The End of the Agency Model - Meet AI-First Innovation

Every step-by-step professional services process will be automated using AI. The question is not if but when the quality will be high enough to replace humans. My goal here is to outline what the AI Revolution means for the innovation function in large companies and the ecosystem around them.

The Industrial Revolution replaced muscle power with Mechanical power. The AI Revolution is replacing Brain Power with Computer power. LLMs are like small coin-operated brains that can perform tasks only humans could before..if you ask nicely.

AI is beating human baselines in some tasks, but not all of them yet. This will eventually be solved as the foundational models become more and more intelligent, but now, you can replace a lot of the intelligence with careful orchestration and data pipelines and achieve superhuman results.

On May 30th, 2023, U+ launched just such an orchestrated system—fifthrow.com —and after 11 months of learning, real-world deployment, and customer feedback, we are steering away from the agency/consulting model and toward AI-first innovation.

The future of the corporate innovation team

For the longest time, the default corporate innovation setup has been a shared service team charged with various tasks, from open innovation to venture building. The team is sometimes connected to the overall strategy and business unit needs and sometimes more autonomous.

The only real difference has been at which point of the process a handoff to a business unit happens. If that business unit hadn’t been involved in the innovation process, the inevitable “death by BU” would happen, and innovation would stall because it was too small, too big of a distraction given quarterly KPIs, and too big of a risk for one’s career.

The macro situation in the past two years forced the divestiture of Horizon 2+ activities seen as cost centers in an environment where showing efficiency for the stock price has been paramount. In parallel, ChatGPT came out, and GPT4 followed soon after.?

Many consultants lost their jobs between 2022 and 2024, then some were hired by corporations to lead innovation in-house – but without outside support and a vague promise of GenAI solving the problem.

While there is plenty of snake oil salesmanship, here’s a proposition on the takeoff of AI-first innovation:

• Phase 0 – Shared service innovation team enabled by consultants
• Phase 0.2 – Consultants enabled by AI preserving the status quo <– WE ARE HERE (on average)
• Phase 1 – Internal innovation teams enabled by Vertical AI platforms
• Phase 2 – BUs enabled by Vertical AI with the support of innovation team members spread in the BUs
• Phase 3 – Autonomous Innovation – Horizon 1+ enabled by vertical AIs that run Autonomously

The future is here, just unevenly distributed – right now, there is a temporary revenue opportunity exploited by the consulting companies being more efficient but charging the same. It’s quite obvious that if you just become a translator of what an AI says, your utility comes into question.

I have been quite outspoken about the consulting business model being an ungrateful way to make money predicated upon the inefficiency of large companies’ ability to innovate. AI is already changing that as we move to Phase 1 of the AI-first innovation transition and enabling the internal teams to follow the standardized rigorous innovation process by themselves.

From the deployment of FifthRow, we see that within about two months of the innovation team becoming superstars by having answers to almost anything a BU might need, the BUs start being interested in having the capability. As fear of the unknown is transformed into unquestionable results, barriers drop.

This will lead to changes in org charts, not only in innovation, as other vertical AI platforms are introduced in the rest of the business. This will lead to dispersing the innovation function into the BUs (which has never really worked before but now has a chance with GenAI) and innovation becoming a core internal capability.

The knowledge ladder of applying GenAI in real-world business applications

To dispel some of the confusion and snake oil salesmanship, here’s a quick glossary:

• Prompt Engineering – Optimizing for the output you want without changing the model
• Fine-tuning – Re-training the model to give you outputs in the format you want
• RAG (retrieval-augmented-generation)/embeddings/vector databases – Providing the model with additional knowledge it was not trained on, aka “grounding” the model
• Agents – The model prompts itself, keeps memory, and can use tools (e.g., browsing, coding)

Sidenote: Most startups that cry “Open AI killed my startup” did some careful/complex prompt engineering that has been subsumed by GPTs or more intelligent models that do what you want out of the box.

So far, the typical implementation ladder has gone like this:

1. Let’s try prompts – it does what I want
2. Let’s try more complex prompts – I’m really surprised by what it can do, but to build something more general, the output is inconsistent, repetitive, and shallow
3. Let’s add my own data (RAG) + complex prompts – now I have something that gives me unique value, but I need to ask a bunch of questions manually to get to the real value
4. Can an Agent do this? – yes, but it is quite expensive and slow
5. Orchestration – can I offload the logic to a database? Or improve my retrieval from RAG? Do I need to pre-process my data differently to achieve better results

RAG + Agents + Orchestration and overall optimization of this system allows for automating any business task provided by:

• Having a technical team who understands the business value you are trying to automate and can think outside of the box of typical system distinction (backend vs. frontend vs. DevOps)
• Having a team of subject matter experts on the topic you are trying to automate to provide feedback every step of the way

It is easy to create a prototype, but it’s hard to create a production-level trustworthy system.

I strongly believe this turns business value creation away from managing people to an engineering challenge of automating existing processes while utilizing the tribal knowledge inside the organization.

This will favor generalists on both the business and tech sides, whose primary north star is an increase in performance/quality in any task for the end customer (could be an internal customer). The easiest tasks to start with are those with high frequency and low required fidelity because these will require only a smaller amount of tweaking to get them right.

Where is U+ and FifthRow going

With almost a year of production-level GenAI development and deployment in Fortune 500 companies, we are delivering on our mission of Increasing the ROI of Innovation. We also believe that we can deliver superhuman innovation performance with continuous development and as the field matures.

With that, we are placing several bets:

• U+ is focusing AI-First venture building = we will continue our consulting practice to deliver as much as we can with FifthRow and amend teams that do not have in-house capabilities to deliver innovation ROI
• We are launching FifthRow Continuum and Flows – more below
• We are standing up our own Venture Studio powered by FifthRow Continuum with a separate pool of capital to invest into ventures identified primarily by FifthRow
• We are raising our hand for any GenAI projects that focus on automating business value – we have world-class no BS experience deploying the technology at scale successfully

FifthRow Flows

Flows is an agent orchestration platform that allows us to chunk any step-by-step task and ground it in private data or online search. It is an evolution of our current platform that allows for bit-sized super custom workflows that we will aggressively automate. Anything from Strategic Foresight to Horizon 1 innovation first.

E.g., it took me 1 minute to set up a flow to analyze some of my claims in this piece using a web-connected agent (this flow will now be available to any subscriber of FifthRow)

FifthRow Continuum

With Flows and the existing platform, we have been able to build the first Autonomous AI Venture Studio —a platform that makes decisions on the next topics to pursue, conducts market research, ideates, ranks all solutions, creates synthetic personas, simulates interviews, finds incumbents, prepares marketing assets, builds a landing page, and launches a market test. Continuum moves at superhuman speed and is the foundational technology for our Studio.

We will continue scaling our impact on the ROI of Innovation, our Flows library and expansion of the existing platform with an additional core product launch coming around Summer 2024.

Happy to chat about GenAI projects, FifthRow, and U+ Studio – book a meeting here