Why Consumption Pricing Makes Forecasting Harder with Devavrat Shah

Summary

Revenue Builders Podcast: Why Consumption Pricing Makes Forecasting Harder with Devavrat Shah

Air Date: May 17, 2026
Hosts: John McMahon, John Kaplan
Guest: Devavrat Shah, MIT Professor & CEO of Ikigai Labs

---

Episode Overview

This episode dives into the unique challenges of consumption-based pricing models in enterprise sales, with a particular focus on how AI can help address forecasting difficulties. The hosts are joined by Devavrat Shah, an MIT professor and CEO of Ikigai Labs, who brings a data-driven perspective to the discussion. Together, they explore why forecasting is especially tricky in consumption models and how AI-powered insights can foster trust and accuracy throughout revenue organizations.

---

Key Discussion Points & Insights

1. Understanding Consumption Pricing (00:53 – 03:50)

• Consumption Pricing Overview: Devavrat Shah starts by outlining the shift from traditional SaaS licensing models (seat-based, perpetual licenses) to modern consumption-based pricing, especially as it relates to AI and token-based API calls.

  • Quote (Devavrat Shah, 01:01):
    “...lots of modern AI pricing is around this word called token-based API calls, very much along the lines of consumption. And that's where one is trying to think through, how does one think about AI as a consumption-based product rather than AI as a traditional value-based product or traditional SaaS pricing...?”

• Data as a 'Massive Cube':

  • Dev discusses conceptualizing all data as a vast cube, with different axes and time series, reflecting how data warehouses treat information. Volume of work and compute can then be directly tracked to usage—underpinning the honesty of the consumption model.

  • Quote (Devavrat Shah, 01:47):
    “We think of data as a massive data cube… For us, it's like all time series, different types of data, how much time in history, what is the future forecast? And each of the forecasted data points is like prediction, query and all. And that provides a volume of work you are doing, volume of compute you are doing.”

2. Forecasting Challenges with Consumption Pricing (02:40 – 04:35)

• The CFO vs. Sales Rep Perspective:

  • While consumption models allow CFOs to estimate total volume and revenue at a high level, it becomes far more complex for individual sales reps and managers to forecast specifics for their territories or accounts.

  • Memorable Analogy (Devavrat Shah, 03:03):
    “...you're forecasting on one hand, take a smooth water. On the other hand, you're forecasting when umbrellas are purchased. It's like two very different types of forecasting problems.”

• Manufacturing Analogy:

  • Forecasting total production volume is easier, but predicting individual customer behavior and timing is difficult—mirroring issues in sales forecasting under consumption.

  • Quote (Devavrat Shah, 03:26):
    “Easy to forecast overall volume, very, very difficult to forecast where and how. But they are all related. And this is where AI can really help.”

3. AI’s Role in Improving Forecast Accuracy (03:45 – 05:00)

• Cohort Analysis & Learning Across Reps:

  • Dev highlights the importance of not viewing sales reps or channels in isolation. AI can identify cohorts and patterns across reps, extracting learnings and surfacing collective insights to improve granular forecasting.

  • Quote (Devavrat Shah, 04:07):
    “But if we start looking at, let's say, sales reps as cohorts now, we can learn from them, each other. Here is one type of sales rep, here is another type of sales rep, here is one type of channel, is another type of channel. And now we can actually start correcting in addition to data that you have in MEDDIC or whatnot.”

• Fostering Trust with Data-driven Forecasts:

  • Rather than promoting blame or finger-pointing when forecasts are missed, AI-driven forecasting can build trust and objectivity in the process.

  • Quote (Devavrat Shah, 04:33):
    “I think there is a huge opportunity to actually empower organizations to work… with more trust rather than more finger pointing... with the help of AI. And I think that would be a good thing for everyone involved.”

4. Hosts’ Takeaways and Open Questions (04:54 – 05:52)

• Complexity of Consumption Forecasting:

  • Kaplan enumerates the factors making forecasting consumption difficult—seasonality by customer, rep forecasts, timing of new bookings—all of which AI is best-suited to aggregate and analyze for accurate predictions.

  • Quote (John Kaplan, 04:58):
    “There's seasonality overall, there's seasonality of each individual customer and their business... there's the rep forecast, there's new bookings that are occurring, there's the timing of the new bookings. So there's so many different factors that go in. And to Dev's point, you know, AI could help by the aggregation of all of that data... predicting what the answer is going to be, whereas right now that's a pretty hard task to do.”

• Indispensability of AI for Consumption Forecasts:

  • McMahon concludes that forecasting a consumption business is nearly impossible without AI.

  • Quote (John McMahon, 05:41):
    “I almost feel like it's not possible without AI to effectively forecast a consumption business. I don't know how you would do that.”

  • Devavrat Shah agrees (05:52):
    “I would absolutely agree with this.”

---

Notable Quotes & Memorable Moments

• Devavrat Shah, on the difference between aggregate and granular forecasting:
  “Easy to forecast overall volume, very, very difficult to forecast where and how.” (03:26)

• John McMahon, on the indispensable role of AI:
  “I almost feel like it's not possible without AI to effectively forecast a consumption business.” (05:41)

---

Important Timestamps

• 00:53 — Devavrat Shah introduces consumption-based pricing concepts in AI and its impact.
• 02:40 — Exploring the forecasting dichotomy: total volume vs. specific account behavior.
• 03:45 — How AI can help through cohort analysis and improving trust.
• 04:58 — John Kaplan details complex variables in consumption forecasting.
• 05:41 — John McMahon and Devavrat Shah summarize the necessity of AI in this domain.

---

Episode Takeaways

• Forecasting in consumption pricing is fundamentally more complex than traditional models—especially at the sales rep or regional level.
• AI holds promise for aggregating, learning, and predicting in ways that human analysts cannot replicate at scale, fostering organizational trust and operational accuracy.
• Without advanced analytics and AI, organizations are likely to struggle in delivering precise forecasts for consumption-based businesses.

---

Listen to the full episode for deeper insights into consumption pricing models, real-world analogies, and practical AI applications in forecasting.

Transcript

[John Kaplan]

Foreign.

[Podcast Host]

Welcome to the Revenue Builders podcast with John McMahon and John Kaplan. This podcast is brought to you by Force Management. Today let's revisit a segment from our episode with Dev Shop. Dev is an MIT professor, the director of their statistics and Data center and the co founder and CEO of Ikigai Labs. He knows what he's talking about. In this segment we talk about using AI to help with consumption pricing models and forecasting.

[John McMahon]

What are some of the ideas, understandings, challenges you think are presented right now with consumption pricing and what, what could AI help with in the future on that?

[Dev]

Fantastic. So I think there's two parts. One is about consumption pricing as a concept. And again I must sort of confess here is that my expertise there are limited. But let me sort of contextualize in the context of AI itself because lots of modern AI pricing is around this word called token based API calls very much along the lines of consumption. And that's where one is trying to think through how does one think about AI as a consumption based products rather than AI as a traditional value based product or traditional SaaS pricing which is licensing per seat, et cetera, et cetera. Right now I think it's a great framework because one way I like to think about, we like to think about in my role as, as. As a leader of company C Ikkai, right, Is that we think of data as a massive data cube. That's how world of data warehouses, world of databases have been thinking about it, right? Different axes, you have tables and all of that. Take a massive cube. For us, it's like all time series, different types of data, how much time in history, what is the future forecast? And each of the forecasted data point is like prediction, query and all. And that provides a volume of work you are doing, volume of compute you are doing. That's how sort of, let's say entire a cloud as a compute has been sort of structured. And what that has done well in my mind successfully is allowed people to understand what is it that they're getting into. They can actually just compute. There is a very honest exchange that's there. Now as that honest exchange happens in terms of overall consumption, it's relatively easier for people to forecast what their future revenues would look like in that consumption model for sitting from a CFO's office. However, if you are selling those deals as an enterprise, sales as a sales rep in a small region, in a small location or small part, small channel, then forecasting those things becomes hard because here is a, you know, you're forecasting on one Hand, take a smooth water. On the other hand, you're forecasting when umbrellas are purchased. It's like a two very different type of forecasting problems. And this shows up in many, many different places. Think of manufacturing. I'm manufacturing, let's say a material, okay, that material I need to manufacture because when my customers come, every customer comes and places one ton of order once every quarter or so. There are a few customers, choppy orders, but that collectively forms into 27 tons of production this quarter and 28 tons of order next quarter. But who? I don't know. The mechanics are exactly like that. Easy to forecast overall volume, very, very difficult to forecast where and how. But they are all related. And this is where AI can really help. Because if we look at these things in isolation, it will be very hard. But if we start looking at, let's say sales reps as a cohorts now, we can learn from them each other. Here are the one type of sales rep. Here is another type of sales rep. Here is the one type of channel, is another type of channels. And now we can actually start correcting in addition to data that you have in medpic or whatnot. So I think there is a huge opportunity to actually empower organizations to work, how to say this with more trust rather than more finger pointing, if I may call it with the help of AI. And I think that would be a good thing for everyone involved.

[John McMahon]

Johnny, what are you, what are your thoughts on this? I know a lot of questions.

[John Kaplan]

Well, I mean there's a lot. Yeah, AI could definitely help because there's a lot of parameters that go involved. There's a historical analysis of consumption by per customer. You know, there's seasonality overall, there's seasonality of each individual customer and their business. You know, there's the rep forecast, there's new bookings that are occurring, there's the timing of the new bookings. So there's so many different factors that go in. And to Dev's point, you know, AI could help by the aggregation of all of that data in order to give you, as you said before, like almost the answer, predicting what the answer is going to be, which. Whereas right now that's pretty hard task to do.

[John McMahon]

I was going to say, I almost feel like it's not possible without AI to effectively forecast a consumption business. I don't know how you would do that.

[Dev]

I would absolutely agree with this.

[Podcast Outro]

Thanks for listening to today's episode. If you enjoy the content, please subscribe, rate and review the show to help us reach more people. This show is brought to you by Force Management, where we help companies improve sales performance and executing the growth strategy at the point of sale. Check out forestmanagement.com for more information.