handles the costly assembly work that makes too many proposals start from zero.

The agent can begin with a short opportunity brief. It receives the client name, sector, stated problem, likely scope, deal stage, and any known constraints. It then searches approved past proposals, relevant case studies, capability decks, rate cards, and prior statements of work.

From there, it creates a tailored first draft with:

• A client-specific problem statement based on the opportunity brief
• Relevant credentials and past work, with the source material available for review
• A proposed workplan structured around decisions, milestones, and governance
• Commercial options linked to the type of work being proposed
• A list of gaps where a partner or director needs to add judgement

The partner still decides what promise the firm should make. They still remove claims that don’t fit. They still shape the relationship. But instead of spending a late evening pulling slides together, they are reviewing a draft built from the firm’s own approved material.

That is a more credible way to pitch strategic AI work to clients. It also demonstrates that your firm has applied the same operating discipline internally.

You can see the operating model behind this on Omni ops. If you want to identify where the proposal process is leaking time first, See Omni for consulting firms.

Research is where strategic AI projects start to repeat themselves

The second cost sits at the beginning of the engagement.

A client asks your firm to assess where AI could change their business. Before you can recommend anything, the team needs a view of the industry, the company’s market position, competitor moves, financial context, public AI initiatives, regulatory conditions, and likely operating constraints.

That is legitimate work. The issue is the way it gets done.

A consultant opens dozens of browser tabs. They download investor presentations. They read earnings calls. They search for competitor announcements. They paste findings into a working document. A manager then has to determine which sources are credible, what is current, and which observations actually matter to the client decision.

For some engagements, the research phase can run for weeks. Then, six months later, the firm begins a similar assignment in the same industry and repeats much of the process because the earlier research was never captured in a usable form.

Your client should pay for insight and judgement. They shouldn’t pay for your team to rediscover publicly available facts that another team collected last year.

The Research Agent in Omni ops is designed for this starting point. It works from a defined research brief rather than a vague instruction to “find everything.” Your team sets the industry, company, geographic market, decision question, time period, and source requirements.

The agent then runs a structured process:

1. It gathers information from defined public and approved sources.
2. It records source links and dates, so the team can trace the evidence.
3. It groups findings into agreed categories such as market dynamics, competitors, customer signals, technology adoption, regulation, and financial performance.
4. It identifies contradictions or missing evidence rather than treating every web page as fact.
5. It produces a one-page brief for the project lead, followed by deeper source-backed summaries where needed.

The output is not a strategy recommendation by itself. It is a faster and more reliable starting point for the people who make recommendations.

That distinction matters. A consulting firm shouldn’t pretend an AI tool has solved the client’s strategic problem. The value comes from reducing low-value collection work so experienced people can spend more time testing the implications, challenging assumptions, and deciding what the client should do.

For ideas on making that shift without launching a large internal programme, review the practical material in our AI insights library. The point is to redesign one repeatable workflow, prove its value, then apply the learning elsewhere.

The knowledge debt behind every new engagement

Every project your firm completes creates IP.

There are decks, interview notes, workshop outputs, models, recommendations, meeting transcripts, issue logs, and final deliverables. Some are highly specific to one client. Others contain patterns that would help the next team avoid a week of work.

Most firms don’t have a real way to use that knowledge.

A shared drive is not a knowledge system. Neither is a document repository with folders that only make sense to the person who created them. Search can find a file name. It often cannot answer a practical question like, “What operating model recommendations have we made for mid-market manufacturers struggling with margin visibility?” or “Which past statements of work included a two-phase pilot before a full implementation?”

That is knowledge management debt. It becomes more expensive as the firm grows because more people create more material, while the ability to find and reuse it does not keep pace.

The Knowledge Agent reads approved decks, documents, and meeting transcripts across the firm corpus. It gives people a controlled way to ask questions in plain language and retrieve answers with links back to the source material.

A project manager preparing for a client workshop could ask for common adoption barriers from previous AI readiness assessments. A business development lead could ask for examples of how the firm has priced a diagnostic phase. A new consultant could ask which frameworks have been used successfully in a particular sector.

The answer should not be treated as final truth. It should be a source-backed starting point. That is why the design needs permissions, document ownership, retention rules, and a review process for sensitive client material.

Done properly, the Knowledge Agent makes the firm’s accumulated work more useful without creating an uncontrolled chat interface over confidential information. It helps protect margin because teams stop paying for the same insight twice.

This is also why an AI strategy for your clients gains credibility when it is grounded in your own operations. You understand the difference between an impressive answer and a reliable workflow because you have built one.

Position AI engagements as a sequence of decisions

When a corporate buyer says, “We need an AI strategy,” they may be asking for several different things at once. They may need direction from leadership. They may need a shortlist of opportunities. They may need a pilot. They may need governance. They may need help deciding which vendors deserve attention.

Don’t try to package all of that as a single promise of productivity.

Build a sequence of decisions into your proposal.

Stage one: identify the bets

Start with business priorities, operating pain points, available data, and constraints. The output is a ranked opportunity set, not a list of generic AI use cases.

For each opportunity, define the user, workflow, expected value, evidence required, risk level, and decision owner. This gives the client a basis for choosing where to test.

Stage two: prove one workflow

Choose an area where the work is frequent enough to matter and bounded enough to measure. Set an operational baseline before the pilot begins. That could be time spent producing a research brief, turnaround time on customer requests, or the percentage of work requiring rework.

Then make the evaluation criteria explicit. Is the aim to save time, improve response quality, increase capacity, improve consistency, or create a new client-facing capability? The answer will shape the implementation.

Stage three: decide what scales

At the end of the pilot, the client needs a decision pack. Continue, stop, redesign, or scale. If it scales, what ownership, controls, training, and data work are required?

This approach is commercially useful because it creates a clear advisory path. It also avoids trapping your firm in a proposal full of unprovable return claims.

If you are building this capability internally before taking it further into client work, Omni advisory can help structure the operating decisions around the technology.

Start with your firm’s own repeatable work

You don’t need to rebuild the entire firm around AI. You need to find the manual work that occurs often enough, costs enough, and has clear enough inputs to improve.

For many consulting firms, proposal creation, engagement research, and knowledge retrieval are the first three candidates because they affect both revenue and delivery margin.

A useful first step is to map the work honestly. Who does it now? How many hours are involved? What documents or systems are used? Where does quality drop? Where does the work wait for a senior person? What information cannot be used because nobody can find it?

Then choose one workflow and define a controlled first version. Include human review. Limit the data sources. Create a measure that matters. Run it long enough to learn, but not so long that the initiative disappears into committee meetings.

Our Deploy Your First Business Agent guide is a practical worksheet for that exercise. It helps you define the workflow, inputs, guardrails, owner, and first measure before you start building. You can also access the direct version here: Deploy Your First Business Agent.

The firms that benefit most from rising AI demand won’t be the ones making the loudest productivity claims. They will be the firms that can show clients a disciplined way to invest under uncertainty, while running their own operations with the same discipline.

Find the first workflow worth fixing

The opportunity is real. Corporate spending is moving. But your clients are right to ask what changes in the business, who owns the work, and how they will know if it is worth scaling.

Your own firm should ask the same questions.

An Omni Audit is a 60-minute working session, not a sales deck. We identify the workflows creating the most avoidable drag, assess the data and operating constraints, and leave you with three outputs: a prioritized opportunity list, a first-agent recommendation, and a practical implementation path.

If proposal work, research duplication, or inaccessible project IP is costing your firm time, Book a 60-min Omni Audit. You can also review the AI audit for consulting firms before the call.

The aim is not to chase an AI trend. It is to build one working capability that lowers cost, protects senior time, and gives your firm stronger evidence when clients ask what AI can actually do.

When you are ready to identify that first capability, Book my Omni Audit.