Make your company self-improving.
AI is a hot-button topic at the moment. But how do we implement AI in a way that adds value?
What is holding back work in your business?
For a lot of larger businesses it’s the bits between teams, between systems, the bits that need checking and approval. The bits that don’t have an owner.
If you’ve been banging on about AI then maybe you’ve seen examples of AI generating sales or support responses in seconds where it might have taken humans minutes. This saves a couple of minutes.
The value comes when we connect them all together and implement loops.
What is a loop?
Loops are workflows that can be automated, improved, and scaled. They may be used by sales, customer support, operations, HR (recruitment), finance, etc. Loops are made up of:
inputs, the ‘sensors’ that feed into the system, such as customer emails, support tickets, product telemetry, and churn/cancellation surveys.
tools, or what the AI is allowed to do, such as calling an API, reading/writing to certain data tables, starting a workflow, or sending an email.
policy, or the guardrails that describe what it can do on its own, what it needs human approval for, and what needs logging.
quality gates such as evaluations, safety measures, and human-in-the-loop sign-offs for certain actions, which are particularly important when the loop’s output goes to customers.
failure analysis to understand when and why the system didn’t work and incorporate fixes.
In order for a system to improve itself it has to remember what went wrong and add this information in some way into the system. This may be in the form of a tweak to the instructions/prompt, adding an automated test, a tweak to the workflow, etc.
How can we implement loops in my business?
We do this all the time here. We measure our work in loops. When a business comes to us to implement AI we’ll look at three loops and measure the results against a baseline that’s been agreed with your leadership team. It’s treated like any other addition to capacity.
Here’s a rough overview of what happens:
- We start with a Loop Audit. This is where we dig deep into a few processes that are repetitive and likely candidates for automation. At the end of this process we’ll have an idea of the scope of work and pricing. During this phase a process owner should be appointed, someone who will be accountable for the loop.
- Next, we bring together everything we know into a format and place that LLMs can access and utilise. Much of this knowledge will be embedded in people’s heads but may be documented to some extent via emails, chat, shared storage, other systems like Raq.com.
- Once we’ve got this ‘brain’ built, we’ll look at implementing loops around it. We’ll start with inputs, what it can do, guardrails (policy), approvals, etc. This will be piloted against real-life tasks and refined until it’s good enough to ramp up usage.
- The loop will be monitored and refined over time. We’ll analyse failures, propose patches such as code, workflows, or prompt/instruction changes, get sign-off from the loop owner and implement changes.
Get in touch
Want to find out if your business is a candidate for AI-powered loops? Book an advisory call with our team. We’ll dig deep into some of the workflows in your business, particularly those that are high-friction and span multiple teams. We’ll tell you what we think is and isn’t ready to be tackled with LLMs and AI.