SaaS Sales Forecasting - Christian Wattig’s Top Insights and Methods

Kirk Kappelhoff

Alright. Hello, everybody. I can see people rolling in now. Christian, Saurav, how are we doing today? Fantastic. Great to have you, Christian, once again. Saurav, how are we doing?

Saurav

All good, sir. Excited.

Kirk Kappelhoff

Good. Good. Well, thank you both for joining. Today, obviously, we are talking about a very big topic. Specifically, we will be talking about the world of SaaS forecasting. This is a big world to talk about. As you can probably imagine, the global landscape has been changing the last couple of years. On average, SaaS has a CAGR of around twelve percent every single year. Eighty percent, in recent studies, of businesses use a SaaS product. And IT departments could have up to twenty-five percent of their spend on SaaS products. So now more than ever, it's an important time to be able to forecast your SaaS. It's a growing business. And so we're really excited to have Christian and Saurav today, lead industry experts on this topic. In fact, this would be a good time for you guys to introduce yourselves. I'll start with me. Who am I? I'm Kirk, director of strategic finance here at Drivetrain. My past has been modeling specifically for IPOs. Now, obviously, with Drivetrain moved from Excel to a system world, the formulaic expressions we use for forecasting remain the same. And that's what we'll be chatting about today. Christian, I'd love to hear a little bit about your background.

Christian Wattig

What brought you here today and made you the FP&A expert? Sure, I don't know if I'm "the" FP&A expert, but my background: I started my career in FP&A at Procter and Gamble. I spent four years there, then seven years at Unilever. After my MBA, I transitioned into tech and had a chance to lead an FP&A team at Squarespace, a website builder company, when we took the company public. That was a fantastic time because we learned how to get the CFO ready to stand in front of analysts and tell the financial story of the business, and we looked at forecasting from every angle. I also worked at a smaller Series B B2B SaaS company in the FP&A tool space. When I transitioned from consumer goods to SaaS, I realized a lot of the challenges FP&A people face are actually very similar and independent of industry and company size — things like: how do I improve forecast accuracy while making the forecast more useful for decision-making? How do I build a financial model optimized for fast iteration? How do I manage a collaborative planning process? Realizing that, I also discovered I really enjoy teaching — what I loved about being a leader was helping my direct reports develop and advance. So I thought maybe there's a way to teach FP&A more broadly. A few years later, I'm now director of the Wharton FP&A certificate program, a corporate trainer, and I teach FP&A full-time.

Kirk Kappelhoff

Wow. Well, Christian, we are looking forward to learning from you today and taking that breadth of industry experience to figure out how we can best forecast the SaaS world. We also have Saurav here — a fellow Drivetrainer, head of product in RevOps. Saurav, I'd love if you could give an overview of what brought you here and what made you a RevOps expert.

Saurav

Yeah, hi folks. I'm Saurav, I lead product and RevOps at Drivetrain. I used to be a venture capitalist looking at SaaS companies in an earlier life, and for the last couple of years I've worked with a lot of our customers on the RevOps topic across the spectrum. So I have a fair bit of understanding of how this is done and have seen customers use these methods up close.

Kirk Kappelhoff

Well, thank you both for being here — we're looking forward to learning from you. Let's talk about what we'll cover today. We've done introductions; next we'll talk about the types of SaaS sales forecasting — there's obviously not a one-size-fits-all, so we'll cover factors like company size and timeline. Triangulation is something not a lot of people in forecasting have heard of — Christian will give an overview of that, and it'll help eliminate what we call forecast bias. Then we'll do a deep dive with real-world examples, so you're not just getting the theory today. We have a poll up now — feel free to answer. What you can expect from this session: first, sales forecasting techniques (the theory); second, the practical piece — once we learn the theory, how do we actually build a model with it; and third, something coming over the next couple of months — deep-dive Drivetrain masterclasses on each of these forecasting techniques. So keep an ear out for anything that specifically jumps out at you. Let's start with the sales forecasting overview. There are two high-level factors to consider: your go-to-market sales motion, and your forecast period or horizon. Sales-led growth — bifurcated across SMB, mid-market, and enterprise — is one major consideration, and that can look different depending on whether you're forecasting the current quarter versus twelve months out. Product-led growth, prosumer motion — a mix of producers and consumers — looks different still. And of course there's always a hybrid or multiple-growth-engine model blending these approaches.

Kirk Kappelhoff

Christian, any comments on what you've seen here? Are there other factors from your industry experience that might be an addendum to these two?

Christian Wattig

Yeah, the factors here are definitely key, but there are others — for example, how volatile your business environment is can also influence your choice of forecast technique. Let me share an example from my time at Squarespace. For those who don't know, Squarespace is a website builder selling to both consumers and businesses; I led the FP&A team responsible for marketing and growth. During the pandemic we saw an uptick in subscriptions, like many companies, and had to figure out how much was a one-off versus sustainable long-term. That was no easy task, because basically all our existing forecasting methods fell apart as the business environment changed so drastically. So what we ended up doing was having the company hire data scientists who developed a machine-learning-based model. After a lot of trial and error, we found that if we fed the model not just our own data but also geographic unemployment and even mortality rates, forecast accuracy came back to almost where it was before. Just another example of how important it is to think about these factors when choosing your forecast technique.

Kirk Kappelhoff

How interesting. Saurav, anything to add — any practical examples, or are these factors pretty comprehensive already?

Saurav

I think these are the major ones — Christian's point on market volatility really resonates. One of our enterprise SaaS customers, when forecasting the current quarter, follows a very bottom-up model getting inputs directly from reps, because that's more reliable for them. That's very different from another customer, more product-led-growth, working with trials, who follow a demand-gen or top-of-funnel approach instead. Very different businesses, very different approaches — that's the key factor to consider.

Kirk Kappelhoff

Definitely, many factors in motion here. Let's look at some of these approaches — I'll give a quick overview of what each means, then I'll ask Christian to comment on the first two, and Saurav on the second two. These are four main methods we typically see Drivetrain customers use to forecast. Judgment-based: a bottoms-up approach where a sales rep looks at their pipeline and estimates how much they'll close each month, quarter, or year. Probability-weighted: pipeline deals weighted more or less heavily depending on their stage — a lead generated yesterday weighted much lower than a deal closing tomorrow. Driver-based: your typical price times quantity — number of users times ACV, for example. And statistical analysis: further down the line, where you might use something like linear regression to project a trend line forward. Christian, that's the high level — I'd love your detailed take on the first two, plus any thoughts on the other two.

Christian Wattig

Sure. It's extremely important to understand the advantages and disadvantages of each method, since you only know which to choose for your situation once you understand the trade-offs. Judgment-based, also called expert-led forecasting: you ask the sales reps for their view on the likelihood of closing. The advantage is that reps factor in context, nuance, and qualitative signals other methods can't capture — an experienced rep can sense, from how a demo went, whether something is urgent for the prospect in a way historical data alone never could. It's also quick and nimble to update — if a rep realizes mid-conversation that a deal has stalled, the model can update immediately. The disadvantage: because it's expert-led, it's subject to human bias, emotional and external pressures — we'll dig into those biases later. There can be inconsistency across reps of different seniority and training levels. There's risk of groupthink — assuming everyone at a certain pipeline stage is equally likely to close, which isn't always true deal by deal. And it's typically more time-consuming, since finance often wants to talk directly with reps to test and challenge the numbers, which takes real time.

Kirk Kappelhoff

Definitely, super interesting evaluation. Saurav, anything to add on judgment-based, especially at the manager and rep level where you've had a lot of experience?

Saurav

I think this works great for companies with high ticket sizes and fewer opportunities. Christian made a great point — reps and managers have the most context, more than statistical methods can capture. But that context also becomes the downside: you're relying on their judgment, and inherent over- or under-estimation bias can skew your forecast either direction. Very common technique, but that's the trade-off.

Kirk Kappelhoff

I like that — the biggest advantage of a method can also be its biggest disadvantage. Christian, back to you on probability-weighted.

Christian Wattig

Probability-weighted is more data-based — you look at historical conversion rates at each funnel stage and apply a standardized probability to every opportunity at that stage. The advantage, almost the inverse of judgment-based: you avoid human bias since it's top-down and data-driven, and it's quick to update since you don't need one-on-one rep conversations. But you lose all the nuanced information a sales manager might have that could actually improve accuracy — and that's often underappreciated. If your sales process changes, or you hire more experienced managers, you need to revisit those probabilities, and companies are often slow to do that, leading to over- or under-forecasting because the rates didn't dynamically reflect the changes. That issue doesn't happen with expert-led forecasting, since everything's bottom-up there. In my experience there's a thin line between making your forecast more accurate and spending too much time doing it without meaningfully improving accuracy.

Kirk Kappelhoff

Saurav, on driver-based — any detail or nuance people might not be aware of?

Saurav

One of the most interesting nuances: technically, any driver can be used to forecast — the real question is which driver actually represents the forecast you're trying to capture. Christian and I were talking recently about a holistic annual model where the drivers were specific initiatives meant to move subscriptions a certain way. Another approach is a sales or quota-capacity forecast, where the driver might be the number of deals reps are expected to close, times ACV, times percentage quota attainment. The interesting part of this method is picking the driver that genuinely captures what you're trying to forecast.

Kirk Kappelhoff

Very interesting, Christian — anything to add on driver-based?

Christian Wattig

I'm a big fan of driver-based forecasting because it forces you to really understand your business — you can't have fifty drivers, you have to narrow to your five-to-twelve highest-value ones, which means understanding the relationship between each driver and sales. It also enables scenario planning: say one driver is a new product feature expected to lift sales — you can build conservative, base, and aggressive scenarios for that driver alone, then do the same for another driver like couponing. That makes it easy to tell a clear risk story to your board: bullish on the new feature, more conservative on couponing since you've overused it before, for example. One of my favorite things with driver-based forecasting is sensitivity analysis — flexing each driver up or down by, say, ten percent, and seeing the incremental impact on revenue, net income, or cash flow. That isolates which driver actually moves the needle most. The danger, though: risk of error stacking. Each driver carries its own margin of error, and stacking many together can multiply that error rather than average it out. That's where a simpler model, even a back-of-napkin approach, can get you eighty percent of the way there — sometimes the extra effort to gain one more percent of accuracy actually makes things less accurate.

Kirk Kappelhoff

Saurav, take us through the last one — statistical. I mentioned linear regression, but I know there are many others.

Saurav

This one's interesting — people from more quantitative forecasting backgrounds often prefer it for its perceived rigor. Two challenges make it tricky to implement: first, data availability — companies with slower sales cycles or generally less data struggle here. Second, data quality — you might have the data, but if quality is poor, you get a garbage-in-garbage-out problem. That said, I've seen these models work beautifully for usage-based revenue forecasting, which is otherwise genuinely hard to forecast — several of our customers use statistical methods specifically for consumption-based revenue.

Kirk Kappelhoff

Super thorough evaluation. These obviously aren't the only four methods out there, just some of the most common we see. What I'd love to do next is talk about triangulation. Christian, I know this is something you've spent a lot of time on and have strong opinions about — could you give an overview, and maybe some practical examples of how you've implemented it?

Christian Wattig

Triangulation is a way to stress-test your forecast and improve accuracy by using multiple forecasting techniques in parallel. It matters which methods you combine. First recommendation: pair a judgmental method with a statistical one, since they have opposite strengths and weaknesses — judgmental methods carry human bias but capture nuance; statistical methods have little human bias but miss that nuance. Combining them finds a better middle ground. Second: combine top-down and bottom-up. Bottom-up is built by business teams — marketing, product — layering up from individual opportunities; top-down looks at the business holistically from finance. Across ten-plus budgeting cycles, they've never matched for me — top-down tends to be more aggressive, bottom-up more conservative, partly for psychological reasons: the people delivering the bottom-up want room to over-deliver, while leadership, more removed from day-to-day execution, tends to be more ambitious. That mismatch isn't bad — it's actually a great conversation starter. You can walk your business partner through your own calculation and find out where you disagree or where you might be underestimating a risk, and that conversation often produces a forecast more accurate than either original plan.

Kirk Kappelhoff

Great overview. I like how judgmental and statistical methods have opposing strengths and weaknesses, so combining them cancels out both sets of disadvantages. Is there a reasonable threshold for how far apart top-down and bottom-up should be before you'd want to revisit one or both?

Christian Wattig

You want to avoid a very large gap, since that just leads to unproductive discussions. What I recommend — and what many companies do — is not sharing your full top-down before the bottom-up is done, since that just influences people to match your number. Instead, share guardrails: roughly how much you're planning to invest in marketing, roughly your headcount plan, roughly what investors expect in growth as a range. That gives people direction without collapsing the two plans into an artificial match.

Kirk Kappelhoff

Super helpful. Quick recap before we move on: your time horizon matters — near-term forecasts lean more bottom-up, longer-term ones need more breathing room. Select the approach that fits your business, considering time horizon among other factors. Once you have an initial forecast, triangulate it rather than relying on one method alone. And compare your triangulated numbers — a big gap tells you something worth revisiting. There need to be guardrails, like Christian mentioned, so top-down and bottom-up can converge on something everyone understands. Let's look at how these methods map to different business types — Christian, then Saurav, your input on each as we go.

Christian Wattig

For a sales-led enterprise motion with few, high-ACV opportunities, an expert-led rep-level forecast makes the most sense — with big deals, you want to spend real time on each one, interviewing reps and understanding probabilities individually. The caveat: this method is most accurate for the near term, since reps simply don't have visibility ten months out. For mid-market and SMB, where there are many more opportunities, the expert-led method becomes harder to manage — and there may not even be a dedicated sales lead yet. There, a top-down, pipeline-stage-probability approach makes more sense: standardized, data-based, good for an earlier-stage business, evolving toward more rep-level input as you move upmarket.

Kirk Kappelhoff

I like that we've distinguished existing pipeline — where I at least know the prospect's name and can gauge probability from the conversation — from new pipeline, where I don't yet know who those future customers even are, which is exactly why statistical methods come into play there. And when you have very reliable CRM data, a sales quota or capacity model — really just a variant of driver-based forecasting with more data behind it — makes sense too.

Kirk Kappelhoff

Saurav, any comments on these so far, especially the second and third ones where you've got direct experience?

Saurav

Great insights from Christian. I'd just add: for a more steady-state, stable business, methods two and three tend to be more relevant, since many of your drivers can be trusted to play out with reasonable historical accuracy.

Kirk Kappelhoff

Good stuff. Next: prosumer companies with a PLG motion — Christian, Saurav, over to you.

Christian Wattig

With a prosumer company, you have a large volume of leads, so it makes sense to look at your demand-gen funnel — website visits, free-trial conversion, paid conversion. That said, this method alone can be a bit shallow, so I'd recommend combining it with a driver-based method that captures other growth factors — like product changes — in a broader triangulation.

Kirk Kappelhoff

Saurav, anything else on this one?

Saurav

Maybe a note on triangulation here specifically: these companies typically have high data volume — large customer counts, smaller ticket sizes, high funnel throughput — which makes statistical techniques genuinely useful for triangulating against a top-down demand-gen number. We've seen several customers do exactly that.

Kirk Kappelhoff

Triangulation really is the big theme of today. Last one: hybrid or multiple-growth-engine companies — is this triangulation in full practice, Saurav?

Saurav

With a hybrid, multiple-growth-engine model, that's exactly where you want to combine different forecast techniques — more complexity means more iteration and experimentation with methods.

Kirk Kappelhoff

Christian, by "hybrid" are we talking, say, true SaaS — accounts times a dollar figure — plus a usage-based component or a services/implementation fee layered on top?

Christian Wattig

It could be that, or, for example, Squarespace sells to consumers but also has a B2B arm with a dedicated sales team reaching out to companies wanting to buy, say, a hundred websites. Either way, you end up needing multiple methods working together.

Kirk Kappelhoff

Makes perfect sense. Let's talk about eliminating forecast bias — Christian, this is your area of expertise; anything else to call out beyond what you've already touched on today?

Christian Wattig

A few common sources of bias worth naming quickly: political or external pressure; status-quo bias — assuming what worked before will keep working; confirmation bias — overweighting information that confirms what you already believe and ignoring the rest; narrative fallacy — seeing a story or pattern in data points where none really exists; Dunning-Kruger — experts tend to underestimate their own ability while newcomers overestimate theirs; and ignoring unknown unknowns — especially in sales, where something can derail a deal in a way you simply couldn't have foreseen, so you build in an adjustment even when you feel confident. Once you identify these biases, first measure them — track your average variance between actuals and forecast over time. Then it comes down to collaborating: going back to whoever gave you the input, or interrogating the data if it's a statistical method, and asking how much rigor actually went into the estimate and what influenced it. I always ask people to walk me through, step by step, how they arrived at a number — that's usually where biases and gaps in thinking surface.

Kirk Kappelhoff

Super valuable — so step one is really just recognizing these biases exist and knowing how to categorize them?

Christian Wattig

It covers the whole process: recognize it, measure it, then address it by walking through the reasoning with people, and by interrogating a statistical model for hidden data biases you might be missing.

Kirk Kappelhoff

Incredibly valuable — we'll include checklist items on everything we've covered today. Let's move into the demo Saurav's going to walk us through.

Saurav

Thanks, Kirk. The example we'll use is an enterprise SaaS company forecasting new bookings for the current quarter, using a bottom-up rep/manager-level forecast, then triangulating it against stage-weighted and forecast-category-weighted models. Let me share my screen. So, at a high level: this quarter's forecast lands around $43.9 million, derived bottom-up from what sales reps have entered. There's also a manager forecast at $40.7 million, used to triangulate against that $43.9 million, plus stage-weighted and forecast-category-weighted numbers in the 48 million range — all in a similar ballpark to the rep-level number. You can filter by rep to see their individual forecast, and start threaded conversations directly on the numbers — exactly the kind of rep-level questioning Christian mentioned earlier. You can tag key deals and reps contributing to a number, and everything shown here is fully customizable to your own business structure. On the triangulation side: alongside the rep number, we've got stage-weighted and forecast-category-weighted figures, broken down by how much of the pipeline sits in each stage or category — in this case, a lot sits mid-funnel, which, if I were VP of RevOps here, would be a bit concerning. The underlying probabilities driving these triangulation models are also visible and fully adjustable. This is the rep-level forecast tracker — what reps themselves see, with role-based access control so each rep only sees their own relevant deals, never a peer's numbers. A rep can review their committed and pipeline deals and adjust their forecast call directly — edit values, mark confidence on specific deals, move something from best-case to committed, and the aggregate number updates automatically and flows through the entire model, not just the reporting layer. Managers get an equivalent dedicated view, seeing all their reps' deals and submitting their own roll-up — which is where that $40.7 million manager number comes from, generally more conservative than the rep aggregate, since managers tend to counterbalance rep optimism. For stage-weighted and forecast-category-weighted: we pull pipeline data directly from Salesforce or another CRM, apply probabilities per stage or category — which can be pulled from CRM data or manually overridden by finance if you think a given stage's win rate is off — multiply the two together, add in already-closed-won, and that produces the bookings forecast.

Kirk Kappelhoff

Super valuable walkthrough, really appreciate it. Let's open it up to a few audience questions. One's for you, Christian — do you have any tips on balancing simplicity and scalability against model complexity in SaaS forecasting?

Christian Wattig

Great question. Model complexity deserves real attention — don't just chase accuracy. Too much complexity risks error-stacking: many drivers each carrying their own margin of error can compound rather than average out. It also reduces transparency — if accuracy suddenly drops, a highly complex model makes it much harder to diagnose why. My recommendation: iterate, and keep complexity only as high as it needs to be to hit the accuracy you actually require.

Kirk Kappelhoff

So the advice is really: don't over-hire into complexity, iterate toward the simplest model that gets you there?

Christian Wattig

Right — you need some complexity, but the main takeaway is to iterate and test different models to see what actually performs best for you.

Kirk Kappelhoff

One more for you, Christian: how do you account for large outlier deals — the kind that close only a few times a year but can make or break your forecast?

Christian Wattig

Good question, one I get often. By definition, an outlier has a low probability of occurring, and we don't have a crystal ball to predict exactly when. What you can do is look at the last three to five years, count how many such outliers occurred, and divide by the number of years to get an expected annual contribution you can bake into your forecast. Yes, if it happens your forecast undershoots, and if it doesn't it overshoots — but this approach gets you closer to reality in either scenario than ignoring outliers entirely, since you'll never have perfect accuracy regardless.

Kirk Kappelhoff

One more before we wrap: any recommended method for handling seasonality — say, strong months in March and July, weaker ones in September and December?

Christian Wattig

You can use any base forecasting method, then apply seasonality as a second step. Forecast your total — say, a hundred thousand new subscriptions over the next twelve months — then take last year's monthly, weekly, or even daily distribution as a baseline and allocate the forecasted total across periods according to that historical seasonal pattern.

Kirk Kappelhoff

Saurav, anything to add on seasonality?

Saurav

I've seen customers build seasonality directly into their driver-based model — adjusting expected close rates or expected rep logo counts upward in historically stronger months. So: use historical data to establish the seasonal pattern, then encode that pattern as a driver in the model itself.

Kirk Kappelhoff

Beautiful. Next: can we track week-over-week forecast movement at the individual seller level?

Saurav

Yes, absolutely possible — sorry, we're losing your connection a little, Saurav.

Kirk Kappelhoff

Christian, are you having connection trouble too? We missed part of that — Saurav, feel free to also drop it in chat and we'll pick it up there. Christian, on the same theme: whether to track weekly or monthly really comes down to business volatility?

Christian Wattig

Exactly — for some businesses, weekly or even daily tracking is essential; for others, that level of cadence would be complete overkill. It's genuinely business-specific, so I'd resist a one-size-fits-all recommendation there.

Kirk Kappelhoff

Fair enough. One more: for PLG businesses, how do you quantify the expected impact of a product improvement — say, a new feature meant to lift free-to-paid conversion — in percentage or dollar terms?

Saurav

This one depends a lot on company maturity. In larger, more established companies that have been through many product cycles, the expected percentage lift is often estimated by the product team as the feature is being built, and that becomes your initial forecast input — then, once it ships, you gather real data and check how that assumption held up. In earlier-stage companies, a lot of that information isn't known upfront, so you're more reliant on a ballpark estimate, iterating quickly and getting real usage data flowing as soon as possible to validate or correct the assumption.

Kirk Kappelhoff

Christian, anything to add?

Christian Wattig

Just that this is genuinely tricky to forecast — it's an art and a science, so don't overthink it. For a small product change, I'd often just let it get absorbed into your existing forecast method. For a major change, that's exactly where you'd apply the more deliberate driver-based approach we discussed earlier.

Kirk Kappelhoff

One quick side note — will this be recorded and shared? Yes, absolutely — recording, a cheat sheet, and masterclass registration links will all go out. If you haven't already, drop your details in one of the polls, or reach out directly if you'd like a Drivetrain demo. Last question before we close: how often should leadership, sales ops, or sales strategy meetings happen — both within a quarter and across the year?

Christian Wattig

I think it's important to run two separate cadences. One should be independent of where you are in the fiscal calendar — as a finance leader, you want a close relationship with go-to-market, ideally weekly touchpoints, even just joining an operational meeting they're already running, so you stay close to the business and spot changing trends early. The second cadence ties to quarter-end and year-end close, often driven by board meeting timing — naturally, meeting frequency increases as you approach the board meeting and leadership alignment becomes more urgent.

Kirk Kappelhoff

Saurav, would you say that holds true in your experience as well?

Saurav

Christian captured it really well — the one nuance I'd add is that sales and RevOps teams tend to meet noticeably more often as you approach quarter- or year-end, but otherwise that's exactly right.

Kirk Kappelhoff

Perfect timing — we've got about a minute left. Everybody, thank you so much for joining. Christian, Saurav, thank you both — any last words before we close? Looks like we're good — thanks so much for joining today's webinar, Christian, thank you again. Keep up with us through our website — sign up for our blog and email updates, and look out for the masterclasses coming in the next few months. Thanks so much to everyone who joined.

Christian Wattig

Have an absolutely awesome day.