Six tactics to help enterprise leadership teams navigate the first wave of Gen AI

GenAI is disrupting the enterprise stack — from underlying compute to end applications — and the reverberations are being felt across industries. LLMs, trained on general datasets, demonstrate specialist capabilities that previously would have required years of domain-specific training. They can automate creative and analytical tasks, thereby significantly increasing worker productivity – particularly for low and medium performers.  

Founders scaling consumer and enterprise tech businesses are urgently working out how they can harness this technology into their playbook. But the path to success is far from straightforward: the landscape is rapidly evolving, LLMs are inherently opaque and their failure points are unclear, with numerous hidden risks to be aware of.

Right now, it’s hard to judge when an LLM is good or bad at something. It could be great at coming up with a limerick but terrible at basic maths.  The problem of where AI excels in some tasks but falls short in others is known as the ‘jagged frontier’. Where a task is outside the jagged frontier, there is a risk that using AI could even negatively impact output.

As a growth investor in companies addressing large market opportunities - everything from online fashion to sustainability and analytics in real estate — we’ve uncovered some proven tactics to help tech leaders join and stay in the AI race. 

1. Buy, test and try, rather than build

We remain in the “first wave” of GenAI and companies can be confident that this is the worst AI that they’ll ever use. Technology will continue to evolve rapidly in the coming years and it is not clear which experimental products (and business models) will create sustainable value. We’ve already seen how technological differentiation can be rapidly eroded with the pace of change in this space.

This isn’t a reason to stay away. Founders should identify quick wins and productivity gains, implement test and learn pilots to learn the new ‘patterns’ of working with GenAI, and prioritise flexibility (both in terms of cost structure and provider landscape). 

For most (non-core) use cases, founders should 'buy’ rather than ‘build’ the required Gen AI capabilities. Building is costly and a distraction from delivering core products to users. And, given the pace of change, anything you build in-house risks becoming obsolete within months.

Fine-tuning of models is time and resource intensive and will quickly become commoditised — perhaps within months. A combination of retrieval augmentation generation (RAG) and few-shot prompting may be sufficient in most cases to incorporate Gen AI into your own product set. 

2. Focus on the quick wins

GenAI is having the most impact in a few functional areas, namely sales and marketing, customer service, and software development, as the experience of our portfolio companies bears this out. McKinsey estimates that three-quarters of the global impact of GenAI will come from just a few such functions. Think customer service, professional writing tasks, idea creation and basic code generation. Unsurprisingly, this is where the majority of today’s GenAI startups are focused, according to a Gartner survey of ~500 recent GenAI products.

3. Spot hidden risks 

There are specific risks for businesses considering using Gen AI in addition to the widely reported general risks of hallucinations and bias. The lure of productivity gains can distract from the lurking legal and reputational risks that could emerge. If they do, companies will find that they cannot simply blame the LLM - and trust in their products may be hurt. 

To safeguard your company’s reputation, human verification needs to remain a priority.  Scaling companies will need to carry out due diligence and control over how content is being generated, as well as considering full transparency with customers or clients. At this stage, it’s vital to encourage humans to use and trust AI.

Not only does keeping human agents in the loop protect against inaccurate, harmful, or sensitive content but it also helps to overcome mistrust about the technology taking people’s jobs. 

4. Upskill your team

For some companies, the knee-jerk response has been to put in strict no-GPT policies. But this just pushes employees into ‘shadow IT’, with employees preparing to use even prohibited tech if it makes them better at their job. Essentially, thoughtful policies are needed that take in the risks but also the realities of Generative AI. So you must invest in upskilling your team and increasing knowledge.

CROs and CMOs can mitigate risks with employee training and strict policies to keep humans in the loop and verify strategy. You will need to make hires but don’t let hiring get ahead of realism. Generative AI can be a skill leveler, bringing everyone up to a higher base level. 

5. Maintain data security and integrity

In software development, all processes need to be updated to be cognisant of AI. This includes robust quality assurance and security processes. You need to understand what vendors will do with your information.

Generated code can contain bugs as well as potential security vulnerabilities — and if it incorporates on open source code, you run the risk of diluting the IP of your code base by entangling it with external code with conflicting obligations or licensing requirements. 

Consider the many legal implications:

• How will your confidential information be used? 
• Will you own the generated outputs? 
• What if the output infringes a third party’s IP rights? 
• How do your products comply with current and future regulations including copyright protection? 
• What is your internal compliance process?
• Does the vendor offer an opt-out of data being used to train the model? 
• Are you ready to take down outputs that infringe on a third party’s IP? 

6. Be braced for rapid turnover of technologies

To make the most out of Generative AI progress, founders and their teams need to brace for an imminent and rapid turnover of both technologies and vendors. Identify quick wins in priority use cases, and adopt a test-and-learn approach. 

Waiting should provide a clearer, more stable picture to make strategic choices. 

The promise of Generative AI is much simpler than the reality of integrating it successfully. As the technology continues to evolve rapidly, your priority should be to keep a realistic perspective on Gen AI's capabilities. That way your AI journey will be all the smoother.

Lead image: Freepik