I don't think anyone should sell AI systems just on the go, opening N8N or any other tool and connecting a couple of nodes and saying, "Hey client, this is for 1,000 to $5,000. Do you like it or not?" That's very unprofessional and I don't think you should vibe code your way through. As well, in this video, I'm going to show you exactly how we've done AI audits. This is something I've not shared in the past with anyone because it's based on internal documentation we have for the employees at Ciola.
And by the end of this, I guarantee you that you're going to be able to make any or do and conduct any AI audit systematically and successfully that guarantees the ROI for any business. That being said, let's jump into the video. So, today we're going to talk about how you can find AI opportunities within businesses that guarantee you an ROI and are profitable and make sense to the executive team and the businesses as well. And how you can go from idea all the way to production.
And we're going to talk about the first step of all of that which is essentially the AI audit. What you see on the screen over here is the CLA internal hub. A side note, this is not for sale. I'm not going to sell you this or tell you there's a subscription to it or so on.
This is something we use internal for this team and also any new member where we have the operational playbooks, the knowledge base, the tools and resources and they can search for help and so on essentially in this database and I've modified it as well just to remove some names and confidentiality of clients over here. But I'm going to give you everything in these videos and the series of videos to come. So essentially what we have I'm just going to give you a tour on the homepage. Then we're going to jump into the operations.
We have the clients where they are within the project. This is just a very systematic approach. Think of it as project management on steroids and then they can go to different segments and if they have any commands or they they want to talk about they want to change the templates, the frameworks or they have suggestions. So how this works is essentially I've built a couple of commands that they can just type in certain browsers or certain certain interface let's say and then things get started.
For example, if you want to start an AI sprint, you just launch a new a sprint by typing this command over here. It's just going to start the sprint and so on. So there's a lot of things over here, but we're going to talk about AI audits in particular and I'm going to go into all the details about AI audits and how we do it from start to end. I haven't seen anyone do it in this details but this is a proven approach that we have done for many many companies and it guarantees to get them to the outcome and by the end of this video you're going to have exactly how you can do it and go sell it yourself.
So MCR which is the mission concept review AI discovery sprint this is the complete playbook we're going to go through everything one by one. So essentially what happens from a very bird's eye view of the high level you have a discovery call with someone for example if you're going and you want to find an AI use case or do an AI audit you have a link somewhere on social media or your website and so on you do a discovery call can be 30 to 60 minutes and you're going to see the interest on the other side in terms of what they're there for and for AI and if they're not interested you end the process if they're interested you start sending the proposal I put an arbitrary number over here which is 10,000 because everyone loves the number 10,000, but you can use any number that you want over here. And usually these go between, let's say, $2,000, $2,500 all the way up to $15,000. You can charge for such discovery sprints if you want depending on the entity.
And usually it takes anywhere between 7 to 14 days. If you're using AI, it can be faster, but just keep that buffer over there, which is 7 to 14 days. So once you send a proposal, you get a payment, what you do is you do an executive call. you sit with the the party or the department head or the responsible entity or the responsible person on the other side and deep dive through a certain questionnaire asking certain questions that's going to feed into everything else we're going to talk about then once you have that transcript or that information that note or whatever it is then you begin the MCR phase which is a mission concept review phase and then day one to day two usually is intelligence you gather all the intelligence the emails and so on then you do assessment. Then you have a gate decision over here and this is very critical and this is what makes it really no risk on them where you say okay we found certain things that we can build out and it's going to be profitable for your business certain AI opportunities if you have not found anything we're going to refund you and we're either we're going to pivot we're going to go again for the next 10 days or we're going to build it out and we're just going to go through this process and this they love it because it's no risk on them.
So you go through the process if it achieves that process you build a proof of concept either through N8N or any other tool that you choose just to show them that it works. You're not shipping any anything production ready at this point. You're just showing feasibility. Then you package it in a way that they can show to their stakeholders or their bosses or whoever you're dealing with because certain entities like certain reporting, certain documentation just to have something tangible to show the upper party that we've spent money over here.
Yes. and it got us something tangible and one we're one step closer now to achieving this AI opportunity target that we have and then you deliver everything to the clients. This is a complete overview of how this phase goes. But now we're going to go into details about each one of these. So the prerequisites over here is the payment confirmed.
So this is what I give to client to my employees and that's why you have this level of detail in terms of what they should have been given you to move to the next step and what's the phase start and end end stage but I'm going to give you all the details over here. So this then the system setup starts where you go and set the system setup and this naming and then you go to use any jet chat GBT or claude or anything for information synthesis and then you do a mirror account for the diagrams and then you decide on the proof of concept platforms and so on. That's the system setup and the time allocation essentially is this much faster with AI but essentially this is what we do and what's the key outputs within each of these days and what's the communication channels and what we're going to give during the process of this client delivery and the must have requirements we have over here which is very critical for the success factors of the project or the MCR phase is that everything or every opportunity that we find should demonstrate a monthly impact greater than 10,000 dirhams you can keep it dollars as well and the complexity should be less than four and we have a certain criteria of gauging that complexity and then the data should be available. What if someone doesn't want to give the data or the information about certain thing how can we build it out and the proof of concept should be feasible that we could build it out.
So the gate decision framework is very simple. At day three and four, there's an assessment completion. We're going to see that if it meets the 10,000 dirhams or dollar threshold. If no, we can go the refund route or pivot to something else.
If yes, then we look at the complexity. If no, we're going to say it's too complex. We need to just choose this at the initial stage because we don't want to go all the way to agents and fail. But let's start with assistance somewhere over here.
Then move our way to agents. Then if it's yes, we go to data available. Is the data for this available or not? If no, we're going to say okay, do you have access to the data?
Can you give us the data? Is there any restrictions over here? If he says okay, I going to give you the data is fine and you move over here. But if no, then we go and check something else as well.
There's a output over here. And then if yes, then we see the proof of concept whether whether it's feasible or not. If not, there's a lot of technical issues. So for any reason, we can't just apply the proof of concept and do it.
So we just tell them okay let's pivot to something else or we proceed to the proof of concept and build it out and test it out. So for risk mitigation this is also for the employees where they can look what if the client data is not available what is the complexity we underestimated it there's a scope creep that comes in what if the proof of concept tools they have limitations and again scope creep over here and the success metric for us is that 80% of the proof of concept should have a success rate so as we build this out it should be giving good data out the timeline should be done within 10 days in maximum the value should be this much in terms of the ROI of the thing that we're going create and the client of course we need to do professional delivery in a systematic way that's aligned with the clients and the stakeholders otherwise it makes no sense once this all is done now we can move to the next stage which is let me go into the details of the premcr setup so what we do is first we make sure that financials are covered and the communication is done we've invited them to the call and we have access to all the information so what we do pre the MCR essentially is as follows We make sure this check checklist is done and then we do the executive call preparation and before the call what we do is we research the company background. Of course we use firecrawl API if it's a big company we use perplexity and we just see the internet of what we can gather as much as we can using agents that we have within Cola and then we prepare the recording setup. We review the discovery call notes in terms of what what the tools whether the tool is working or not to get the transcript on the call of course asking for permission as well.
And then during the call we go through this exploration phase where we're going to see and deep dive into that particular thing that they talked about. The main objective here is to find the pain points the bottlenecks and how the process goes and who's responsible for that process. And after that call we have an input file which is the raw files that's going to be used. I'm going to show you in a bit for what, but we have that transcript file, the emails, anything that's not confidential, we're going to keep it on a side.
And then we're going to essentially synthesize this information into a nice package, which is our initial condition or something we're going to start with for this AI audit. And these are some potential red flags over here. If the client cannot articulate clear the clear business problem, if the required data is inaccessible and you know the things essentially that may lead us to double think what we're doing over here and the quality gates are before starting is the business gate. It it should have a clear problem that's worth $10,000 plus 10,000 dirhams plus per month.
It should be technically feasible. I know I'm repeating myself but this is very very important and the data should be accessible. the stakeholder should have that buy in and say okay this makes sense to me I'm I'm convinced I'm committed I'm excited about this use case and the scope should be clear at this point and the common preparation issues are over over here so once this is all done then the funds self starts and begins where we have day one to two which is intelligence gathering and this the main objective over here is to synthesize everything using AI of course we're not going to do it this manually because it's going to take so much time But we leverage AI as much as we can to create one single file which we call init.md but you can call it initial doc conditions in initial documentation whatever it is. So we create the client workspace. It has a certain structure that we keep essentially it's the inputs the raw inputs and the initial file which is what's going to be created over here and the deliverables and the reviews and the work which is the prototypes and so on.
So the intelligence that we can gather can be the transcript call the emails that we had in between the company research that we're going to do using certain AI agents that we have the stakeholder mapping which we can use LinkedIn profiles and other places where we can get that hierarchy of who's who responsible for what mainly to know their influence their interests their risks what they have done recently and so on and it blows their mind if you just have all this information uphand and before like starting everything and also You're going to see if there's any ex existing documents that they share with you. Sometimes they do share, sometimes they don't. Sometimes they give you screenshots. So this is just to put everything in that raw file that we have.
Then we synthesize and this is very very important and very critical step that we try to spend as much as time as possible over here to have that enough context to start with because once we start with the context that will define everything else and everything that comes after this. So the inputs are the raw files and then we create this init file over here using the form template that we have. Essentially if the employees are checking it out, they're going to check it over here how the template looks and then once the AI outputs that template, they can check what the template is. So what happens in this step is that we extract the problem based on everything we have and we documented for example what's the exact cause or exact language that was used by the client.
What's the problem statement in this structure? What's their current process of how they do things? What is the pain points they have which is right now that's that's very like annoying them and they need and they're very frustrated by and they need to fix it. What's the bottlenecks that may occur in the future that may make a lot of problems for this exact process.
This is the problem extraction. Then we have the AI opportunity scan which is also based on another part over. So you can see over here we use certain knowledge bases. It's all over here the information but how we do things and it's constantly updated as well based on the feedback of the clients and what we do and what we go through.
So then we map the problems to AI capabilities. AI can a either need classification, extraction, generation which is gen AI and so on for content creation or analysis where we gather insights from a certain amount of data that we have and for each opportunity we write it in this format over here and then we analyze at the end and see which opportunity is the most likely or the lowest hanging fruit that we're going to go after. And point number three is we look at the business requirements and we write it in a system engineering way where we say the system shall do certain in a measurable way and the system shall not do something that they don't want and this is the performance target essentially in terms of speed and so on. I don't know if there's examples over here or not but I'm I'm going to tell you.
So essentially it calculates for example the hours saved and how much does that hour cost per employee and that's how we get the monetary number to it. and how do how much do we reduce the error in terms of if this is an AI it's doing it and if it's a new revenue opportunity or not and what's the customer satisfaction over here so it looks something like this block over here as you can see and then step number four is the initial process mapping we can use myro or any flowchart tool we use myro where we have different states in the diagram we put green dots we put red dots we look at different let's say pain points bottlenecks and whatever is an opportunity that's worth $10,000 plus. We put it the green dot and those is what's presented to the client. And step number five is quality check is of course we're going to check everything before sending it just to make sure there's no typos, the professionalism and so on. And this is just common issues that may occur and this keeps on adding on based on what we see that happens usually.
And then the end point is they have this init file which is the initial condition or initial file that we're going to be starting with a complete process workflow and we have those AI opportunities that's validated and will provide that ROI that we have and then only then we move to the next step which is I'm going to go to now the assessment. So during the assessment after that phase is done and very important to note over here we don't move to the next phase if we don't first finish the first phase and this is a system engineering kind of way of doing things where we do the assessment during the MCR and the output over here is to get the SRS MCR which is means in very very simple terms is what do we want to get out of this process that we're going so the starting point as we mentioned is the initial file that we just created the process diagrams and the bottlenecks or the pain points that the AI will fill and integrate and is worth doing instead of human labor. So we create this document over here. This usually takes us eight hours with AI much faster just to provide to them something that has the following elements.
The first element is a mission concept which is single single statement saying that this AI system shall do this particular function to enable this business outcome that we want within this performance metric that there is. So an example over here is the customer support AI shall automatically classify and respond to customer inquiries to reduce response time from 4 hours to 5 minutes. This is just one statement we have at the top of the document. Then we draw simple diagrams here and there just showing what's the system boundary, what's the external disturbances that may occur, what's the data inputs, what's the data outputs, what's the user interactions and so on.
This is just an example. You have the AI system over here. You have customer emails coming in. you have a knowledge base that's coming in then you have these outputs human escalation or an automated response or add it within a CRM system and get things from the CRM system to just for context for the AI system. Then we set up the functional requirements.
The functional requirements look something like this where where they say is the system shall automatically categorize incoming requests into which is let's say emails into technical support billing inquiries product questions and escalation required looks something like this. The system shall generate contextually appropriate responses with 90% accuracy for common questions less than 30 seconds of response time integration with existing knowledge base and so and so on. So we get into details over here but not too much details because the full details comes in the next phase of the project management. This is just defining the AI opportunities but within the next phase we go into much details about this.
This is very high level and then we go to the non-functional goals which is the performance. What performance you want? What accuracy we want? How much uptime do we want?
What security level do you want over here? What's the user experience standards that we're going to follow over here? And then we do this AI feasibility assessment which is all based on the knowledge base again which is what's the complexity what's the solution type are we starting with assistant co-pilot autopilot agent and we try to start with assistant or the lowest automation or lowest integration kind of tools at the very beginning then move upwards what is the proof of concept tool what is data requirements and what is the success criteria I know by now you're thinking okay this is a lot of things you're going through but with the help of AI and if You really want to guarantee that ROI within a business where you integrate AI and not just integrate AI for the sake of it because you got an idea. This is the exact process systematic process that we found to guarantee that success.
So step number two is you apply AI card analysis. You use these knowledge files and then you map the capabilities of the AI with the task at hand or requirement at hand. For example, if we want email routing, that's an classification AI capability. And then you look at the tools that exist.
Of course, we have an extensive list of tools and it keeps on updating. And I'm trying to be very tool agnostics as I explained this to you. So let's let's take another example over here. The requirement is generate responses.
It's a generated capability. Custom DPT can do it. The complexity may be three over here. So the complexity scoring happens 0 to2 is very simple. 34 is moderate and then 5 to six is very complex and for this 10th day delivery for this proof of concept plus the requirements plus the AI opportunities of course is going to be less than four.
We don't want to go into very complex things in 10 days because that's a completely not believable to any business. So then we come to the tool selection. We have a very extensive list of tools. So then we have the risk assessment in terms of what's the technical risk over here in terms of data, the integration, the tools they have, what's the limitations and so on.
What's the business risk over here and what's the timeline risks over here that we have in terms of this is very critical to be done because the person has a presentation to do for his boss in 4 days. So we need to proof of concept within 24 hours or whatever the case is. And then step number four is the gate review where we have this checklist for saying that okay did we pass this phase or not and essentially what happens is we say okay did we find opportunity what was the complexity and go through all this and the decision criteria is something like this and then it creates this review file within that client workspace that I showed you if whether it's a go decision this is the recommendation go and create the proof of concept right now or it's a no-go decision where okay let's go back and look at a different opportunity or do something else. And the common issues that we found during this point are these over here.
And this is essentially the end point of this process and how we finish the the process of the assessment of the MCR. Then after this once everything passes, MCR says it's great AI opportunity is great. This is going to is not that complex. The data is available.
The executive or the client is happy. Now it's time to go and build this proof of concept out. And this is the starting point over here. And we're going to use everything we built so far for the next phase.
What we're going to do is whether it's a chatbased tool, whether it's an N8 tool, it does not matter at all. You can use whatever that delivers the result. And you just set up the environment. I'm going to go through quickly through this whether it's NAN and so on.
And then there's a trigger for any workflows. There is this trigger that's going to happen. And then there's this AI process that's going to happen based on the use case. Then the decision is made whether it's going to do route one route two which is a logic in there and then we're going to track certain metrics just to show that okay this is what you have coming in 500 emails let's say per day and this is you can see the classification and how fast it happened and we met the requirements we set in the first phase what I show you day one and day two which is the high level requirements and the assessment phase so this is just going through how you're going to develop this what you're going to do and then you're going to write these certain documentations over here in terms of what was the configuration What was the test credentials?
What was the notes over here? What were the core features? And you can get as much as advanced over here you want. You want to use cursor and cloud code.
You can you can use whatever you want to build this out. So this is the P which is the proof of concept. If it so let let me just go because this is important part then we're going to go to the next one which is what if it fails? We're going to document the failure mode.
We're going to analyze the root cause and then we're going to try to fix that. What if it partially works? it does not meet all the requirements. We're going to see what we can fine-tune to get it to work. But this essentially is the process that we go through.
And the final point which is the most important point that people miss is that once you're done the proof of concept work, how can you package everything that they are happy with and they see as tangible and not okay this is the thing you work take the link go and try it out and see what happens but more of something they can present a presentation or a document or a PDF or something like this. you can see over here where they can look into and pass on to their team. So essentially what we do is we give them a presentation. We do a presentation for them which is around 60-minute presentation showing them okay this is the process. We do a complete MCR thing.
So we show them the process. We show them this is where we started. This is the requirements review. This is the requirements we did and this was the assessment and this is what we built out and this was the result and this is what if we've passed the requirements we've set or not.
And this is the presentation or the deliverables where you can test it out one and the report that you can pass on your team. And then very very important is the next step. So if you're just selling MCR the AI audit, this is where you enter the next step saying that okay now this has proven to give you that AI opportunity that you want. The next step over here is to actually set the system requirements extensively and go through that review which is the next stage of the system engineering which is the let's say the system requirement review.
It has you set all the system requirements. It's like a very big Excel sheet that you're going to write that has all the requirements that's has to be met as we build the full integration out when you get the production ready. This is just a proof of concept that I mentioned. So you get the documents done, you create the one pager, you create the proof of proof of concept access, you package everything, and you give them something that looks like this, which is an executive report, an executive fact sheet.
Yeah. Proof of concept access guide, the implementation proposal, if they were going to proceed, which is the next one, which is the next phase, and then a read me essentially telling them all the package, and then you do a final quality check, and then you hand it off to your clients. Of course very very important in the knowledge base everything is pulled from there. Anything we do in this playbook which is for the MCR phase only.
We have it for every single phase. For the MCR phase we pull information from the knowledge base and we put also some parts in the information where it says this is the feedback of client XY Z in this industry that we face this issues and this was the requirement. So it's getting smarter by itself and this is a living document as well. So this is exactly how you do an AI audit.
I have not seen anyone do it this way in a system engineering approach. And this exactly is how we did space systems and made sure they work. And this is just one phase out of