![](https://www.facebook.com/sharer/sharer.php?u=https://www.barefootsolutions.com/news/barefoot-bytes-newsletter-11/&picture=https://www.barefootsolutions.com/wp-content/uploads/2024/01/Barefoot-Blog-900-x-600-px-19.png&title=Barefoot Bytes: Grow Without Hiring"e=Barefoot Bytes: Grow Without Hiring&description=Barefoot Bytes: Grow Without Hiring)
Welcome to the June 2025 edition of Barefoot Bytes. In this installment, I’ll be laying out the playbook on how to deploy AI across an organization to boost the efficiency and productivity of your employees.
I don’t think I need to convince people that it’s imperative to deploy AI at your company. As a rule, once my Mom becomes fluent in a particular technology, we have reached mass adoption. So I want to dive right into the how of it. There are so many tools, challenges, and opportunities that it’s hard to know where to start. Even harder is how to successfully design and deploy solutions that drive a significant return on investment. Having helped a number of organizations with this now, I’m seeing some commonalities and starting to develop the playbook.
Ultimately, this is about way more than just a single use case. But I think it’s important to find one or a few that really resonate with members of the team. It’s easy to get lost in all the possibilities, but having a specific focus adds the ability to remain accountable to specific success metrics. A law firm might want to help their business attorneys write agreements faster. A pharmaceutical company might want to generate clinical trial protocols more efficiently. A defense contractor might want to crank out more proposals in a given year. While there will be many more use cases by the end of this process, it’s very helpful to identify something that has an easily understood and significant ROI.
Calculating the ROI can get tricky and often depends on strategic thinking from the C-Level. I had a client ask me, “If this is making my employees more efficient, but I have the same number of employees with the same salary, how exactly am I seeing a return and not just another expense?” Fair question. The answer comes down to – what would your employees be working on if they suddenly have more hours in their day? Ideally they would be increasing productivity and output, driving additional top line revenue growth. Here are some examples for how to calculate the return.
An attorney estimates that she could save 6 hours per week, or 300 hours annually. The firm has 100 attorneys, the average bill rate is $500/hr., and the utilization rate (the amount of their time that is actually billable) is 80%. So that’s
300 * 100 * 500 * 0.8 = $12,000,000
The pharmaceutical company thinks they can produce 10 additional Phase 1 trial protocols each year. Average revenue per trial is $30,000. So that’s a $300,000 return.
Similarly, the defense contractor thinks that they can increase their SBIR proposal output by 25%. So that’s 5 more proposals, 2 more projects, and an additional $2,000,000 in annual revenue.
In all three cases, we now have justification for a significant investment for only a single use case. And at this point we commit to tracking these metrics with a quarterly review to actually prove the ROI.
If you’re curious, here’s an anonymized whitepaper we commissioned for a client to help calculate ROI.
So we’ve identified the killer use case and justified the investment. Next we need to identify the specific requirements. This ensures the solution meets specific needs, constraints, and success criteria. Stakeholders are aligned, minimizing the risk of implementation.
For this platform to be more than just a private ChatGPT, it’s going to need to be connected to other systems used by your company. Some of the common considerations and popular examples:
Single Sign-on – Microsoft365 or G Suite
Document Management – Microsoft Sharepoint, OneDrive, Google Drive
HR – Workday, ADP, BambooHR
CRM – Hubspot, Salesforce, Dynamics
Accounting – Quickbooks, Netsuite, Xero
Business Intelligence – Power BI, Tableau, Looker
External Systems – USPTO (law), Sam.gov (defense), or cdisc.org (pharma)
Will your employees be using confidential information with this AI system? If not, then utilizing a third party LLM like ChatGPT is likely a great option. But if they are dealing in confidential or regulated data, then it will be important to explore solutions that you can deploy “on-premise” in your company’s cloud infrastructure. This is to protect from the risk of allowing proprietary information leaking into external model training data. If you think you’re protected with ChatGPT Enterprise because the agreement says they won’t use or keep your data, think again.
This is probably the biggest risk when deploying a solution like we’re talking about here. We want to make sure that existing access controls are honored and enforced in a new system. If an attorney at a large firm is not allowed to view certain client documents due to a potential conflict with one of their other clients, it’s extremely important that this platform doesn’t allow for that to happen.
The way to do this is to require that your users have to authenticate with each of the integrated systems with their own user account. Using master API keys is a no-no. This is fairly straightforward if everything is done through your SSO provider. Otherwise, the platform will need to handle authentication for each of the integrated systems. And when storing files in a RAG database, every chunk needs to be tagged with a user ID and enforce that queries only return results that are tagged with that ID.
I thought I was clever for figuring this out but OpenAI announced the same design pattern a couple of weeks after I did. Validation I suppose.
A robust logging framework needs to be included, which will become important in a later step.
Once we have the requirements documented and signed off with key stakeholders, we’re ready to start exploring solutions.
With the number of options out there, I find it unlikely that you will decide you need to build a totally custom solution for this. So armed with the requirements, it’s time to do some research. This can be a bit annoying as everyone wants you to do a demo before you can understand features and pricing.
Here’s a prompt to help kickstart your research:
I am the CTO of a national intellectual property law firm. I want to deploy an AI platform for the attorneys and paralegals at my firm to boost their efficiency and productivity. This would be an AI Assistant with a chat interface. It needs to have single sign-on with Microsoft365 Entra. It needs to integrate with Microsoft SharePoint. I also need to include a custom integration with the USPTO.gov API, so the platform needs to be customizable as I doubt any options in the market today include that integration. We will be dealing in confidential information, so this needs to be able to be self-hosted in Microsoft Azure. Given my requirements, can you do some research and provide an evaluation of some potential solutions that are available in the market
I asked OpenAI’s o3-pro model this and the response was pretty impressive. It took like 5 minutes but it would have taken me 5+ hours.
Going through the technical bits of deployment is outside the scope of this article, but there are some important organizational steps to make sure that the deployment is successful.
Create some documentation that helps orient new users to the platform.
Identify a small group of beta testers. These should be employees across different departments that are fairly sophisticated and experienced with using LLMs. They will identify issues that will need to be addressed.
Now run a second round of beta testing, but this time with less sophisticated users. You will find that they have a different set of problems. For example, they might be lousy at writing prompts or not understanding the documentation.
If you have more than 50 employees, I would suggest a phased rollout department by department. Each might have unique challenges that you want to address, and that could be overwhelming across many departments at the same time.
Fully baked training materials will be required at this point.
The critical factor here is adoption. Achieving the calculated ROI depends on it. If nobody is using this, it will be a failed project and the ROI will not materialize. Review the logs and do some analysis. Who are your power users, casual users, and non-users. Figure out how to convert non-users into casual users and casual users into power users. Go back to the metrics we committed to measuring during Step 1 and let’s keep ourselves honest.
Finally, there need to be feedback mechanisms in place so that the platform can be continually improving. Users should be able to ask questions, report bugs, and request new features. If new systems are being evaluated, their compatibility should be an important consideration
Beyond the killer use case, now is time to find other ways that an AI platform can drive value for the business. Some of the common scenarios I’m seeing:
Think of this as your new AI operations platform. Rather than a complicated ERP implementation, this can be the control center that your team logs into first thing every morning. Imagine,
“I have an hour to respond to emails. Can you please prioritize my emails in terms of importance and provide the top five?”
“I just got off a call with Client A. Can you upload my call transcript into the CRM?”
“Can you provide some market research based on this RFP?”
“I’m writing a patent application for Client B. Can you review their previously awarded patents and provide the commonly used definition for a GPU?”
Your employees need this.
This was a long one so just a few bytes:
Cheers!
Share this article
4241 Jutland Dr., Suite 300
San Diego, CA 92117
