The Return of Decision AI

It's been a busy month for AI labs. Both OpenAI and Anthropic have released new versions of their most powerful models. Meta's new personal AI assistant has attracted plenty of hype from the people who jumped on the OpenClaw bandwagon at the start of the year. Yet, the hottest new AI model sweeping Silicon Valley comes from a previously unknown startup founded only last year. Jev by TypeSafe AI is a new kind of model designed to solve one of the key challenges holding back agentic automation.

Decision Models

Jev is a System One Model. It doesn't generate text. It generates decisions. The input into a Jev query is the same as any other AI model: either structured data or unstructured text. The difference is the output; it returns only a pre-defined picklist value or a probability score. The valid output values from a Jev query are defined in advance, and training examples need to be provided in order to guide the model on the expected decision criteria. In that respect, it fills the same niche as a machine learning model by attempting to find patterns from arbitrary input.

One key difference is that Jev is trained on the same types of general purpose knowledge sources used to train LLMs. That means it can parse written information to extract relevant market and business context. This was the problem with traditional machine learning (ML). It didn't understand the data being shared with it. As such, every ML model had to be trained from scratch. Training datasets had to be hand crafted, and irrelevant data removed from the input in order to avoid skewing the results. Such data preparation is a laborious process, meaning that ML training is lengthy and time consuming. 

The Purpose of AI

More recently, developers have tried to turn LLMs into general purpose classification engines that understand business context. This is the starting point for many agentic workflows. Trouble is, all too often, agents serve up hallucinations instead of accurate outputs. TypeSafe AI claim that removing the language element of an LLM eliminates the risk of hallucinations. That's critical in driving AI adoption. The risk of unwanted results is a key reason why Generative AI is not delivering the ROI expected of it. The technology is just not designed for automated decision making.

The limitations of generative AI have allowed black box AI models from technology vendors such as 6sense or Demandbase to embed themselves into enterprise tech stacks. Plenty of leading martech tools are purchased for their embedded AI models, rather than platform capabilities. Many point solutions are used for scoring leads, tiering accounts or attributing opportunities. All these tech vendors are using bespoke ML models trained by teams of data scientists. They have the in-house expertise to categorise, organise and train the ML models powering their products. That does have its advantages. An embedded ML model is trained on far more data than any in house team could hope to gather, allowing platforms to fine tune their models across clients and scale the outputs across sectors. 

AI for Automation

Proprietary AI models aren't going away. They will remain vitally important for big data decisions such as marketing mix modelling or purchase intent calculation. Yet, black-box ML models don't help with the kinds of small scale business-critical decisions intended for custom AI agents. Whether it's for account classification, lead grading, or nurture stream routing, the training data in off-the-shelf models just isn't specific enough to each business's specific context. Jev overcomes this by allowing users to bring their own context and get a usable answer. 

In doing so, Jev opens up agentic decision making in a way we've never seen before. Traditional machine learning lacks context, which means it struggles with dirty or complex data. Generative AI lacks accuracy, which means it struggles to provide a reliable answer. System one models serve as the missing link, combining context aware input with reproducible output. You don't have to rely on Jev either. The techniques used to train Jev are well understood. Competing System One Models are already emerging, which will allow anyone to adopt the underlying technology. In that respect, Jev really is a game changer for agentic automation.

HubSpot Launches New Agents

There was plenty of discussion about AI security at Dreamforce a few weeks ago, but little in the way of interesting product announcements. Meanwhile, in Boston, HubSpot announced a self-updating CRM and a suite of nurture agents for both marketing and sales users.

Operationalising ABM in the Real World: Your Step-by-Step Guide

Finding it difficult to navigate the complexities of ABM. A new eGuide from CRMT Digital cuts through ABM hype, explaining how to implement and run an ABM program in practical terms. Boost ROI with practical approaches to account-based targeting, personalisation and engagement.

Marketing Operations Roadmap Matrix

A Marketing Operations Roadmap is an essential strategic tool that aligns marketing priorities with your overall business goals. Any roadmap begins with understanding where your marketing organisation is right now, and then defining where you want to get to. This free Marketing Operations Roadmap Matrix from CRMT Digital allows you to do exactly that.