This year’s explosive proliferation of AI tools on the market has opened a lot of doors on the path toward automation and efficiency – two objects always on the minds of agency owners. While the abundance of options has greatly increased in the last several months, the path to actually incorporating these tools effectively into professional service operations remains murky, with a number of both tactical and strategic questions to be considered.
It’s fascinating how, almost overnight, about 80% of the emails, LinkedIn messages, and other business solicitations I receive are focused on AI based tools to solve whatever problems plague my days. Outreach from AI companies has finally eclipsed the number of offers I receive for outsourced SEO and WordPress development — an amazing feat! As an agency owner, one of the really fascinating aspects of this change is that it’s happened faster than the industry has been able to adapt. The consensus view seems to be that AI based tools will provide value and can transform parts of our businesses, but how and when and where these transitions occur is still really ambiguous. Related to this fast pace of change is that the infrastructure that surrounds traditional professional services engagements and workflows isn’t necessarily equipped to embrace AI tools. The systems and internal processes an agency uses to execute client work, the legal arrangements that drive client outcomes and agency approach, and the formalization of tools and training for teams, are lagging well behind the rapidly evolving capabilities of the tools.
Through that lens, here are three specific areas that GRAYBOX is thinking through for ourselves and our clients: The legal questions raised by the use of AI, how our Teams are adopting AI tools, and how we, as agencies, can work to incorporate AI tools more formally into our process flows.
As a former attorney whose only actual (and fleeting) interest in law was intellectual property, the big questions AI raised for me were less about evolving how we work or the potential commodification of certain services, and instead about how eventual court battles will play out with regard to ownership of AI-generated outputs. Given my background, I’m often first tier legal review for our contracts. In at least 80% of contract negotiations, and nearly 100% for our larger enterprise clients, Intellectual Property ownership is a large part of our conversations. The question of who will own the finished product is incredibly important. Here’s a fun scenario...
A lot of intellectual property rights in our space boil down to copyright laws. Whether it’s written content for a website, or written code for an application, copyright law tends to be a central driver. Let’s assume an article is being written on a really specific topic — the geologic processes behind the soil and weather combination that make the Dundee Hills AVA in the Willamette Valley (Oregon) exceptional for growing grapes for wine. While there are lots of differing online reviews of the wineries, many promotional websites touting the area, etc. there are a pretty limited number of scientific answers to the question of geography and soil. Let’s say ChatGPT is used by two independent journalists who both had similar ideas for this article, and use similar prompts. There’s a not-insignificant chance that the AI-generated outputs are a close, or exact, match. If that happens, which publication owns the intellectual property rights between two identical (or nearly identical) articles?
Lawmakers are asking an even more fundamental question: is it legally possible for intellectual property rights in AI-generated outputs be assigned to anyone? In the U.S., a human must be the inventor of a patentable invention, or the author of a copyrightable work. AI isn’t a human, so there’s a building consensus among lawmakers that AI generated outputs cannot be assigned to anyone. Despite AI companies saying they will assign ownership rights to users, those rights may not be theirs to assign because the output (ie: our Dundee Hills article from above) wasn’t written by a human, it was written by a model trained on various sources of data and leveraging various learning algorithms to output what it perceives to be relevant responses to a prompt. If a human was not the author of the copyrightable work, it either may not be copyrightable in the first place, or those rights may not be assignable to anyone for ownership (ie: nobody can enforce a claim to ownership).
This example is relatively straight forward when compared to the potentially more complex web of code repositories and license restrictions that come into play with AI tools that help developers write code. I’ve seen some examples via Copilot where the returned output is ~100 lines of code, is an exact copy of code from another source, and makes no mention of the license restrictions to which that code is subject. Delivering that code to a client with a claim of “we wrote this, we own the rights in it, and we’re assigning those rights to you” may be very, very far from the truth.
For enterprise clients, startups, and many others, ambiguity in the ownership of their key intellectual property assets is often a non-starter. As such, the question of how professional services firms both incorporate the use of AI tools to build efficiency, while also satisfying their clients’ desire to be the undisputed owners of their IP, will be a tricky one. Likely it’s an answer we’ll struggle to gain clarity on for several years to come.
If you’re a professional services agency, and you believe your team is not already leveraging AI tools in some capacity, you’re almost certainly wrong (lawyer-Jon dislikes committing to absolutes like “you’re definitely wrong” and hedges with equivocating word selection instead. He also apparently writes in the 3rd person occasionally). Rather than being concerned about being replaced by AI tools, about 60% of employees say they are excited about the use of AI in their jobs. When you think about it, this makes a lot of sense given where the tools are today: They’re not yet in the “replace humans” level of quality, so they’re not professionally threatening. But they’re definitely in the “help humans shed boring tasks” level of quality, so they’re very appealing. Given that many tools are still free, and unburden professional services employees from the rote and the mundane, it’s not surprising to see excitement about their adoption.
Given the constant barrage of professional services staff by AI tool providers, firms should take a few tangible steps to stay ahead of (or at least keep pace with) this rapidly evolving space.
- Define clear policies for your employees: In all likelihood, your employees are not thinking about the intellectual property implications of their use of Copilot or Bard; they’re thinking about how much time the tools are going to save them. It’s the responsibility of the agency to have an opinion and clear policies for how and when AI tools can be used. Draw the line where you are comfortable: maybe internal initiatives are acceptable, but client deliverables are off limits. Maybe use of AI tools is ok for clients, but only for engagements where clients have specifically opted-in to new contract language. Craft policies that both protect the interests of the agency, but allow your staff the flexibility to be nimble in their experimentation and adoption of new tools that can improve their productivity.
- Develop a framework: Specific, targeted policies will not keep pace with the evolution of new tools (weird, lawyer-Jon used a very declarative phrase. He must really mean it!) Your agency’s philosophical approach to evaluation and adoption of new AI tools must be sufficiently flexible to accommodate many types of inputs and outputs, by different categories of users, in use cases you have not yet considered. Rather than building a list of approved AI tools, build a system that supports the discussion, evaluation, and, if applicable, incorporation or deprecation of AI tools into your workflows.
- Button up the legal aspects: As a follow-on to the Legal Questions discussion above, it’s important that agency contracts have incorporated a stance, and appropriate language, on the use of AI outputs in their client deliverables. Previous wording such as “our work is our own” may no longer be true. If your current IP language states that you assign all rights to the Client, you may no longer be entitled to do that. Engage your attorney to evaluate your current contract language and build a legal framework that allows you to adapt to keep pace with the ecosystem. For agencies like GRAYBOX, that includes a broad and flexible framework that extends beyond just code to also include copy, UX and designs, graphic elements, and all other components of digital project execution.
How do we incorporate AI into our businesses?
Ironically, I didn’t write this article by typing a prompt into Copy.ai, ChatGPT or Bard. I didn’t even use it as a starting point. It’s a telling indicator that I’m writing a piece on how AI will be incorporated into the professional services world, and my go-to approach was to personally craft sentences to convey my thoughts. That approach is rooted partly in age (I’m 43), partly in that former lawyer mind of mine (precision of communication is really important to me), and, most importantly, in the question of “How and when can we effectively start to leverage these tools in our day-to-day workflows?”
Despite the bevy of AI options newly on the scene, businessesare still struggling to effectively and consistently incorporate them into their workflows. There are a lot of interesting use cases and impressive proofs of concept that individual employees and users find for their day-to-day. But for the business to adopt an approach that is consistent, applicable across its multiple scenarios, and sufficiently resilient to adapt, there has to be a material positive impact to make the shift. Understanding where to look for opportunities and how to evaluate them for impact takes time, and agencies are still in the very early days of that exploration.
As an approach to starting the process of incorporating AI tools into your agency, a few things businesses can do include:
- Identify some use cases and start experimenting: Thus far, we’ve found that the more narrow the use case, the more effective the tool at building efficiency. The other side of that coin is that narrow use cases are only narrowly applicable. You improve only a small part of a larger process, or a single small process within the larger set of agency workflows. Efficiency is about more than simply “does it save me time on this task.” Instead, efficiency should be viewed for the agency engine as whole, and a tool’s impact to that engine’s overall performance. Look for that sweet spot of focus and impact that actually moves the needle on efficiency.
- Align use of tools with the importance of the output: The ultimate quality of some outputs matters a lot more than others. Current AI tools are not a panacea for all agency processes. They still require human intervention and review. There’s a really important question to ask when evaluating if some output of the agency is a good candidate for incorporating AI tools: How much do we care about the quality of that output? It’s possible that you just need something shipped, and 80% efficacy is more than fine. In those cases, a quick prompt and cursory overview of the output might be a great time saver. Alternatively, some use cases might relate to mission critical outputs. The thought, planning, organization, and execution may be so bespoke as to make the use of AI tools less efficient. Ultimately, evaluating the value and importance of the AI-generated output is a key step in determining where to incorporate AI tools into your processes.
- Evaluate your workflows: If you and your team step through your existing operations and workflows, and don’t find any areas that seem primed for testing AI tools to build efficiency, consider evaluating those workflows for some fundamental changes. Outside of the truly bleeding edge top of the professional services pyramid, there are almost always opportunities to push downward into the efficiency realm. If your current workflows don’t seem to support the inclusion of AI tools, there’s a high likelihood that there’s room for improvement in how you approach those things that would subsequently open the door to improvements through the incorporation of AI tools.
- Plan for evolution: The evolution of AI tools is not going to mimic ERP or CRM software deployment — don’t assume you’ll pick one tool and stick with it for 3-5 years. The ecosystem is moving too quickly, and will continue to do so. Don’t build a process flow around something too narrowly focused. Build a system and framework that is adaptable so that as the ecosystem improves, your processes can slot in the new latest and greatest.