How We Plan & Scope AI Projects in our 100k /Mo AI Agency

22.68k views6650 WordsCopy TextShare
Ben AI
👇 Join my community and get all my templates, 1-on-1 tech help & more: https://www.benai.agency/com...
Video Transcript:
hi there so in this video I'm going to show you exactly how we plan and scope our AI automation projects in my AI agency now I know this might not sound like the sexiest topic but if you want to build your first automations or even scale your AI agency having a planning and scoping framework in place is really crucial most people jump straight from idea to building but in my experience planning and mapping out projects before building directly impacts the quality and the success rate of your automations and down the line will really improve client
satis action and even conversion rates so in this video I'll show you the four-step framework we use in our agency to plan and deliver AI automation projects now if you don't know me yet I'm Ben I've been running an AI agency since 2023 I also run a community with Founders professionals and Builders active in the AI space and if your business looking to adopt AI into your business you can also book us in for a free call in the description below now before starting a quick disclaimer this is not the only way to plan and
deliver AI automation systems this is just what works for me and my team and hopefully it can help you and second of all I'm going to focus this framework on uh delivering uh automation systems to clients now even if you're not doing this for clients or not doing it for clients yet and you're just trying to build your own automation systems I think this framework will really help you understand how to um first of all plan out these projects and second of all improve the outcome of your automation systems and lastly this framework can really
be applied to any of the the no code uh automation softwares whether you work on on nadn relevance AI uh make.com or or any other this uh should be helpful for any of these now as I said before most people jump straight into uh step three the development phase uh which is understandable it's the most exciting one but in my experience and I've learned the hard way it is really important uh to cover those first two steps first right so first we have uh the project scoping and in the project scoping we're really trying to
get clarity first of all on the specific uh problem we're trying to solve or the need we're trying to fulfill right because in the end any of these automation systems that's what we're trying to do right we're trying to resolve either business or personal need or or problem right and we also want to get clarity of course on the objectives the specific requirements and the metrics uh for the for the automations then from there we can uh dive into the system design and in this part we're actually going to design and establish a process right
because very rarely are we actually automating an existing perfectly well-defined process or system uh most of the time we have to establish and design the process in the first place and then we can actually automate it right most businesses are very messy and don't have well- defined processes in place so once we have done that we can actually make a road map and uh with that a proposal and from there we actually dive into the development where we build test debug step by step and Implement error handling at the end and then lastly of course
we deploy and optimize again rarely are these automation systems perfectly optimized for business at the first try which which is perfectly fine but we do have to uh plan it into our process and iterate and improve based on real world usage and feedback so let me go through them one by one so you get a better idea so in the project scoping what we're really trying to do is get all the necessary information to be able to design a system and to judge the feasibility of the project right is it actually doable now key here
is to mention that we're only trying to get the necessary information not all information right automation projects can get very complex very quickly and normally lots of little details and and feedback and and extra information is necessary from clients to be able to scope out a project 100% right we're not trying to do that we're only trying to get the necessary information to be able to design the system and to judge the feasibility now we usually try to do that in maximum the first two meetings and from there we send the proposal and wait for
client payment and all the extra details will figure out in the onboarding call which is the first call after payment and even during uh client fulfillment now why do we do this because again as I said automation complex become very uh complex very quickly and usually lots of little details and back and forth are necessary with clients so if you do this all up front right you try to scope out the entire project before closing a client you're first of all going to spend a lot of time uh with meetings uh waiting for uh client
feedback your emails that can literally take weeks multiple uh meetings you're adding in lots of frictions you have to wait for for clients to send emails back with with specific data points Etc and in this process lots of deals fall through right I've learned this the hard way and I've seen it happen around me too um you don't want to spend so much time on on scoping out a project for a client not to convert at the end right so it's really important in my experience the smaller the time window between the first call and
the proposal sent the higher the likelihood of conversion right so of course we do need the necessary information to judge the feasibility and to design uh a system but we don't need all the details we can do that during client fulfillment right so very important so what are the necessary details in my experience it is first of all again defining the problem right really important a lot of times companies can't actually really Define the problem that well at the beginning and you have to dig through a little bit a lot of clients come for example
with Solutions I need this uh but not attacking the problem and in the end they hire us to resolve business problems or business needs right so it's really important for us to really understand the problem so we can design the best system to resolve that problem and the most efficient for example a client comes with you know I need a voice agent for inbound leads but his actual problem is uh conversion rates right so that's the end the problem we're trying to resolve so maybe there's a better way maybe that's a voice agent but maybe
there's another another way right I'm going to give you an example of of this uh in a second then second we want to define the outcome now this is also really important which becomes a lot of easier of course when we have def find the problem is like what should outcome be now it's very important here again to get examples and specific requirements from the client on what the outcome should be first of all because we can use that to reverse engineer uh our system or our our build right because in the end we have
to work backwards from an end result and Define what's steps we have to take uh in the middle to get to that final outcome and second of all it aligns expectations right if if we get a specific outcome from a client we know that if we can deliver a similar outcome that the client will probably be uh happy with it right so really important to get that that clear now clients again will not always have output examples but even if they don't let's say they ask for a a personalized email system it can be a
good idea to ask them look write me one or two uh emails that are good personalized emails in your industry and because then we can work backwards from those emails and also know that if we can produce an email like that the client will most likely be be happy now sometimes of course a client will also uh rely on your domain expertise to actually write a good personalized uh email but then you want to probably first Define the outcome to the clients so your line expectations then the third one that's really important is uh defining
the triggers and input data so when should this automation this specific automation or project run right and what data do we have access to right really important to know what data we can work with to get to the final outcome right then fourth of course we need to know uh which softwares uh need to be integrated into the system uh also again to uh judge feasibility do these softwares actually have apis and second of all also to create an onboarding document for example where they can find the API Keys Etc and lastly we also have
to Define uh the volume and the budget for the specific automation how often uh is this this automation expected to run or should it run right and what's the budget of usage cost for this specific automation like let's say it's a it's a process that's not really solving a big need in a company and it's going to cost quite a lot in in usage costs well then it's probably not a very good project for the business and it won't help their Roi right so it's important to to Define this in those first meetings to to
again align expectations but so get let me give you a quick example so for context this was a recent client of ours uh that had a coaching business and I think it's a good example because because of the uh scoping and and and planning phase we actually ended up with a a different system than the client originally asked for um now basically how this Co the problem of this coaching business was uh they ran a lot of LinkedIn uh ads uh and got in lots of leads through those ads um but their sales reps were
overwhelmed with the amount ofad leads and basically their workflow was the qualified lead should be booked into a meeting and the unqualified one should be sent to a lower ticket offer right now the big problem was as I as I mentioned here right the sales reps were overwhelmed with volume don't have enough time to manually review leads and set up meetings with qualified ones right this of course resulted in uh slow follow-up times for the qualified ones meaning uh they left some money on the table there uh it impacted the conversion rates and of course
after that the Topline Revenue right now client suggested an AI voice agent to automatically call qualify leads to set up a meeting I put here some uh example questions you can you can ask if you if you talk to a client to get to the the bottom of the problem right I'm going to make sure to share all the documents and the presentation of this this video um in the community if you're interested so anyway after finding the the problem right we try to define the outcome now again I put in some some example questions
here but the outcome of course for them the ideal outcome is qualified leads automatically were right away booked in uh for a meeting and the unqualified ones are uh sent a personalized email with a low ticket offer now I tried to ask for uh the qualification criteria which is obvious that we want to know what is a qualified lead and an unqualified lead uh and that's the funny thing they didn't have that defined well yet so really important I didn't wait for those qualification criteria before sending the proposal right I know I can figure that
out once we get started right in the onboarding get them some time let their team talk but first we're going to close the deal and we know once we have that we can Implement that right then uh for the triggers the input data uh pretty obvious here of course the it should be triggered as soon as a lead filled out the lead LinkedIn lead form right and from that LinkedIn lead form we got the following data points which is the name email LinkedIn URL and the company URL which is optional which is also important by
the way to to know right if there are optional Fields because if we count this in and we design a system based on we know we have uh the company URL uh our system will break right so that's why these input Fields have it well defined what you always have available is very important because otherwise it's going to be difficult to design a system then for the softwares right and of course we need the LinkedIn ads go high level was their CRM and they already had a syn flow uh account for the voice agent but
that was was optional uh we could switch that for another one through they said uh and then lastly for the volume and budget so they around 150 New Leads per day so that would be the amount of times this automation would run and for the budget of this automation the alternative was to hire more sales reps to be able to handle all of these leads and so budget really won't be an issue for this one because I know this automation is not going to cost uh more than one or two employees so if you're new
to this uh I can imagine the hard thing is in the beginning understanding what Pro projects are actually feasible and what not right and as I think the intuitive thing for people who are new to this and maybe have their first client or their first two three clients is they want to scope out everything right to make sure that it's actually possible that they can actually do it uh before closing the client but again right I highly recommend to keep the time window between first call and proposal and uh client paying as short as possible
even though you might not know what you're doing yet right so why is that because first of all you have to show Trust and confidence to these clients right uh and you can only do that when you actually can talk confident with them second of all most of the time you will figure it out during the process and thirdly in the worst case scenario you buy them back their money and you've you've learned a lot right because during these client projects you actually learn the most right this is at least the approach I took with
the first four to five clients I didn't really know what I was doing but I just jumped in uh showed confidence tried and I was able to figure out figure out all of the the projects during during the project and it's also where I learned the most right so highly recommend you take a little bit of risk it's not risk in the end right uh but just jump in and and try it because that's where you're really going to learn which brings us to the second point the system design which I've broken down into four
um steps right first we want to as I said before we want to reverse engineer right and think backwards from the outcome right what do we need to do to get to the outcome so we already know we need this company needs quality qualified leads need to be booked in uh as quickly as possible and unqualified leads need to be followed up with ideally with a personalized Eno sequence or something um to uh to the low ticket offer right now in order to get there of course we first need to know what is a qualified
lead right so we need lead qualification or lead scoring system and in order to qualify or score leads we actually need data on those leads right so we need to research leads right which brings me to the second point which is break it up right this is really important uh I think also when you're starting break these larger projects or any project really up into smaller uh parts or phases how we call them or sub projects right and basically the way you want to think about it is like what's the shortest path to deliver value
or a win to the client now why why do you want to do this because first of all most we're still in the early adopter phase right most companies are extremely skeptical still with Automation Services with AI in general Etc the quicker we can show value and we can show that these systems actually work to the client the more likely they will stick with us right that first of all and second of all we don't want to scope out the entire large project just yet again because we might need more details to actually do that
um and second of all we over complicate right we want to if you come from startup worlds you want to have an MVP approach what's the minimum valuable product that delivers value to the company right and build on top top of that right itate improve and add on top of it right so if we look at this project I would break it up into three phases three sub projects which is first the lead research second lead qualification and third uh the actual voice agent or email sequences right now why because the lead research most likely
will already have lots of value to the company right because if I can find relevant data points for the sales reps put them in their CRM they'll probably already save a lot of time uh when when researching leads or they can quick quickly tell which leads are qualified uh and in which ones not is it an ideal system no no but it is better than their current system so and second of all they'll probably be better prepared for the meetings so that would be my first phase right second phase would be lead qualification and the
third one that I would add another point if you have a client that wants multiple or lots of automations which happens too right they have lots of ideas in in mind start with the one that's the quickest and easiest to deliver with high leverage for the company right that really delivers a lot of value to the company in general I won't I wouldn't recommend running multiple automations in parallel at the same time because again right MVP approach we need to show something iterate improve rarely is it perfect on the first try and if you're going
to run three or four automations at the same time it's going to be very chaotics so in general I recommend going one one by one and focus on the highest leverage uh ones first of course some companies that need to do a lot and have a big budget you can you can start doing this but I think the best when you're when you're starting out is just to go one by one now the best way to design the system is to quickly use uh a diagramming software right even if it takes you 10 15 minutes
it's going to really help you save time in the long term uh because you already understand dependencies and things like that through diagramming so I use fig Jam which is uh free it's from figma U but some people use whimsic call draw. iio I think think it's also free uh or Lucid chart and I highly recommend you do this especially if you get started so let me show you a quick example of how how I would do this for this project for example so again I've already broken it down in my head into three phases
so I'm not going to work backwards all the way from The Voice agent Etc I'm going to first focus on phase one which is the lead research right and and we know the outcome right because we already have uh that information in our scoping right so the outcome ideally is uh an enriched enriched CRM or enriched lead right in the CRM right that would be the outcome now what data points do we need in that outcome we got that from the scoping call all right so these are relevant data points for the sales reps so
we can just add them here so we know what we need to get to and we also know that the trigger is LinkedIn add so we can add in the trigger I can give it a different color so we know it's a trigger and we know the input data or the input Fields LinkedIn ads or the name our email the LinkedIn URL at the company URL but that one was optional right so now we know this is the information we have and this is the outcome the ideal outcome right so what's the shortest path from
A to B right that's what we have to Define right now now if you're completely new to this uh it's going to take you a bit more time and figuring out if what's possible what not if you have more experience this will be a lot more a lot easier and eventually for these simple ones you don't even need to because you've already done them a few times so you know you know how to resolve this but it's still good to to to plan this out so I already know these three data points right lead summary
the current job title and the years of experience we can get easily with the LinkedIn scraper right which you can do with applify right which is you can use in any of the no code automation software right so we can use the LinkedIn scraper and we basically can cover these three data points already the company size and the company summary we don't really get from uh the LinkedIn scraper because of course we're scraping the personal LinkedIn right so there's we can for example uh if the company URL is available right which is a dependency or
which is logic condition uh then we can scrape the website for example right scrape website where where we can at least get the company overview company size not necessarily so if company URL is available right then we I already know that if scrape this LinkedIn personal LinkedIn profile you sometimes also get the company LinkedIn URL uh from the company they currently work for now it's because I already know that's why this is not uh always possible to to perfectly uh Define at the beginning right when you're building you're going to add some things and you're
going to see some new things but it's still good to get that initial idea right so for example uh we can say here scrape company LinkedIn Ur but only if right only if uh if company LinkedIn URL is available now we know this now if both are not available we have to go directly here right and maybe in this case we don't have the company size yet so we can add in another condition right and basically add in that same one and also scrape this one right so if the company LinkedIn Ur is available we'll
also scrape we'll scrape both the website and the company LinkedIn URL and then we get all the data points right now of course we need to uh add a language model first right to extract all of this data right so L&M and this extract data points all right now there is a possibility we don't have access to any of this by but we know at least for this route and for this route we have all the data points now of course it can also be this route in this route we miss the company size right
and in this route we miss the company overview and Company summary right so that's why I put this asteris there because we won't always have access to this in the future right now of course what we can do is add in lots more scrapers Google search scraper Etc to try and do everything we we can to try and find this right but again we're designing an MVP first right this is already going to help this company a lot more than what they currently have which is nothing right or they have to manually look it up
right so we're first going to start with this and we can always add on top of it and it's important to note this because now we also know that if we're going to then add in the lead qualification later we we can also tell the company that these are the data points we can work with and these are not always is available right now I'm thinking about it already if this is not available you can probably say it's an unqualified lead because it doesn't currently work for someone but again you can let them decide that
so this I would Define as phase one right and I already know this is quite easy to do right so this won't take me longer than a week or two right and we can show a clear win to them in a week or two right and from there of course you can uh you can add on top of this right so we can go with the let's say we keep this in in phase one right then we can go with the lead qualification right and here we already know we need some we need some data
from them right it's like what is a qualified lead right so this is a dependency which we put into the The Proposal right and again that needs to be updated right and that would probably be pH too again that could probably be done in a week but again usually these systems need uh once they're deployed uh people use them a bit they'll see oh Ben can can't you add another data point right so we have to actually adjust this that's usually in the real world how it goes right can we add in another this and
then we find another scraper we can add so we can get another data point things like that instead of Designing this whole email sequence already for for unqualified leads we are still at the beginning of this automation right and this could I could deliver it in a week but they could ask for optimizations and it could take another one or two weeks right so we don't want to spend too much time on scoping out all the the work at the back right so lead lead qualification updated CRM with the lead score right that would be
phase two and like that we continue right and of course like from here we' go you know let's say the voice agent right and uh email sequence now again here there are going to be lots of dependencies again right like what happens if someone doesn't answer the voice agent uh where does it need to update in which calendar uh what happens do we have to send them to an email sequence if they don't answer what happens if they're not interested things like that so we don't want to go through all the details we already know
this is going to be a little bit of a bigger project so we can sort of give it a timeline so I'd give this one to two weeks once two weeks and this probably a little bit longer uh but that's what we then can put in the proposal right and I can already tell you we delivered this to the client uh after a week or two they loved it and it already helped their sales process so much but just having these data points available in their CRM then we added this part which actually took quite
a while for them to figure out their own lead qualification system and I helped them out with him with it a bit and after that uh they actually decided they don't even need the voice agent because this lead qualification system worked so well that it could just filter by the top leads that reduced down the amount of leads for the for the sales rep so much that they could easily handle uh the amount of leads and we ended up only adding the the email sequence for the unqualified leads right but it just shows you that
you know breaking it up uh into into phases is really important right and also to show quick wins because again these guys had no idea what's was possible with AI they were bit skeptical after only seeing this which is really basic right they they were blown away right but this just magically appearing in their head into the CRM uh they they they were they were amazed that's that's how easy it is a lot of the times so that's it from there we basically once we have an idea of this and of course you get faster
at this and after a while especially for the simpler ones you don't have to map it out completely you already have it in your head but uh from there we we send a proposal right and I'm going to show you a quick example too so then you can add those phases into the proposal right and usually we add an onboarding doc or we do it in the onboarding call right where we actually set up um the Integrations with the software right so let me show you a quick example so this is an unrelated proposal right
but just so you get an idea so we have the breakdown of the phases right then we have phase one with with an estimated uh time frame per phase right now in this case it was an even larger project so we broke uh phases up into sub phases even right but you get the idea and then we have the sub phase for example this case lead research and update CRM right the action what are we going to do right and the output what can they expect right what's the end result of this first phase and
the dependency now in this case none but let's say in that second phase dependency can be we need to have the lead research but also we need to have the lead qualification criteria right and for the third one we need um an example over a personalized email Etc right now that's that's sort of how we lay it out then and now for prices we do uh always do subscription based pricing uh it works by far the best in our experience because again uh most of the time when you deliver let's say this first phase uh
in this case for example the client loved it but they wanted some extra data points um which is perfectly fine right we want to add and improve on the system but if we've already quoted them for specific project like that and they want to add things on top we have to quote them again we adding friction to the process so because they're on a subscription model they have a lot there's a lot less friction we can improve ad on existing uh projects and they will understand too that the exact timelines won't be complied with if
they want to add something to uh the the the first automation Etc so it becomes a a lot more flexible and and a better relationship in my experience if it's Subs subscription based pricing because if you do this project based you either have to scope out all the needs in the entire project uh completely which is going to take you lots of time right and if they then want to add something on top or or something doesn't work then you would have to add another quote on top of that and again it just adds friction
so highly recommend the subscription model and that's it uh and then we do the onboarding call where we uh explain how to to integrate the softwares right we always set it up in their accounts and uh that's it and then the next phase of course is we're actually building it now I'm not going to show you actual building I have many videos on my YouTube channel if you're interested in learning how to build but I'll show you quickly the steps we take when we are developing these these systems first we build and test step by
step right so we make sure that every node or module or step Works before adding the next one uh now again you you can predict sort of the edge gases and where can go wrong but also don't over optimize right again have the MVP minimum viable product mindset in mind when developing this right if the automation fails maybe on two of 100 uh leads right because the LinkedIn can't be scraped Etc it's fine right it's still going to add a lot of value uh to to the company and it's a win for the company right
because before this they had nothing right so we can add two or three more weeks or whatever onto the automation by getting it perfect but if we can get uh it a 90 or 95% solution uh in a lot shorter time frame we should aim for that right of course there are are are exceptions where there are really important tasks that canot absolutely nothing can go wrong right that's a different scenario but in general right in for the MVP uh approach right then and then iterate and improve of course uh on the system once it's
in production and and being used in the real world right and once you get feedback from the client which is really important for you to be able to improve the system right then once you've built the the entire phase or the the system then you're going to test it on ideally 10 to 100 examples right you'll need usually need to do some debugging there and improve the system there already uh then we usually add in error handlers right um we always add in these retry mechanisms for API modules especially if you have large um volume
automations apis fail it's it's very common right uh because we're relying on their servers and and things go wrong the easiest way is um most of these these no code uh automation softwares already have have these uh error handlers built in right I can show you quickly make.com but we always add in uh these break modules uh for um apis right so let's say the Google Docs right so how do you do this I can show you quickly here you have these are basically error handlers right so we have break commit ignore resume roll back
in general break is enough right and basically what break does is if it fails it will try uh it will try again with a delay right so here we can Define how many times usually three with a 15 second interval we already uh resolve 90% of these these uh these issues right you can also add in ignore Etc but really we're not we don't use any more than this in NN you can also really easily do it right in any API modor in any any modor really you can go on settings and then retry on
fail right again Max tries and wait weight between right so we're always adding this make.com if a scenario completely fails usually you get notification anyway and uh in any then you can also add in separate workflows to actually um notify you when specific errors um happen in your workflow Etc uh I'll probably make another video on on on error Handler it error handling if you're interested in it here have an example right but uh I'll make sure to to share these these workflows too in the presentation so that's it and then lastly uh we we
deploy of course and and that's where it it sometimes just starts right after getting feedback after the company trying it out um you usually they get lots of ideas on how to improve the system how to add things to the system uh how to iterate on the system Etc so you'll have a cycle here of improvement right and adding things and of course if everything's running really smoothly we can continue uh with the next automation or the next phase in this case and expand on the current automation now I highly recommend uh when you're working
with clients to open up a a slack Channel with them or invite them into your slack channel so have a direct line because this this is really required with these automations you're really working inside of their business so it's really important uh you're in line you understand you constantly need uh some things from from them Etc so of course you don't want to overdo it right you don't want to be spamming them uh you want to be professional right uh we we usually try to limit our our our requests to once a week but uh
in general it's good to have sort of this open line and for us slack has been working the best so that's it I hope this uh this helped some of you uh I know it was very client focused but I think if you're just building your first automations this also gives you a good idea of how to plan out your first automations yourself and uh maybe how to eventually uh sell this to clients too uh also soon I'll make a video on on this whole onboarding process how we do it inside of our our agency
because we basically automated the majority of these these steps uh from the meeting transcripts to the proposals Etc so I'd love to show you because I think it's really interesting and uh yeah thanks again for watching and uh I hope to see you in the next one bye
Copyright © 2025. Made with ♥ in London by YTScribe.com