Back to all articlesBuild, Buy, or Wait, with Brian T. O’Neill
Lucas ThelosenLucas Thelosen
Podcast

Build, Buy, or Wait, with Brian T. O’Neill

Directionally Correct: Data 2030 is an educational podcast for data and analytics professionals, hosted by Lucas Thelosen, co-founder and CEO of Orion by Gravity.

In this episode, Brian T. O'Neill of Designing for Analytics takes the buyer's seat and asks Lucas the questions he hears most often from enterprise data buyers. They dig into whether to build, buy, or wait on AI analytics tools, how a pilot should measure success, and why counting adoption alone can mislead.

Brian T. O'Neill is founder and principal of Designing for Analytics, where he helps B2B analytics companies translate technical complexity into commercial clarity. He works on the vendor side, with founders and product leaders whose products are technically strong but hard to sell. He also hosts Experiencing Data, a top 2% podcast globally per ListenNotes, with over 200 episodes interviewing CEOs and founders of B2B analytics and AI products like Orion as well as CDOs and heads of enterprise data science and analytics. Follow him and his podcast at https://designingforanalytics.com/ or on LinkedIn at http://designingforanalytics.com/linkedin

What you will learn

  • Why building an AI analytics tool in-house is quick to prototype and costly to maintain
  • Why feature parity no longer wins deals, and why speed to clear value matters more
  • Why internal teams often underestimate the product design work behind software people want to use
  • How analytics that recommends actions and tracks decisions can tie to ROI
  • How a pilot can measure success through the end user's experience, from frustration to delight
  • Why adoption numbers alone can mislead, especially when a tool pushes insights to users

Chapters

0:00 Welcome and Brian takes the buyer's seat
2:49 What buyers should ask, and the build versus buy debate
7:21 The real cost of building and maintaining your own tools
10:41 Why internal teams underestimate product design
15:01 An AI analyst that runs a sales forecasting workflow
16:45 Why not wait for the AI labs and data platforms
21:30 Value beyond cost savings, and recommended actions with tracked ROI
27:08 Descriptive versus prescriptive analytics, and data literacy
29:44 Should people record their decisions? Accountability and proxies
34:07 Competitive culture and who a tool like this fits
38:27 Why not wait six months, and the widening AI gap
42:08 Measuring a pilot, and why adoption alone can mislead
48:26 Advice for buyers, measure the end user's experience
52:27 Keeping analytics from bloating, and the data team's role
54:15 Product and strategy over engineering, and the question of a moat
56:51 Closing thoughts

Transcript

Welcome and Brian takes the buyer's seat (0:00)

Lucas Thelosen

Welcome to another episode of Directionally Correct. With me today is Brian T. O'Neill, a fantastic person I met on the topic of data product management. We really thought about how to manage data as a product, and he has built a fantastic community around this. He's actually advising a lot of B2B analytics companies, just like Orion by Gravity, on go-to-market and how to position yourself, how to message yourself, and how to design your product for the audience, for the data people out there. It has been really interesting talking with him. I actually got to join his podcast a while back, which was really fun. I think it's one of the top podcasts in the world, top two percent, if I remember correctly, globally, which is an amazing achievement, with over 200 episodes. And Brian has offered to come on and actually be a very skeptical enterprise buyer and just give me the hardest questions that I get all the time and see how I react. I don't have the answers here, so let's see how I do on the spot.

Brian, thank you so much for being on the show.

Brian T. O'Neill

I'm happy to be here. It's good to get to do one of these and flip the roles a little bit here.

Lucas Thelosen

Yeah, exactly. Well, why don't you give me, or us, a little bit about your background? I only touched the surface.

Brian T. O'Neill

Yeah. So I run a company called Designing for Analytics, and I help B2B analytics companies build data products that sell themselves. The primary lens through which I see that world, and how I do that work, is trying to translate technical complexity into commercial clarity. Most of my work tends to be more on the quote-unquote vendor side, but for many years I was serving both vendors in my consulting work and also the enterprise data crowd, who are the buyers of solutions like Orion's. So I have a perspective now from both, from the in-house data analytics and data science teams that are using solutions like this to build internal enterprise value, as well as the people building the tools for them to do that kind of work.

One thing I've been changing on my own podcast is mixing my guests between founders, operators, CPOs, and CEOs like yourself, and enterprise data leaders, who come on to talk about their experience buying versus building, how pitches are being received, and why they're not adopting the new software platforms and things that are coming out. There's so much stuff happening right now. So it's kind of fun to have them back on in a different kind of role, instead of talking about data product management and how they work with their teams internally. We're talking about their experience now as buyers, and sometimes as users, of solutions like yours.

What buyers should ask, and the build versus buy debate (2:49)

Lucas Thelosen

No, amazing. I often think this is an amazing time to be a buyer. All of a sudden prices are dropping, features are being added, and the market has never been this competitive, so it's a really good opportunity. At the same time, the market can be quite flooded, because all these AIs generate all this content. Marketing-wise, everything sounds very similar. It's really hard to cut through the noise and understand what's real and what's not real. So what are some of the top questions a buyer should be asking right now? If you're with a legacy product, you're probably overpaying, because those vendors don't adjust their prices. If you go out in the market, you'll see much, much lower prices with feature parity oftentimes. But still, how do you get to the bottom of whether this is a good product for us?

Brian T. O'Neill

Yeah, there are lots of angles on that. I think obviously the elephant in the room is buy versus build. The top use case for copilot tools right now is writing code. We can do it ourselves. I've had just a couple of interviews recently about this, and I've gotten multiple perspectives. On the one hand, you have VPs of data who say, "We know enough about what we don't know that supporting an internal build like this, now we're basically becoming a software company. You have to sustain and maintain this. You have to feed it and its features. You have to think about support when it goes off the rails, all these kinds of things, on top of your day job. Is that really a skill that we have in house and that we want to do?"

On the other hand, I just recorded an interview with a former chief data and analytics officer at a bank, and he said, "Frankly, we've had these bake-offs before, as they sometimes call them, the buy versus build. And I had a team of three people in a weekend get us something that everybody arguably thought was better for our needs than what the vendor was pitching to us." In that particular case, I don't know the details enough to know how deep it needed to go, or the maintenance of that, and sustaining it and patching it and security and all those other kinds of things that technical founders know about, which I think end users sometimes aren't thinking about. In this case, it was good enough.

So what I think is happening is that the value has to be obvious. This is a place where I work with a lot of B2B technical founders. They typically come from a technical background, software engineering, data science, computer science, finance, things of this nature. And I don't think feature parity is really the way you compete anymore, because you can copy stuff so fast now. The real issue becomes how fast and obvious the value is to the user, not necessarily the scope and breadth of the product. Whether that's in the marketing of how you talk about it, in the user experience of the product, or in how you sell the product, it's probably a little bit of all of those. It gets to how obvious that value is.

What I think I'm hearing that's working is getting in fairly quickly with obvious value fast, where the value is either obvious to the user because it's new value they didn't even know to think about until they saw what you offered, or they have some fairly discrete metrics, a plan of measurement for how they would evaluate a solution like yours or anyone else's. They're data people, and they're dogfooding. I don't think a lot of people are always dogfooding, and this is the funny thing with the data community. We're always saying, "Do you need to be data-driven in your decisions?" And when it comes time to look at these things, I'm sure there are founders wondering, "How could they not have gone with this?" or "I can't believe they're moving forward with that." It's not all about the numbers. We're human beings. Data people are still human beings. There are lots of angles we could talk about here, but that's kind of my state of the union, so to speak, as of October twenty-first, when we're talking. And next week it'll be different already.

The real cost of building and maintaining your own tools (7:21)

Lucas Thelosen

Yeah. The build versus buy. I feel like I'm back in 2009, when everybody was building an app. There's an app for it. You build an app for it, and I want to build an app for it too. And I was frustrated. I'm a CEO of a well-funded tech company, so I get about 15 emails and LinkedIn messages per day for go-to-market tooling. Some software that helps me with my sales, and we'll get you leads, and all these things. It's really hard for me to cut through. And then I think, I'm a unique company, I feel like in some sense. Everybody feels a little bit like they're unique. So why not just build it myself? So I opened up Claude Code and I started vibe coding. It was fun. I set it up over here on a workstation, and it is really fun. You can prototype something quickly and show it to the team and they're excited, and I think it's a good experience.

The problem then becomes actually maintaining it. A couple of weeks later, there are features I didn't realize I really needed, and how do I add them now? So it becomes a bit of a challenge. And then I looked up, okay, the price is actually not too bad for some of these off-the-shelf products that I can get, where there's a team dedicated to thinking about this. So I've been thinking, internally, what are the things we're going to build ourselves versus buy? We actually considered GitHub. We use GitHub a bunch, and there have been a bunch of outages, so we thought, maybe we should build our own. There's open source already available anyway. And Slack is super sensitive, because all our internal communication goes over it. So why not build our own Slack application? We did a prototype, and it cost about $600 in tokens. It was super cheap to build our own Slack application. But then, do we want to maintain it?

So I always get to that. How much time do I dedicate now to maintaining this application versus paying someone for it? That's what I get to with build versus buy. I do think it's a good experience to try to build something, especially something you're passionate about, something you're interested in. I do think that will go down a little bit, especially as people move on and there's maybe something running, an agent running around that is doing things, and we're not exactly sure what they're doing. That gets a little scary as an executive in a company. How many agents are running around, and how many jobs are scheduled? Where's my control panel so I can see what's happening in my organization?

So it'll be really interesting to see that over the next year, through 2027. I'm sure there will be some discussion around visibility into everything. Anthropic and Meta both published a great paper on how they built an analytics application internally and what it took. The prototype was easy, but then it took X number of engineers. I think in one case about twelve, and in the other case almost two dozen, or a little over twenty people, to get into production. And now there are a couple of engineers full-time assigned to maintaining it. That's of course larger organizations. You might not need quite as many for a smaller organization. But I think that's a thing we don't really talk about yet. It's so exciting, it's so new, it's fun to build something quickly. But is that the thing, the core to your business, that you want to maintain?

Why internal teams underestimate product design (10:41)

Brian T. O'Neill

I had very similar client conversations recently. It's a CRM company that I'm working with, and they felt like, you know what, we should just use our CRM product for our own sales CRM and stop using the provider they were currently with. Some of it is, when someone onboards, we need to touch these drop-downs, which activates something in the product, and it's easier if it's all in there. There are some fair arguments for that. My feeling, having come up through a product design background and thinking about user experience through the lens that I do product management consulting now, is that there's so much you're probably not thinking about, even with Slack. You mentioned Slack, and we tend to think of the chat interface and the real-time part of Slack. But then you get into, I need to search the messages, and I need multi-threaded messages. Oh, and when should we notify someone if we haven't heard back from them? I don't know if you've seen it, there's a workflow diagram that went out showing when Slack decides to send a notification or not, and whether it's email, text, in-app, or whatever. There are probably 85 nodes on that screen of thinking that went into the experience of when it's appropriate to send a notification. That's just one aspect of Slack. There's the real-time chat thing, which we tend to think about right now, but there are all these other downstream things that add up into a product, into an experience that's worth buying.

So I think that's the trap with the build thing. The thing we can picture in our head, we can get to really fast. And I think that's sort of true. If you can picture it in your head, you can make it. But there's a little bit of overconfidence that what we can make is actually right for everyone else. This gets into your internal data team, for example, building a solution for business users. The issue is assuming that the analysts know how to design good software. We feel like we do. Everybody has an opinion about design and how the screens look, and even the workflow. Everybody has an opinion about that, so we tend to be confident that we're doing a good job at it. But designing is really about someone else's perspective, whether or not you think it's the right thing. It's very much an act of empathy. You can luck your way into producing something pretty good sometimes, and if it's a tool for you and your team, you'll probably do a fairly good job, because you know what you need immediately in the moment. So I'm not saying no one should be trying to design their own products. But when you're doing it for somebody else, that's a whole skill, and it's hard to know if it's working or not.

My general sense is this. Given how hard it's been for internal data teams to often show value inside their organization, isn't the tenure of a CDO like one year, two years? There's high turnover there, which suggests the business doesn't see the value of all this investment in machine learning. Remember, data science was the thing five years ago or whatever, the job of all jobs, and it was going to save us and all of this. I've been hearing this for 20 years from internal data teams, show the value. Now you're going to build software and then show the value with the data inside it. It's a two-hop leap. My general sense is I would really question that if it's beyond small application things. Are you really spending your team's time on their best stuff? Are they leveraging their best skills in the best way possible for the organization, or are they now doing a bunch of skunk-works projects and kind of creating a mess? "We solved this problem, but now, like you said, I have to babysit this thing." We made an agent that's going to go out and do this marketing thing for us every week. And it's like, well, who's watching that and deciding on the evaluation for that? Who's managing that overhead now?

It goes on, and it can snowball really quickly. It's so easy to spin this stuff up now. So anyway, those are some of the things I would be thinking about when I'm making a buy versus build decision.

An AI analyst that runs a sales forecasting workflow (15:01)

Lucas Thelosen

Yeah, no, I think that's super valid. You brought up some really good points there. As data people, we were so focused on building beautiful dashboards, and we'll continue to think about that. We're really focused on architectures and what the foundation underneath is. But then also thinking about the user experience and all of that, it becomes a lot. We use Orion internally for our sales forecasting. Orion actually reaches out to each sales rep. It checks with them, saying, "These are the numbers I'm seeing. These are the accounts you talked to. Is this correct? Do you want to update anything before the meeting?" So when the meeting comes around, everybody knows, everybody agreed on the number, and there's no more "Oh, actually, I didn't update my records." That's a really cool analytical workflow. Almost every organization has something like this. Maybe not for the sales team, maybe for the marketing team or procurement or whatever. Having this AI analyst that actually goes to each person, checks with them, makes sure this is what they want, asks if there are any changes they want to make to the systems, and then moves this analysis up the chain so it can be reported upwards. I don't think any data team would have thought about setting this up as a product, like building that as a product. It's a workflow that does a lot of things. It's more like a project manager, almost, that goes through and checks with everybody.

So it's really great for us. It really smooths out our actual meeting, and we can streamline that tremendously. Well, besides build versus buy, what are the other questions you think an enterprise buyer should be asking right now? For any kind of technology vendor, but maybe specifically in data analytics.

Why not wait for the AI labs and data platforms (16:45)

Brian T. O'Neill

Yeah. Claude Dashboards, Lucas. Why don't I just wait? I already have accounts with ChatGPT, I mean OpenAI, and Anthropic. They're going to come eat your lunch, so I'm just going to wait. Why do I need this interim solution when every week they're in another domain, they're in another space? Maybe we should just be there. Then we don't have to worry about all these tertiary products that need to talk to each other and connect to all these different data sources. I'll just wait. It's too dynamic right now. Push back on that.

Lucas Thelosen

Yeah. No, I think that's a great question. I think that's a super valid question. I can share this. I actually just met with Anthropic a couple of days ago, and we talked about exactly this question. And maybe I make this question even harder for myself. Why not use Databricks Genie, or Snowflake Cortex, or Gemini on BigQuery? Between those giants in the field, each a trillion-dollar company, each needing to make hundreds of billions in revenue to justify the valuations, why a purpose-built analytics solution? I think that's a very valid question to ask. Talking to the Anthropic team, they said, "Yeah, we built this for small companies. This is not for companies that actually have a data team. This is some of the SMB market." But of course they need to go into enterprise. They need to charge a decent amount of money at some point for this. Databricks and the other database providers are of course focused on the data engineering side, but they still need to move into the business as well. So why is something needed in between?

I think that is really interesting. On the one hand, you have a company that wants to use as many tokens as it can. On the other hand, you have a company that wants to use the database as much as possible. So it gets expensive. On both sides, you have someone with a strong incentive to drive up usage. I think it's really good to have something in between, a Switzerland that actually optimizes token usage and optimizes queries run. There are things we can do through advanced caching. We have a great layer in between that first of all makes it really fast for the business user to interact with data, but also means we don't hit the database all the time, and we can federate across databases. So from a pure technology and financial perspective, it makes sense to have something in between that is neutral. Plus, it allows you to switch out ChatGPT, and maybe you want to do it through Slack directly instead of going into one of those desktop applications first. On the database side, you can do it across BigQuery and Databricks, or across spreadsheets, whatever your data sources might be.

That's from a technology and financial side. But also from the data team's perspective, I think of Orion more and more as the cockpit for the data team to control the business user's experience with data. You can think of it, and we're big into space analogies here, like a Houston, a station where you can see everything that's going on, the questions being asked, how they're being routed, and whether they're being routed to the right context layers. When marketing asks something in Slack, it's routed through the marketing context. When Sarah asks something in Claude, it's still routed through the marketing context. So you can see questions coming from different points of origin, routed to the correct context layer, not just data semantics but also business context. What are the goals for the quarter? What should they be focused on? What's her span of control? So then the analytics that come back are very actionable for Sarah, for marketing. So that's the control tower. And as a data team, you now become a context engineer. If you have Claude connected to Databricks, you're out of the loop. There might be someone in engineering who set this up. But in Orion, you are the context manager. You promote context up a level, to the team level or the organizational level, and Orion works with you on making that experience better.

You've mentioned a couple of times how important design is. Internally, one of the first things we built is that Orion administers itself with you. It suggests changes all the time, like, "I'm seeing business users using more of X, Y, and Z. Maybe we should do this, or add a data source over here, or change the layout of this table." It suggests edits. If you use dbt, it can update the dbt model file or something. So it really works with you as an administrator of its own product, which I think is a really cool use of AI. Now you have your copilot, in essence, your copilot for data analytics experiences for business users.

Value beyond cost savings, and recommended actions with tracked ROI (21:30)

Brian T. O'Neill

I get all that. Some of that sounds defensible to finance and procurement, about why there's a cost-savings angle here. But from a new value creation perspective, as opposed to "this is a cheaper alternative to doing it in-house," what do you think? This is a question I've heard buyers asking sometimes: how are other teams measuring it? Even in a POC. I don't know if this is data buyers genuinely not knowing, or wanting to see how the seller of the solution thinks your solution should perform. You should count these metrics, not those. That's the wrong thing to count. Man-hours of analyst time saved is not the right metric. I could understand how maybe they're asking a founder like you to help them look at the right metrics.

But if I were to demo, if I'm thinking about bringing in a solution like yours, let's say I heard in the elevator at CDOIQ or some conference that your name was Orion. All I heard was it's one of those AI for BI, talk-to-your-data solutions. They do dynamic dashboard development in there. There's some kind of shared semantic layer. It sounds really interesting. You should check it out. End of elevator ride. So we're thinking about that. We want to see what's in the market. We've been sitting on Power BI and Tableau forever, and we know the market's changing. We don't know where we should be. I'll take your call, Lucas. So tell me about your product. Why should I need you? Are you just going to help me get rid of my analysts? Because I don't really want my staff to lose their jobs, although I'm supposed to cut costs, but I'm also supposed to create value. So how can you help me beyond automating the act of dashboard building and doing one-off reporting and things like this? How should I think about demoing you? And how would I know if this is right for my business?

Lucas Thelosen

Yeah. I think that's a really good question. ROI in analytics has always been a challenging one. I have a little unfair advantage here in that I ran product at Google for data and AI, and I was at Looker beforehand, so I know a bit about internal analytics. Now, one of the challenges I think almost all legacy dashboarding products have is repeat users, people who come back every single week, ideally every day, but at least every week. The marketing team is meeting once a week, so they should be looking at data at least once a week. And people don't. People don't want to look at data. They want to make decisions based on their relationships in the company and the gut feeling they have. That's the reality that we data people don't like to hear.

So one thing we set out to do, and that's why we focus on business context so much, is ask who Sarah is, what she cares about, and what her span of control is, in addition to the semantic layer. We want to make it actionable. We want to say, here's what we're seeing. We peeled back the onion of the underlying metrics and why this is happening, and here's what you can do about it. Here are your three options today, or your two options today, and here's the estimated ROI based on these options. If you do this, that's the estimated ROI. If you do that, that's the other one. The easiest example is that you can either try to focus on increasing revenue or on increasing profitability. Those are two paths that are both very defensible. But which one are you going to choose in your role somewhere down in the organization? You take steps and actions every given day. Now we have an ROI calculator that sums that up. Here are the decisions Sarah took, and we can track that. She decided it was option A this time, and this is how much money she contributed with that decision. And you can have a leadership board, so you can actually see this. All of this is approximate ROI. This is not hard math, but it's really interesting to see.

And this was not possible with dashboarding products, because dashboarding products just show your number. They don't show you what to do about it. So we really wanted to take it much further, to the action you should be taking. I think it's really cool as an executive now. We have this large retailer that shifted its focus a little bit for this quarter. You can give Orion that context, that I want to shift the focus a little bit. We're not pivoting, but we're adjusting the goals for this quarter by focusing more on apparel, something we haven't done as much before. That is a subtle change in analytics, and in how Orion, for example, would then talk to you about what's happening. Previously, in traditional products, you would have to go into each dashboard and each slide deck and figure out how to talk about the performance of this category versus the other one. If you're not aware that it's a strategic goal to push one as a loss leader, maybe, you constantly recommend going back to the bread and butter of the business. Now we're able to influence hundreds or thousands of employees to be aligned to the goals of the company for the next couple of quarters. That's a really exciting thing for an executive. You can say it in the all-hands meeting, you can say it six more times, but for people to actually take action, it's great that you can put that into the context of a product like Orion. Now all the analysis for every team, for every meeting they have, is slightly adjusted to be focused on the goals you have.

Descriptive versus prescriptive analytics, and data literacy (27:08)

Brian T. O'Neill

Can I ask you a question about this? What I'm hearing, and this goes back a decade, I remember these old slides about historic analytics, then descriptive analytics, predictive analytics, and prescriptive analytics, which used to be synonymous with machine learning but is more "here's the number and what to do about it." Or in your case, "here are three options given the number that's provided here." So are you saying the value of Orion, since I didn't see your pitch in the expo hall and we're talking on the sales call here, is that instead of providing primarily historical rear-view dashboards and tooling, I'm going to be able to give my business users forward-looking recommendations and options? They'll be able to track the decision they made, so we can see how that's performing and what kind of ROI is coming with it. And the context will be spread out, so if marketing decides to double down on spend over here, or they're going heavy on video now instead of advertising or whatever, then another department would know about that, or somehow that information, when relevant, would surface to another department. Something along those lines?

Lucas Thelosen

Exactly. If you look back at the talks I gave over the last decade at Looker, I was always talking about data literacy and education and getting people to actually understand what this line graph means for their job. It was so hard to build the data culture. You're really reliant on data teams taking care of jobs that are really outside of their scope, but they're champions of data in the organization, so they're doing it. So I was just thinking, how can we make every organization as data-driven as they were? I was at Uber, I was at Amazon. And now, because you can add the business context to it, you can make it incredibly actionable. I can tell the foreman at the construction site, this is what the data is telling us about your day to day. This is what you should be mindful of. And here are the options, the decisions you can make, and the consequences of those decisions. So you can make it incredibly practical. You don't have to look at a line chart anymore. You don't have to understand how a pivot table relates to your job. You can actually hear what the recommendations are, because Orion understands what you do day to day, what actions you can take, and the consequences of those actions.

Should people record their decisions? Accountability and proxies (29:44)

Brian T. O'Neill

Got it. Can I put my skeptic hat on? I want you to push back on this. That sounds amazing, but I'm not sure my foreman wants to tell the tool what decision he made, because now there's a recording of the decision I made. That's great if it's good, but you're getting into accountability now at the human level. How good am I at making decisions? The higher up the stack you go, the more people lean on experience. And what's the price of making a poor decision? I would wonder if that takes a fairly mature, very intelligent, very safe organization, where the team would be willing to do that, unless the product provides outsized value to offset all the risk that might be associated with that. We're not looking at it right now, but these are just some things I would think about from a design perspective. How can we get someone to record their decisions? Not everyone's thinking, "I want to place bets." I actually use this bets model that you hear about in product management a lot. When I work with teams like your own and talk to product people, you're here to deliver outcomes, not outputs. And each output we make, each feature we make, is essentially a bet. It's a bet to see if we can improve the lives of customers. If it's a user experience bet, or a business bet, we'll see more revenue, or we'll cut cost if we cache this thing and we don't hit the database here. We actually think we don't need to do that more than once a week, but we're not sure people won't stop using the product. So our bet is this and this, and then you go and measure and you see that. That's a fairly mature concept, I think. And I wonder, with analyst teams, especially people working more in cost centers and stuff like that, is there a resistance to recording? What's the downside of this?

Lucas Thelosen

Yeah. I think of it as we're making data-supported decisions, data-informed decisions here. So there is a lot more safety to this than making decisions that aren't at all based in data. I don't think you should blindly go with whatever the system says. Companies are not built just because machines are really smart. They're built because there's a human element to them. I actually think what we have here now is a lot more human-centric than the dashboards of the past, because we added the business context. We added who Sarah is, what she cares about, which dashboards were never aware of. But there is a hesitancy, maybe not a hesitancy to record, but it's sometimes challenging from a technical perspective to actually record the actions to take. It's really easy when you're in marketing and you should shift spend from this cohort to that one, or from this platform to that one. You can just add a button there, and they can say yes, and that can be recorded. That's kind of what we did at Uber, and then it's super easy to track the ROI. But there are so many things, like an account manager. You go into a meeting with a customer, and Orion points out that there's upsell potential, and if you mention X, Y, and Z, and it gets a deck ready for you, does the account manager actually bring that up in the meeting? Do they actually send out the order form for the upsell? Those are all things that are hard to track. What we do see in the data is, 30 days later, did the account value go up? Are there proxies for "okay, a decision was made and an action was taken"? It is approximate. In inventory management and supply chain, it's a bit easier, because it usually happens pretty quickly. But we have to use proxies, because it's not necessarily an action you're taking in Orion with a button you're pushing, although you could take it that far.

I haven't had that much pushback from people being worried about their actions being recorded. At least now you have a trail of "here's what this incredibly intelligent machine recommended, and here are the reasons why it recommended this, and that's why I made this decision." There's a clear trail to that, versus "I just showed up on Tuesday and I did what I wanted to do," where there's nothing backing you up. Now you have something that's actually backing you up. There's some data.

Keep reading

View all