What Your Clients Now Assume

We have a large public sector client where our primary contact is an extraordinarily busy administrator running several high-profile programs at once. It’s been a good relationship, but every so often a request lands late Friday for technical input on a proposal or presentation due Monday.

One of those emails apologized for the short notice, then qualified the apology: “I’m sure you can just AI something in an hour, right?”

That was the first time I remember anyone using “AI” as a verb. And, while I use AI constantly when writing (including this piece), I never use it the way that sentence implies, least of all on high-stakes deliverables where the details matter and funding is on the line.

Still, “I’m sure you can AI it in an hour” is what many clients are thinking right now, even if they don’t always say it out loud.

There are three assumptions bundled within that sentence, and they’re worth separating.

Staffing

“You can AI this in an hour” is a staffing judgment made by someone who should not be dictating staffing. No consultant worth their salt would uncritically accept a client’s prescription for their human staffing mix, and there’s no reason clients should be specifying the human / machine mix either. But they’ve started to, because AI made the mix feel visible in a way that human org charts never were.

Acceptability

On the question of whether AI output is acceptable as a consulting deliverable, clients tend to go to one of two extremes:

“Don’t you dare use AI on my work.”

or

“Just ChatGPT it, I don’t care.”

Neither statement should be taken literally as a quality standard. Of the two, the second is especially dangerous in that it sounds like permission to deliver a quick and dirty deliverable with AI when in reality, no matter what they say, the client still expects quality. A better translation of their underlying intent would be:

“I think AI should dramatically reduce your cost and turnaround time, while I continue holding you responsible for the outcome.”

In these conversations, your firm’s goal shouldn’t be to convince clients that AI is inadequate. It’s to make sure the client appreciates what you add on top of AI’s default capabilities, and what a reasonable AI-assisted turnaround time looks like.

Competence

Clients often assume that AI is capable of delivering at an expert level… and, frankly, those clients are often right.

Today’s models are capable of dispensing advice based on the consensus of the internet at least as well as any undifferentiated generalist consultant. If your value proposition is just competent synthesis of publicly available best practices, that is now a commodity with a near-zero marginal cost, and no amount of positioning will fix it. Generalists without extraordinary gifts of client interaction are in real trouble, and I’d say that in a room full of them.

Where clients get it wrong is in cases that call for something more than a generic generalist: a contrarian view that cuts against the online consensus, ultra-niche domain expertise that was never published anywhere a model could learn it, deep discovery work full of unknown unknowns, and the resolution of highly nuanced edge cases. In those situations AI will still generate responses (that’s its nature) but what it generates might have the sound and sense of credible expertise without the substance.

There’s a perverse irony underneath this. Frontier models were trained on mountains of high-quality professional writing: websites, magazines, books. Plenty of my own writing found its way onto the internet back when there were gatekeepers, which means it’s presumably been assimilated into the AI hive mind. That professionalism is precisely why my writing superficially resembles AI output, leading clients to conclude there’s no difference between the words of a model and the words of an expert.

And that conclusion is entirely understandable. Clients have absorbed the productivity claims AI companies have been making, and it may require some painful trial and error learning for them to discern between smart-sounding slop and actual expertise. In the meantime, your job is not to convince clients that AI is incompetent: It’s to educate them on where general AI competence ends and your own defensible expertise begins.

What Work Is Gone, and What Isn’t

Some of the work consultants historically billed clients for was mechanical: rough drafts, first-pass research, summarization, initial document review. Now, with AI, cost and turnaround expectations for that work should move, and firms pretending otherwise are in trouble.

Clients are already using AI to handle their own grind-it-out work in minutes at near-zero cost, and they’re transitioning faster than their vendors. The same Thomson Reuters survey found 47% of corporate legal teams are already using GenAI, against 41% of law firms.

So, if the self-service floor for mechanical work is rising, what will remain?

• Mechanical production has largely been commoditized: rough drafts, first-pass research, summarization, routine review: this work is either gone or on its way out.
• Expert judgment applied to AI output has not disappeared, but the production model has changed. Assessing and correcting machine output is real work, and a buyer’s inability to perform that review themselves will still justify a consultant’s price tag. For example, my company recently negotiated an agreement where both sides were using AI agents to propose and negotiate terms, then each side independently verified the result with their attorneys, to confirm the language meant what we thought it did.
• True niche expertise is defensible, but not defensible forever in its current service form. If your firm possesses knowledge that generic models don’t, that creates an advantage. But once that knowledge can be encoded into a reliable system, someone will eventually do so. This creates an opportunity for a firm to become the go-to AI solution provider within their niche, provided they can create and market a quality solution before a rival or a software vendor does.
• Certain work will remain fundamentally human-led because of physical judgment, observation, liability, or catastrophic downside. You can still use AI to assist, but today’s models cannot do the core task or legally sign off on the final review.

How to Position AI-Accelerated Services

While the best consulting relationships are collaborative, collaboration does not mean giving the client a vote or a veto on every aspect of how the work gets produced. And that’s just as true of AI-assisted work as of unassisted human work in the past.

Clients are entitled to set requirements around outcomes, confidentiality, security, acceptable risk, and sometimes the use of particular technologies. They are also entitled to understand how AI materially affects the service they are buying. But there is a difference between transparency and surrendering control of your operating model.

According to the Thomson Reuters survey, 40% of firms have been told both to use and not to use AI depending on the client, and we’d encourage consultants to push back on both ends. If AI is now part of how your firm delivers better work, faster or more consistently, you shouldn’t feel the need to apologize for using it. And you shouldn’t invite clients to dictate your human-machine staffing mix.

Instead, you need to explain where AI fits, address legitimate objections, and price the service in a way that keeps everyone focused on the value and quality of the result.

Own the Human-Machine Mix

About a year ago my father needed shoulder surgery. When the surgeon said he’d be using a surgical robot to make the incisions, my father was nervous. The surgeon’s answer: “For this procedure, you do not want me doing it manually.”

The doctor explained how the robot improved the success rate, but otherwise didn’t invite negotiation regarding the toolset. In the end, the result (in terms of restored shoulder mobility) exceeded my father’s expectations.

That’s the move for AI-accelerated consulting firms. Not “Is it okay if we use AI?” but rather “here’s how we use AI as part of this particular service and why”, then take full responsibility for the quality of the result.

Engage With Objections

The flip side of clients who expect faster and cheaper because of AI are clients who prohibit it outright.

A professional association client of ours initially refused any AI-generated imagery in their training materials. This same client had complained for years that stock photograph libraries rarely depicted the work of their profession correctly.

We showed them how AI image generation could finally deliver authentic representations of the environments where their members worked and the tasks they performed, then asked what specific objection they had.

Their actual concern wasn’t a fundamental objection to the technology. It was a potential violation of intellectual property, which happens to be the area of law I studied. We showed them how our generation and review process ensured no violations of other parties’ IP, and they approved. They soon became quite enchanted by AI’s ability to finally get images of their profession right.

Sometimes a client’s opposition is categorical. But most blanket AI prohibitions are proxies for more specific, addressable concerns like “no copyright risk,” “don’t expose our data,” or “don’t give us obviously synthetic garbage.” Once you uncover the actual concern, you can solve the real problem.

Price the Value

Many practice management gurus encourage consultants to adopt value-based (or volume-based) pricing instead of hourly billings, and it’s a strategy that has always worked well for our practice.

By setting a fixed fee for a predefined result with milestone based review gates, our team is incentivized to work as quickly as possible, short of sacrificing quality, while clients can plug a fixed number into their budget, without the need for the client to monitor or audit our team’s activity.

This offers a solution for how to price AI-enabled work, so long as all parties can rationally weigh the value of the services delivered against alternatives, with two of those alternatives being “hire a competing vendor” or “DIY with AI.”

If the value of your service was concentrated in things AI now does well without you, then the value has changed and the price should follow. Any analysis that concludes “hold firm” in the face of a viable do-it-yourself AI alternative is delusional.

But if a client needs a $120,000 solution to a $12 million problem, and AI alone will not produce that solution, there is no reason to move.

Either way, reasonable flat rates make the human / machine mix your business, rendering the question of “how much of this did AI do?” effectively moot.

This is where the AI-enabled attorney we worked with went wrong. By billing discounted AI hours as a line item alongside the human hours required to clean up AI mistakes, he converted “AI is saving you money” into “you are being double billed for unusable machine output that a human then had to redo at additional cost” – even if the total came out below comparable unassisted work.

Building an AI-Enabled Practice That Can Actually Deliver

Many of our workforce training clients are organizations experiencing rapid growth, creating a need for more formalized training programs. Often, their leaders will wax nostalgic about when the organization was smaller and everyone knew everyone else by name, or worry that growth will undermine the operating model and culture that made them successful.

My standard response is: “You are already becoming a larger company. The only question is what kind of large company you want to be.”

The same advice applies to consulting and professional-service firms in fields being reshaped by AI. Client expectations have already moved. Competitors’ cost floors have already dropped. Asking “Should we adopt AI?” is nostalgia dressed as strategy.

The more useful questions are how deeply AI should enter your practice, what you need to change once it does, and what capabilities you actually intend to own.

Decide How Deep You’re Going With AI

Right now, many organizations are asking their incumbent consultants for AI strategy advice simply because they don’t know where else to turn. Naturally, every consultant would like to be the client’s guide into the AI age. For many firms, though, that ambition begins and ends with a new page on the website and a deck of generic strategy guidance the client could get for free from a chatbot.

Most firms operating at this level seem to recognize the disconnect. A Teamwork.com survey found 39% of client-service firms admit exaggerating their AI capabilities to clients, while 52% say they feel ill-equipped to advise clients on AI at all.

You do not need to become an AI development company to remain valuable. But you do need to decide how far down the stack you intend to go.

At one end, a firm may simply be AI-literate: able to advise clients on how developments in AI affect its own domain. Further along, a firm may use commercial AI tools to accelerate its own work. Go deeper and the firm may begin designing AI-enabled workflows and configuring systems around its methodology, or building client-facing tools. At the far end, it may take responsibility for integrating, operating, and maintaining substantial AI systems.

Those are different businesses requiring different capabilities.

The danger is that AI makes it unusually easy to do a half-assed job, and mistake experimentation for expertise. Traditional software generally stops working when you feed it nonsense. AI will produce a coherent response anyway, talking back as if it understands your intent (which it sometimes does.) But this superficial fluency can mask deeper issues with the AI system’s output / task performance that can continue silently, undetected for years.

The same landmines show up repeatedly:

• Reliability at scale. A workflow that behaves beautifully in a partner’s hands during a demo may behave very differently across forty employees and ten thousand uses.
• Standards versus user instructions. An agent that does whatever the user asks is not the same as an agent that does what should be done. Encoding (and enforcing) predefined workflows without overly constraining the system constitutes much of the real work.
• Knowledge architecture. Your terminology, methodology, decision rules, exceptions, and relationships between concepts need to be represented deliberately. This is unglamorous work, and it is also the piece that has to come from your domain expertise.
• Data and access controls. Once AI systems touch client information or operating systems, security, permissions, confidentiality, and governance stop being somebody else’s problem.

A firm doesn’t necessarily need the ability to build all of those, but you should at minimum know the difference between using an off-the-shelf tool, configuring an existing enterprise platform (like Microsoft Copilot Studio), or commissioning a custom application. There are $400 / month AI solutions, $40,000 solutions, and $400,000 solutions and recognizing the factors that drive scope (number of users, complexity of tasks, volume of operations, security concerns, integrations, extent of knowledge management work required) is critical.

You should also know what kind of talent each requires. “We need an AI person” is no more useful a hiring specification than “We need a computer person.” I’m currently talking with a scientific research organization that hired PhD-level machine-learning specialists and software developers, built a platform, and neglected the knowledge layer almost entirely. The system responded to everything. There was no crash, no outage, and no obvious defect anyone could point to. But the answers were often wrong before because everybody was focused on model and platform performance and nobody was making sure the system was being fed the correct inputs at the correct times.

Redesign the Practice, Not Just the Tasks

The most consequential effects of AI do not come from making individual tasks faster. They come from changing the economics and structure of the process around those tasks.

• If an AI system equipped with your playbook and evaluation criteria can guide a junior employee through an analysis that previously required a senior consultant, with no appreciable loss of quality, is that now your standard operating procedure?
• If a lead consultant becomes productive enough that completing a task themselves costs less than explaining it to an associate, do they still need the associate?
• If a well-designed system can perform part of your service without human labor at all, do you continue staffing that work out of habit?
• And if all of this changes the actual cost of delivery, should it change the assumptions your firm uses to estimate projects, allocate staff, and set prices?

These are not questions about software features. They are questions about your operating model. The firms that get the most out of AI will not merely bolt tools onto existing processes. They will revisit where expertise is required, where review belongs, which decisions need a person, which activities can be delegated to a system, and which work should disappear altogether.

That can be uncomfortable because professional-service firms have traditionally organized themselves around the availability and billing rates of people. AI introduces another form of capacity whose marginal cost, scalability, and skill profile look nothing like a junior analyst or associate. Eventually, the org chart has to acknowledge that.

Rethink the Talent Pipeline

There is another problem hiding inside those productivity gains: uch of the work AI automates first is the work junior professionals historically performed while learning how to become senior professionals.

First-year lawyers reviewed documents before they negotiated major agreements. Junior analysts conducted research and built first-pass models before they advised executives. Associates prepared analyses, sat in on client meetings, made mistakes, had their work marked up, and gradually learned what experienced practitioners notice that novices do not.

If AI removes that work, firms may save a great deal of money in the short term while quietly dismantling the apprenticeship system that produced their future experts.

At the same time, AI can make your existing experts dramatically more productive. If a senior consultant can suddenly do 1.5, two, or four times as much high-value work with AI assistance, that looks wonderful on a utilization spreadsheet. But it also increases the firm’s dependence on that person. You have concentrated more revenue, client knowledge, judgment, and delivery capacity in fewer people.

That makes retention even more critical. If your best people are generating a substantial “AI dividend” for the firm because technology has multiplied their output, you might want to cut them in on some of that upside in order to keep them. Otherwise the firm captures all of the productivity gain while asking its most valuable people to carry a larger share of the business, which is not an especially durable bargain.

The temptation, meanwhile, will be to stop hiring and developing juniors because the senior team can now handle more work with fewer people. That may improve margins this year while leaving you without a bench five or ten years from now.

Law firms are already wrestling with this problem around junior associates, but it applies well beyond law. If the machine handles the easy cases, newcomers may encounter only the difficult ones, without having accumulated the experience that once prepared them for those situations.

The answer is not to preserve low-value work purely as a training exercise. It is to become much more deliberate about how judgment is developed.

That may mean structured apprenticeship, greater client exposure earlier in a career, simulation, guided review of AI-generated work, deliberate practice around edge cases, or systems designed not merely to give employees an answer but to expose the reasoning, standards, and decision points an expert would apply.

AI can eliminate a rung from the career ladder while making the people already standing near the top much more productive and much harder to replace. A sustainable firm has to solve both problems: share enough of the productivity gains to retain its strongest people, while continuing to invest in the bench that will eventually replace them.

Decide What Could (or Should) Become a Product

Once you have encoded a process well enough that an AI system can execute substantial portions of it reliably, you face another question: should your consultants continue operating that system on the client’s behalf, or should the client be able to use it directly?

The obvious trap here is what I think of as the “chatbot-ification of expertise”: upload your playbook, methodology, or training manual as a PDF, put a chat interface in front of it, and declare that you have turned your expertise into a product.

But a chatbot that can answer questions about your methodology is not the same thing as a system that can apply your methodology. The real value of expert practice is rarely contained entirely in the source material. It lives in the sequence of questions you ask, the information you gather, the standards you apply, the distinctions you make, the exceptions you recognize, the decisions you refuse to leave to the client, and the actions that follow from the analysis. If the AI product merely waits for users to ask good questions, it may be transferring the hardest part of the expert’s job to the person least equipped to perform it.

A useful AI product therefore needs to encode more than knowledge. It may need to guide a process, collect structured information, apply decision rules, challenge bad assumptions, evaluate work against explicit criteria, interact with client data, track activity over time, and know when a situation should be handed back to a human expert. In other words, the goal is not to build a chatbot that has read your methodology. It is to build a system that can do something useful with it.

There is also the question of enterprise readiness. A prototype that impresses a partner is not necessarily a system a major client can buy. Once AI touches the client’s data, employees, workflows, or systems, you enter the less glamorous territory of security reviews, confidentiality requirements, access controls, procurement, approved vendors, auditability, data retention, and sometimes regulatory requirements. This is another reason firms need to be clear about how deeply they want to go with technology. Advising a client on how AI affects your field requires one capability set. Operating a system that handles confidential client data across thousands of users requires another.

Even if you develop (or contract) the technical capability, productizing a service can still feel threatening because it may cannibalize billable work. Sometimes, however, it is the only way to keep a client who would otherwise build their own AI solution or buy a rival’s alternative. And on the positive side, an AI product can generate license income between human touchpoints, reach new audiences within existing client organizations, and serve clients who could never afford or logistically accommodate the traditional offer.

One of our clients, a sales training firm, invested tens of thousands of dollars developing an AI coach that helps their clients’ staff, primarily within financial-services institutions, develop and execute client-acquisition strategies while tracking accountability for day-to-day activity. It is not a chatbot dispensing sales advice: it is their methodology encoded into a system, gathering information about the salesperson’s actual pipeline, guiding the planning process, applying the firm’s standards, tracking execution, and supporting employees between live coaching sessions.

When they brought it to a Global 50 financial-services organization, the buyer told them they were the only consultancy in their niche that had shown up with something real rather than another PowerPoint deck.

That is increasingly the opportunity for firms with genuine niche expertise. Your proprietary methodology may remain highly defensible even as generic professional knowledge becomes a commodity. But preserving that advantage means resisting the allure of simply putting a chat window in front of your intellectual property and calling it a product.

While not every aspect of your consulting practice should become a software product, the decision whether or not to productize should be made deliberately rather than assume that every valuable piece of expertise must continue to be sold one expert-hour at a time, or that every body of expertise becomes a product the moment you give a chatbot a PDF.

Don’t Get Stuck on the Cleanup Crew

One of my favorite jokes has Napoleon calling a consultant: “My army just invaded Russia. We believe our strategy and processes could be improved. What can you help us do with the soldiers we have left?”

That is where a lot of clients are about to find themselves with AI: budgets spent, schedules blown, systems deployed, and results that are nowhere near what was promised.

Cleaning up the output of poorly designed AI systems will be a substantial near-term source of consulting revenue. In many cases it will also be an easier sell than convincing a client to pay a premium for an AI-enabled service or product in the first place. On a rescue job, the failure is visible, the sunk cost is painful, and someone senior is looking for a person who can fix it.

But there is a danger in becoming too good at cleanup.

I spent years as the designated rescue person at a consulting firm, parachuting into projects that were behind schedule, over budget, or about to lose the client. It earned me several heartfelt thank-you emails from the CEO, but it never earned me a project to lead from the beginning: I had made myself too useful in the fix-it role.

So by all means take the AI rescue work, there will be plenty of it. But whenever possible, write the forward-looking redesign into the engagement as well: what the client should build instead, how the process should work, what role your firm should play, and what has to change so you are not having the same conversation again in eighteen months.

Some clients will refuse. That’s useful information too. The goal is not to become the firm clients call after the AI project fails, it is to become the firm they should have called before they started.

Conclusion

Make no mistake: nearly every expert consulting firm is now in a race against the clock for transitioning to AI-enabled delivery or productization. However, the clock that matters is client-driven, not technology-driven.

While some niches will cross the threshold later than others, once a credible provider demonstrates that work that once required three weeks and five consultants can now be delivered in three days by two, expectations will be reset forever. Before then, firms have some latitude to define what good work should cost and how it should be delivered. Afterward, they are arguing against a demonstrated alternative.

The answer is not to race toward the cheapest possible delivery model, nor to tell your team to “use Claude” or hastily bolt a chatbot onto your website. It is to get very clear about where your expertise still creates value, where AI genuinely changes the economics of the work, and what combination of people, systems, knowledge, and judgment produces a result clients cannot easily replicate on their own.

For some firms, that will mean using AI quietly behind the scenes to make experienced professionals more productive. For others, it will mean redesigning workflows, changing staffing models, or turning proprietary expertise into systems clients can use directly. For most, it will mean some combination of all three.

Eventually, “AI-enabled professional services” will sound as redundant as “computer-enabled professional services.” The interesting question will be whether the firm uses the available tools well enough to deliver something distinctive, reliable, and worth paying for.

Until then, there is an advantage in getting there before your clients have to ask.