5 minutes; 2 minutes for Overview and bold text.
Overview
This is the third in a series of posts about AI/LLMs at DISC. Here are the first and second.
8/16/24 Advisory: See DISC’s service, Deploy Your LLM Professional “GPT.”
This 3.5 minute video, How To Create Custom GPTs For Beginners, outlines steps in making GPTs for most professional services (details may differ now).
(DISC’s clients have read the rest of this post in their 2024 plans, except for the final ORRR section.)
Imagine hiring a dedicated professional for each main area of web marketing, each of whom comprehends ~95% of her profession (more than even the best human pros), recalls and applies all your specific instructions and general guidelines, produces at 100x human time, and the total cost of all these professionals is $20/mo. That’s AI/LLM properly set up. Human experts are still needed to:
- carefully train these new AI/LLM “employees” with web marketing rules and sources;
- select and input the firm’s files (website sections, internal sales and profit data, etc.);
- review output quality, and tune the prior two processes accordingly.
The first two tasks comprise the bulk of ~12 months labor. Subsequent years mean that web marketing firms will be needed for ~1/2 the time of prior years, with that time probably declining yearly. Integrating the separate main fields of all AI/LLM-powered web marketing (e.g., SEO, PPC) with clients’ internal and market changes will comprise the main role of web marketing firms in 2025+.
Contrary to quick intuition, the most precisely rule-governed processes are last to be handled well by AI/LLMs. In web marketing, this order of AI/LLM tractability is: (1) content-level SEO, (2) CRO and ROI reporting, (3) PPC, and (4) website and server management, programming, and technical SEO.
Passing Savings to Clients
Web marketing firms that are truly focused on their clients’ interests will start now to set up systems that enable AI/LLM to greatly reduce clients’ costs in 2025 and after. Good web marketing firms that use AI/LLM will be able to serve more clients to compensate for less revenue per client. For at least 5 years, domain experts (in marketing, law, accounting, most professions) will remain crucial in orchestrating AI/LLM efficiencies and results. But the stark fact is that much fewer of those pros will be employable–estimates range from 40% to 80% reduction over 5 years. Proficiency with AI/LLM systems will determine success for most businesses and organizations.
The Two Main Parts of AI/LLM Systems
ROI reporting systems track quarterly OKRs while the following two AI/LLM systems are implemented to produce years of high ROI.
- General preparation for AI/LLM assistants that will increasingly handle middle and upper management functions: This includes preparing firms’ cloud file systems’ for AI/LLM assistants. Such systems align with decades of best practices in networked file management and even in library science, so are beneficial regardless of AI/LLM. Google and Microsoft AI/LLM assistants or standalone LLMs like ChatGPT with plugins comprise the primary commercial goal of venture capitalists’ LLM investments. AI/LLM assistants serve to expedite and amplify human expertise as well as operations management. This setup will help achieve web marketing goals and will also make firms well disposed to AI/LLM’s help in all they do.
- Careful deployment of specialized AI/LLMs to amplify and expedite human expertise in each field of web marketing: These fields are integrated under the decades established model of target markets and buyer journeys. Google’s multimedia MUM & Entity models replace the old keyword model (illustrated in Google’s recent–and somewhat deceptive–Gemini announcement). LLMs properly setup can handle the specialized details of each service area while keeping in “mind” those two sets of models. The order of main web marketing services to first benefit most from AI/LLM is: SEO, CRO, PPC, and lastly ROI Integrations.
ORRR
At DISC “ORRR” refers to the Objections, Rebuttals, Replies Rule. We ask our workers to process salient threads of ORRR before proposing system changes.
Won’t rapid developments in AI/LLM undermine current setups?
As in most tech fields, sufficient understanding of the technology shows the arc of future change.
Years ago Google introduced BERT (a change in organic ranking based on GPT-like Transformers) and the MUM & Entity models, both of which were predictable antecedents to the AI/LLM revolution. Google’s Gemini LLM will operate on the same foundational tech, meaning that the two main parts of AI/LLM systems described above could be ported efficiently to a competing platform if ROI analysis justifies the move. Such a move is akin to migrating a website to a new WordPress theme or a new CMS.
Before Open AI released custom GPTs, tons of third-party plugins were hawked, most of which merely wrapped ChatGPT in pre-written verbal “prompts.” Sufficiently informed people knew that most of these would be rendered obsolete by OpenAI’s arc of change, and that’s exactly what happened.
Of course GPTs will need periodic tuning, which comprises the bulk of what professionals will do in 2025+.
Aren’t AI/LLM “Token limits” too small to succeed?
AI/LLM Tokens are units of meaning equivalent to a single word. Current ~100,000 token limits are sufficient for most web marketing GPTs. For context, the novel The Adventures of Huckleberry Finn is ~82,000 words (tokens). These limits will grow–in 2023 they more than tripled. Even if token limits expand well beyond most firm’s input-output needs per service, experts should still strive for economy of inputs or “eloquence.” For example, which sections of Google’s 3000 pages of SEO help should be selected as inputs guiding work in an SEO GPT?
True, AI/LLM assistants for Google WorkSpace or Microsoft 365 must encompass much more than current and near future token limits. However, the primary and heavily-invested goal of such giants is to make AI/LLM assistants that handle firms’ total cloud and workflow. Several methods work now, and the frenzied pace of improvements will accelerate. Therefore, good hygiene in clouds’ information architecture will poise firms to benefit now and in the future.
Wont AI/LLM’s cloud access risk security and privacy?
You use cloud storage now, right? Risks here are essentially the same. Most businesses accept such risk in exchange for the enormous efficiencies of cloud storage and workflow. As for private clouds on a firm’s own servers, consider how many of those have been hacked–like at the FBI, NSA, and too many Fortune 500s. The likes of Google and Microsoft know that their bread and butter depends on security; their massive economies of scale produce far better security than most firms can engineer themselves.
Why not wait for web marketing GPTs for sale on the ChatGPT store or equivalent?
Waiting may make sense in some professions standardized and regulated for decades, like law and accounting, but in web marketing there are few top minds from whom I might buy a pre-made GPT. One would still have to adjust the generic GPT to what matters most for a given firm, then select and upload documents, vet the output, and tune. Good web marketing firms will stay apprised of worthy enhancements to roll-out quarterly or yearly. Given how few web marketing firms are owning up to the truth of AI/LLMs impact–just as most did not admit that the MUM & Entity models replace the old keyword paradigm–I question whether a GPT store will ever have good SEO GPTs. If any come to exist, finding them in the vast rubble of mediocrity will require the very expertise one seeks. Lastly and most importantly, truly good web marketing firms whose GPTs I would trust probably wont sell their core intellectual capital for the pittance of typical store revenue.
Please don’t hesitate to contact DISC with your concerns, questions, or suggestions.
(No LLM was used for writing, only for a quick proofreading.)
