Support Stack E16: The 14-Factor Checklist That Makes Fin Actually Find Your Content with Dawn Perrott (Fin)
In this episode of Support Stack, Dawn Perrott, AI Knowledge Manager at Fin (formerly Intercom), shares the 14-factor content readiness framework she uses to make help content work for an AI agent. A human reader will push through a confusing article; Fin won’t. If a section is ambiguous, an image has no alt text, or the steps are buried in prose, Fin skips it or gets it wrong — and the customer is the one who feels it. The 14 factors are the things that consistently break AI retrieval and human comprehension at the same time.
Dawn walks through all 14 — from disambiguation and alt text to question-shaped headings, self-contained sections and the “jobs to be done” opener — then shows the framework in action. She runs a single Operator prompt against a real article, and it reports exactly which factors pass, which fail, and what to rewrite. She’s saved the prompt to a text expander so the whole check is one keystroke, and re-runs it after each edit to confirm the fix has landed.
The throughline is that Operator does the heavy lifting, but your product knowledge is still the quality gate. It runs the check and drafts the proposals; the knowledge manager approves what’s right and pushes back on what isn’t. If you own Fin content, run a help centre, or you’re an Intercom admin trying to lift answer quality, this is a practical look at getting your content ready for an AI agent — the how and the why, not just the what.
🔗 Resources mentioned:
AI knowledge management workflow — from setup to content readiness
Optimise your help centre for Fin AI Agent (includes a downloadable Style Guide example)
NPI: How to ensure your agent is ready every time you ship
Want daily, practical tips on getting the most out of Fin and Intercom? Subscribe to my daily email list.
Episode transcript
Conor Pendergrast (00:00)
Hello and welcome to episode sixteen of Support Stack. So, Dawn Perrott from Fin, AI Knowledge Manager. you are back for a second episode to talk about another exciting topic. We teased this in episode fifteen, didn’t we, Dawn? Yes. So today we were talking about how many how many factors do you think would be a good number of factors for a content readiness workflow or checklist, Dawn?
Dawn (00:22)
These is I’d go with the fourteen.
Conor Pendergrast (00:34)
Fourteen is great. Fourteen is good enough. You can have one every day for two weeks and you got a fortnight worth of worth of content readiness at your fingertips. But let tell tell me all about this. Like why why is this so important? Can’t you just can’t you just guess? Can’t you just have operator guess about what it what changes it should make?
Dawn (00:54)
So each factor represents something that affects whether a customer can understand your content and whether Fin can retreat can retrieve it and use it accurately. a human reader might push through a confusing article, but Fin won’t, Finai Agent won’t. So if a section is ambiguous, if an image has no alt text, if the steps are buried in prose, Fin will skip it or get it wrong. So the 14 factors aren’t arbitrary. They’re the things that consistently break AI retrieval and human comprehension at the same time.
Conor Pendergrast (01:28)
Super. And so you’ve developed this you you’ve developed this with within Fin, the company, and have applied this to your content. And now you’ve also like you’ve made it available in the help the the help center. So I’ll I’ll put a a link to that in here. So it’s specifically in this section, the 14 can factor content readiness framework. But I think what might be useful is just if we briefly Talk through the the fourteen factors and then see see the see the framework in action. Does that sound good? Cool.
Dawn (01:59)
Yeah. Let’s go through the functional factors now.
Conor Pendergrast (02:02)
You go for it.
Dawn (02:04)
So this is the content readiness I’ve saved to my test workspace, but it’s exactly what I have saved in my Fin workspace. so we have 14 factors. So the first one is disambiguation. So this is to avoid references like this screen or the field shown above. Every fr every reference must be self descriptive, no emojis pointing, because Fin isn’t not isn’t able to understand that. The second one is visual content text. Every image must be have descriptive alt text, a trailing colon, and image with no alt text is not acceptable. Describe the visual and the text too. So this is when Fin is sending images in its replies. By adding the alt text to the images, Fin will know exactly what is in that specific image. Are you familiar with the adding alt text to to article content, Conor, or will I show you what that looks like?
Conor Pendergrast (03:03)
I have seen that. I I suspect people will will know what it is, but what what I’ll do is just just in case, we’ll include a link to that in the in the show notes as well, so people can see what that means. Alt alt text being alternate text, right? It’s just it’s just in this case giving Fin or or a screen reader as well, for example, for people who are visually impaired, it’s a great way of giving them information about what is in the image. Because ultimately fin
Dawn (03:18)
Yes. Yep.
Conor Pendergrast (03:32)
Fin’s not looking at the image at all, right? In the help article.
Dawn (03:36)
No, Fin isn’t looking at the image necessarily, but it’s Fin is pulling the context of what content is around that image. So by specifically putting the alternative text on it, you’re making it far more accessible to all.
Conor Pendergrast (03:43)
Mm. Yeah, that’s perfect. So it’s a good accessibility step and it’s a great step to also improve Fin’s ability to use the images that already exist in your help center. Perfect.
Dawn (04:00)
Exactly. Then number three is undefined terms. So this is defining abbreviations and product specific terms on first use. So if you’re going to use GDPR, I suppose GDPR is quite a universal one and two FA, but you never know who your audience is. And it’s always good to explain what those are initially, and then you can abbreviate them further down in the content. number four, structured enumeration. So Multi-step processes must be in numbered lists. Lists of item lists of items must use bullets, bullet points. Never describe steps in flowing prose. number five, query answer symmetry. Headings should mirror how a customer would phrase a question. Prefer how to or question form over bare noun phrases or bare imperatives. Number
Conor Pendergrast (04:47)
that one’s a little bit more maybe a little bit less clear for me just just just reading through it. ‘Cause it’s a l it’s got complicated words. Query, answer symmetry and bear imperatives. Now, I’ve never I’ve never had a bear involved in Fin. Haha.
Dawn (05:00)
I know this is rather than just ha heading your sections and your content what a product is, like let’s just say tasks. You have a section on or let’s not say tasks, because that’s let’s say notifications. You have a section called notifications, and Fin isn’t going to be able to read that when it’s answering or pick it up as easily if it was. in answer or sorry question format. So rather than saying notifications, say how do I set up notifications?
Conor Pendergrast (05:35)
yeah. Yeah. Yeah. Okay. So you’re you’re you’re exactly at a your the customer would phrase a question a certain way, use that phrasing.
Dawn (05:44)
Exactly. Give give the headings context because your H ones, your H twos, your H threes, they’re all really important for Fin’s answer quality as well. Fin will Fin has a priority for answer retrieval and it is looking f at those signals. So it’s important to make sure you have have that outlined as well.
Conor Pendergrast (06:07)
Perfect. And then six self contained sections.
Dawn (06:10)
So every section must make sense if retrieved independently from Fin. Avoid as described above, then openers are referenced to f references to prior steps without restating them. So again, context is key. It’s very important that you give context to Fin what’s what’s happening in this section if it if you’re just launching into something entirely new. audience specification, state who the context is for and what permissions or plan access is required. So this is a check every time you’re generating new content. it’s to see if there needs to be an audience applied or call out applied to the content saying who this article is for. So are you on the plan?
Conor Pendergrast (06:49)
Hm. And so that’s am I right that that’s both checking the actual like fin AI agent audience as well as just saying in the article, in the content itself, who it’s for, or is this just for the content side?
Dawn (07:04)
This would just be for the content side. so usually what I will say, I will if I’m creating some new content for a new feature that’s going out, and if it’s for a specific audience, I will say that in the content and it will be called out. It’s something that’s important to question when you’re generating new content, because you need to make sure it’s it’s pointing to the to the audience that that it relates to. But Some some articles are for everyone, and it doesn’t have to be for someone who’s on expert plan. it would be cool if it was also picking up on the audience applied to the content. you you how we use that functionality in our own workspace would be we would actually we would say to operator
Conor Pendergrast (07:35)
Yeah, yeah. Great.
Dawn (07:53)
This update is in relation to Fin standalone. Can you update the Fin standalone content? Or we would say this up update is in regards to the intercom help desk. Can you update specifically the intercom help desk content? So it is actually able to pick on, pick up on this is for this audience and this is for this audience.
Conor Pendergrast (08:10)
Got it. Okay, that’s great. Yeah. Super.
Dawn (08:13)
So number eight, entity distributions. So re repeat key entities, product feature names, and article core topics throughout, not just in the opening. Again, for context for Fin, semantic chunk boundaries. Each section should cover over focus topics, break long sections with meaningful subheadings. Number 10, restate questions, tables and standalone blocks must include enough context context to be understood without the heading hierarchy, and add an intro sentence before tables.
Conor Pendergrast (08:23)
Mm-hmm.
Dawn (08:41)
Again, that’s an important thing for accessibility. overview jobs be done. The upper opening paragraph must state the jobs the reader will accomplish. Use this article to do y. is this is stronger than this article covers X. Instruction completeness, step by step instructions must be completed end to end, including what happens after the final step. number 13 limitations and workarounds. Document known limitations, gaps, and workarounds explicitly say which one. which ones and what to do instead and numerical clarity. I in regards to these 14 factors, I wouldn’t say one is the most important. I think they all are doing a specific job to help to help with the customer in the AI.
Conor Pendergrast (09:19)
Yeah, yeah.
Dawn (09:22)
I can show you an example of content I queried against an article from my test workspace and I can see I can see exactly where it failed and where it passed. So if you actually if you’re looking for prompts on how to query operator, I have them in the AI Knowledge Management setup article. And I have a section that’s dedicated to what prompts to use. So the prompts I use most. So
Conor Pendergrast (09:28)
I think that’s a great idea. Perfect. Okay, great.
Dawn (09:49)
This is the one that I actually have used in the example. So how does this article measure up against the content readiness framework? And just to speed up my workflow, I’ve actually saved these prompts to text expanders. So if I wanted to use this, I would just go like I would type my sh my abbreviation and it would auto-generate quickly because I use this I use this quite regularly. So this is just an example article that I generated for our demo today. And
Conor Pendergrast (10:02)
Grace. Super. Mm-hmm.
Dawn (10:18)
When I queried it against that prompt, how is this article against the content readiness framework saved in your memory? and basically what it did what the pro what operator executed then, it looked at the content readiness framework against all 14 factors and it told me where where it was passing and what factors were passing and what factors were failing.
Conor Pendergrast (10:38)
Super. Yeah, that’s really helpful. So it’s so this is going step by step through the whole content framework and saying, listen, here’s what’s what it’s not living up to according to the standards that we’ve set. And will I mean all of these fourteen steps will impact on Fin’s ability to read the content, interpret it correctly, and use it in the correct context to to answer customer questions and just create really frustrating customer support experiences for for for your customers. And so then it then it goes through and it suggests the suggests the rewrite. Now I presume I presume there’s gonna be cases where it can’t it can’t fully suggest what to change because it might lack the context itself on what like what an abbreviation might be, for example. But it’s doing does is operator like comparing with the rest of the knowledge base in in this in this step as well and saying, okay, well I don’t know what JTBD means in this page. But I’ve just searched and I’ve understood that J T B D means jobs to be done because of this other this other knowledge knowledge article.
Dawn (11:41)
So if there is something ambiguous that operator doesn’t know what the answer is, it will ask you the question, is this for this audience? Is this for this audience? It won’t it won’t add in content based on something else necessarily. It will always ask for the clarification from whoever is accepting the proposals, because at the end of the day, it is.
Conor Pendergrast (11:50)
Super.
Dawn (12:08)
knowledge management knowledge manager sorry the knowledge manager’s judgment that’s still actually very important to how AI is answering and the customer experience. the workflow and the process to update content has been made it’s like a lot quicker, but at the end of the day it’s still really important that the the content that gets added is accurate and that it is up to date.
Conor Pendergrast (12:35)
Yeah. Yeah, ul ultimately so so am I right, like you came you came to this role from the customer support team and have a huge and as a result you have this wealth of knowledge about your company, Fin, your products and your customers. And like that’s that’s what you can apply. Operator that makes that a lot more efficient, but your expertise as an AI knowledge manager and your pa massive past experience in working directly with customers is what makes this so effective. If you didn’t know any of that, you would just come in and operator could suggest something and then you wouldn’t you wouldn’t know that that didn’t make sense in the in the product context. So you you still need that expertise and you still need that that customer insight to to do a really great job.
Dawn (13:25)
Yeah. operator runs the check and reviews the proposals. I’m the one that approves what’s right and push back on what isn’t. And those fourteen factors give you a systematic checklist, but your product knowledge at the end of the day at the end of the day is still the quality gate.
Conor Pendergrast (13:42)
Absolutely. I would say how I would probably look at this is this is another way where I don’t get to be lazy because I don’t know about you, Dawn, but my default in a lot of cases is just like do the the quickest path to get something done. And certainly I remember writing help articles in the past, that were nowhere near as structured as as they need to be now to work effectively for an AI agent. And and having a specific framework like this fourteen fourteen factor content readiness list is is perfect to to force me to be less lazy, but also in a way where I don’t have to do a huge amount more work, which is important for me.
Dawn (14:22)
For sure. It’s really it really has sped up my workflow as well. even on existing content, I will run the prompt and it will flag things that are not making sense that we already have in the help center and it will tidy that up. another thing I just want to mention is sometimes it’s the case that not all 14 factors are updated in one go. So you may need to apply you may need to use the prompt. multiple times. So I will sometimes what I will do is I will if I’m working on one article, I’ll obviously go through the first check and I’ll approve the proposal. And then to make sure that it has actually updated the content, I will ask it to reevaluate it. So it will run through those 14 checks again and it will tell me if it’s been updated or if it hasn’t.
Conor Pendergrast (15:11)
Super. Yeah. And and I mean often if you I guess if you run the article through, you’re gonna want to rerun it again anyway, because now you’ve got a brand new article that you wanna make sure that that all that all works. Yeah, this is really interesting. Okay, so it’s running through again. Yeah. Okay. What is there anything else you’d like to share about the the the content readiness workflow? Anything else that like I I gotta be honest, this seems like something that should just exist in operator.
Dawn (15:39)
There is a I believe there is part of the LLM that there is a content check there already. this is just doubling up on it to make sure that it gets applied.
Conor Pendergrast (15:46)
Brilliant. Yeah, yeah. And there are and like you know, like we said, there are probably gonna be some some disagreements with with with parts of it. Some of them seem pretty pretty obvious and and all like the image old text one. Obviously that’s something that you should always happen. that it should you should always have. Well this is fantastic. I I can see this being another really helpful episode on to to improve the quality of content and get get get Fin working even more effectively. but also like this is a great framework even outside of Fin as well. I hate to say it, but but other AI agents are available. Not not that they’re any as as good, but but this is fantastic. Dawn, if if people would like to to learn more from you, that you have other maybe you have other things to share in the future, where where should they go? What’s what’s the best place to find you?
Dawn (16:40)
Yeah, if folks want to follow along for more knowledge management tips, they can find me on LinkedIn. Dawn Perrott is my name.
Conor Pendergrast (16:47)
Lovely, and I will put a link to that in the down below in the YouTube description. So cast your gaze downwards and and find her there. Click the linky and press the follow button. this has been a really interesting episode. We’ve got one more about Operator that I’m excited to to share with everyone too. But for now, if you well, if you wanted to see that episode, make sure to click the smash that subscribe button. Give me a like as well to feed my ravenous ego. And if you want to get daily emails from me about Fin, you can do that customersuccess.cx/daily. yeah. Thanks Dawn and thanks everyone for watching. See yas.