The way I frame it - Copilot makes one person faster. Infrastructure makes the whole team faster. Copilot = 1:1 help. you ask, it answers, you get a little more productive individually.
Architecting gives you scale, standardization, and automation. You build it once, and it runs for everyone. The Analysis Dossier generates hundreds of reports without anyone writing a prompt.
When pitching internally, I focus on the output and ease of use: “What if every seller had account research ready in 5 minutes instead of 2 hours—without doing anything except ticking a box?” That usually gets attention faster than explaining the architecture, and in fact what I had started with in my enablement sessions.
The distinction between AI that assists vs AI that performs is spot on. Grounding the Analysis Dossier in real source data through APIs completely sidesteps the hallucination problem. How do you handle the maitnenance burden when these workflows span multiple systems? Does your team need to understnad the full pipeline or can different people own diferent segments?
Great questions! And yes - when I figured out how to grab the full texts from 10-Ks (SEC API), and uploading the PDFs through ChatGPT native assistant function for analysis, it was a game changer. Made the results so much more specific, relevant, and salient.
I actually maintain the Analysis Dossier workflow myself, which doesn’t take a lot of time. Most of this is fixing any issues (rare, maybe timeout issues if it’s being run too much? But Zapier reruns those automatically) or new requests to add sections or something.
Most of the time I spend is on improving the report, like figuring out new ways to make it accurate. Last week for example, I split up a part of my workflow for accounts with and without an Annual Report PDF to streamline the process. I had it all in one because it was easier to build at the time.
Next week I might take some time to add Perplexity APIs for accounts without 10-Ks, earnings calls, non public accounts since their web research is so good.
This can and is distributed across the team - mostly one other person right now, who can access, review, update the workflow in Zapier. But it definitely takes some skill and training for folks who don’t have a natural inclination. Most Ops folks are well primed to understand it though!
Love this Lily! I feel like we've been saying this for decades now regarding marketing tools & technology: the value is realized from the strategy, alignment to GTM goals, config & practical application/integration of them, not from the capabilities of the tools themselves. Love that you're building that AI-forward culture in your org! I'd just add that for marketers unsure on where to start with that practical application of AI or for teams that don't have a GTM Engineer or mature Ops function, it's always helpful to think outcome first. It could be small or large - like 'we want to reduce the demo no-show rate' or 'we want Sales to have more impactful first calls to improve conversion to Stage X opps' - but that level of focus makes it much easier to build the systems & processes to deliver on that goal and test what works/what doesn't. Then you'll find a level of momentum & excitement internally about 'what else can we improve!' to further inform your AI infrastructure. And by then, they'll be 50 more AI tools on the market to look to for inspo 😁
Thank you! And yes, exactly, this isn’t a new insight for ops and martech folks. We’ve been saying for years that tools don’t solve problems, the right processes and strategies do. AI just raises the stakes because the hype is louder.
I love your point about starting with outcomes. That’s how the Analysis Dossier got started - leadership asked “how can we make sellers more efficient at account research?” Not “how can we use AI?” Starting with that one use case can totally build momentum. Once people see one workflow actually working, the “what else can we improve?” conversations start happening naturally.
And yes, by the time anyone reads this there will probably be 50 new tools to evaluate 😅
Couldn't agree more. How do you pitch 'architecting' over 'Copilot'?
The way I frame it - Copilot makes one person faster. Infrastructure makes the whole team faster. Copilot = 1:1 help. you ask, it answers, you get a little more productive individually.
Architecting gives you scale, standardization, and automation. You build it once, and it runs for everyone. The Analysis Dossier generates hundreds of reports without anyone writing a prompt.
When pitching internally, I focus on the output and ease of use: “What if every seller had account research ready in 5 minutes instead of 2 hours—without doing anything except ticking a box?” That usually gets attention faster than explaining the architecture, and in fact what I had started with in my enablement sessions.
The distinction between AI that assists vs AI that performs is spot on. Grounding the Analysis Dossier in real source data through APIs completely sidesteps the hallucination problem. How do you handle the maitnenance burden when these workflows span multiple systems? Does your team need to understnad the full pipeline or can different people own diferent segments?
Great questions! And yes - when I figured out how to grab the full texts from 10-Ks (SEC API), and uploading the PDFs through ChatGPT native assistant function for analysis, it was a game changer. Made the results so much more specific, relevant, and salient.
I actually maintain the Analysis Dossier workflow myself, which doesn’t take a lot of time. Most of this is fixing any issues (rare, maybe timeout issues if it’s being run too much? But Zapier reruns those automatically) or new requests to add sections or something.
Most of the time I spend is on improving the report, like figuring out new ways to make it accurate. Last week for example, I split up a part of my workflow for accounts with and without an Annual Report PDF to streamline the process. I had it all in one because it was easier to build at the time.
Next week I might take some time to add Perplexity APIs for accounts without 10-Ks, earnings calls, non public accounts since their web research is so good.
This can and is distributed across the team - mostly one other person right now, who can access, review, update the workflow in Zapier. But it definitely takes some skill and training for folks who don’t have a natural inclination. Most Ops folks are well primed to understand it though!
Love this Lily! I feel like we've been saying this for decades now regarding marketing tools & technology: the value is realized from the strategy, alignment to GTM goals, config & practical application/integration of them, not from the capabilities of the tools themselves. Love that you're building that AI-forward culture in your org! I'd just add that for marketers unsure on where to start with that practical application of AI or for teams that don't have a GTM Engineer or mature Ops function, it's always helpful to think outcome first. It could be small or large - like 'we want to reduce the demo no-show rate' or 'we want Sales to have more impactful first calls to improve conversion to Stage X opps' - but that level of focus makes it much easier to build the systems & processes to deliver on that goal and test what works/what doesn't. Then you'll find a level of momentum & excitement internally about 'what else can we improve!' to further inform your AI infrastructure. And by then, they'll be 50 more AI tools on the market to look to for inspo 😁
Thank you! And yes, exactly, this isn’t a new insight for ops and martech folks. We’ve been saying for years that tools don’t solve problems, the right processes and strategies do. AI just raises the stakes because the hype is louder.
I love your point about starting with outcomes. That’s how the Analysis Dossier got started - leadership asked “how can we make sellers more efficient at account research?” Not “how can we use AI?” Starting with that one use case can totally build momentum. Once people see one workflow actually working, the “what else can we improve?” conversations start happening naturally.
And yes, by the time anyone reads this there will probably be 50 new tools to evaluate 😅