A B2B sales forecast built on your own conversion history
Five members of the B2B Sales work group on the methods, the data and the commit rules behind a forecast.
Updated
A forecast earns trust when every number traces back to conversion rates in your own history, and a deal reaches commit only on evidence from the buyer.
- Build it twice. Kyle Poyar builds a top-down version from the board's number and a bottom-up one from ICP accounts, rep capacity and real conversion rates. The gap between them shows which assumption has to change.
- Project from pipeline cohorts. Cassie Young groups pipeline by the quarter it was created and applies each cohort's close rate. In her worked example, a $2M plan needed about $16.5M of new pipeline from a team that generated $6M of pipeline last quarter, which showed the miss before the quarter began.
- Let buyer evidence decide commit. Peter Kazanjy ties every stage to a buyer action you can verify, and keeps a deal in commit only with a mutual close plan and 1:1 contact with the economic buyer and one other stakeholder. Kris Rudeegraap drops deals well past the median cycle out of commit.
- Keep the history and score the call. Andy Mowat snapshots opportunity history into a data warehouse, then compares the leaders' week-2, week-6 and week-10 calls and the RevOps model with final bookings. He sets the accuracy band from the team's own past misses.
Group chat with Kyle, Peter, Andy, Kris, Cassie — the B2B Sales work group
Question:
How do you build a B2B sales forecast you can trust?
Kyle PoyarFounder of Growth Unhinged. Former Operating Partner at OpenView
Build the forecast twice and make the two versions argue with each other.
The top-down version starts from the number the board wants and works backward. The bottom-up version starts from what's physically possible: how many ICP accounts you have, how many at-bats each rep gets, real conversion rates. When I talked with Emily Kramer about this, her point was that most finance and marketing teams only build the top-down one. Then you look at the ICP list and realize you can't send a million emails when you only have 5,000 target accounts.
YouTube
The gap between the two numbers is the useful output. It tells you which assumption has to change, whether that's headcount, ACV, or win rate.
Here's how I'd get the inputs right:
1️⃣ Use your own won/lost history. Pull the last 18 months of deals and calculate expected ARR per 100 opps (win rate x average ACV x NRR x 100), cut by segment. Your fintech mid-market deals and your SMB deals shouldn't share one blended win rate.
2️⃣ Run a physics check on every rep. On the podcast we called this "cowboy forecasting": price times the number of deals a rep can realistically close in the days they actually have. Plans that need 6x coverage when the team lives at 4x are fiction.YouTube
3️⃣ Forecast from accounts by stage, with opportunities as one input. Counting target accounts at each stage (aware, engaged, evaluating) and the conversion rate between stages shows you next quarter's pipeline before reps create the opps.
TL;DR: you can trust a forecast when every number traces back to a conversion rate you've actually seen in your own data.
Peter KazanjyFounder of Founder-Led Sales Factory. Co-founder of Atrium and author of Founding Sales
Founders: the forecast is only as trustworthy as your stage definitions. If a stage moves because a rep feels good, the number is fiction. Tie every stage to something the buyer did that you can verify.
At Atrium our stages were discovery, then "data lit" (the prospect actually OAuthed Salesforce and their data lit up), then preview, then commercial discussion. "Data lit" is binary. Either the account is connected or it isn't, so nobody can argue a deal into that stage.YouTube
From there, the method:
- Measure stage conversion by cohort. Of 10 first meetings, how many get a second meeting, how many a third, how many close? My target shape: roughly 7 get a second meeting, about 4.5 get a third, 2 to 3 win. Those are your leading indicators.
- Get enough at-bats before you believe the math. Two wins out of 10 is a good start. Run 50 to 100 first meetings, and if you reliably convert 15 to 25% of first meetings into customers, your conversion rates are real enough to forecast against.
- Pull out the deals that never had a moment of truth. If a prospect hasn't had a real chance to say no, their yes doesn't count yet.
LinkedIn
- Scrub against red flags: no date-driven milestones, a vague "collaborative" decision process, no current spend in your category. Two or more flags means the deal comes out of the forecast.
X
- Make commit mean there's a mutual close plan with the buyer: milestones, an owner for each step, target dates, known blockers.
Then grade the pipeline weekly on deal size × close probability, and inspect the big, likely deals hardest, because a miss there is what breaks the number.LinkedIn
Cassie YoungPartner at Primary Venture Partners. Former Chief Commercial Officer at Sailthru
Kyle's conversion rates get much more trustworthy once you add timing, so I build the forecast from pipeline cohorts. Group pipeline by the quarter it was created, then measure how much of each cohort closes in its first quarter, its second, its third. That gives you real conversion curves you can project forward.
Here's the worked example I teach. Take the open pipeline created one, two and three quarters back and apply each cohort's historical close rate. In my example that adds up to $979K in expected bookings against a $2M Q1 plan. The remaining ~$1M has to come from pipeline created and closed inside Q1, and that same-quarter close rate is only 6.19%. So you need about $16.5M of new pipeline. The team generated $6M last quarter, so the plan doesn't work, and you know that before the quarter starts.YouTube
Kyle's coverage physics check also needs a cycle-length lens. Required coverage is 1 divided by win rate, so a 25% win rate needs 4x. A company at 1.6x for next quarter on a six-month cycle won't hit Q1, and spiffing reps in January won't fix the coverage problem until July. The honest moves are to lean on expansion in the installed base now and to fund pipeline creation today for the quarter after.YouTube
Two more rules:
- Keep the forecast separate from the plan. Forecasts can change, and plans never do. Kyle's two versions shouldn't collapse into one number, so layer the targets: the sum of bottoms-up seller capacity is higher than the leader target, the leader target is higher than the company target, and the company target is higher than the board target. That buffer is what lets you hit the board number when deals slip.
- Be intellectually honest with your board. The worst thing an exec can do is tell board members the number they want to hear. It never ends well, and a forecast is only trustworthy if the board believes you'll flag the miss early.
Andy MowatCMO at Inflection.io. Former VP of Revenue Operations at Carta
I trust a forecast when it's built on history: how pipeline has actually converted in past quarters. Rep gut calls come second. So my method starts with pipeline coverage, and I've used it as the rallying metric at every unicorn I've run RevOps for.
- Get marketing, sales + finance to agree that pipeline is the number everyone is measured on.
- For each past quarter, measure how much pipeline you brought into the quarter, how much of it closed and over what time period, and how much got created and closed inside the quarter.
- Use those ratios to set the coverage you need going forward, then compare the reps' commit call against them. When the two disagree, that's the deal review conversation.
Apple Podcasts
Where most teams break is the data. CRM opportunities show you the current state, so you lose the history of how deals moved. When we launched our second product, we couldn't answer basic questions like sales cycle length or win rate. Reps would add the SKU to a renewal opp, have a discussion, then remove it, and Salesforce kept no record of that. We fixed it by snapshotting opportunity history into a data warehouse. Once you can map stage movements over time, your stage conversion rates become real numbers, and your conversations with reps get a lot sharper.YouTube
If you can't show what your pipeline looked like 60 days ago, you have nothing to check this quarter's forecast against.
Kris RudeegraapCo-CEO and co-founder of Sendoso. Former Senior Account Executive at Talkdesk
Pete's commit rule, a mutual close plan behind every deal, is the one I'd fight hardest for. When I went on Sales Talk for CEOs, my point was that assumptions are what hurt forecasting. They're how you get surprises close to the finish line, and nobody in sales wants a surprise in week 12.YouTube The host said he'd been running these plans since the 90s, before anyone called them MAPs. He told the customer exactly what to expect, when, and who had to do what, and the deals kept moving.
I'd add two checks on top of that:
- Use deal age as a check against rep optimism. At Sendoso we've watched this across thousands of customers: every day a deal sits open, it gets less likely to close.
- Put Ops in charge of the data. I've said Ops will be the skill that matters most in sales over the next few years.
The rep's call still matters. BUT... it should be the last input, after the evidence.
Kyle PoyarFounder of Growth Unhinged. Former Operating Partner at OpenView
Pete's "data lit" stage carries over to a problem a lot of teams now have. With usage-based or credit-based pricing, closed-won and revenue are two separate events.
In a seat-based SaaS model, the booking turns into revenue. In a true usage model, usage turns into revenue, and that revenue may later turn into a bigger booking. A sales forecast that stops at the signed contract will miss in both directions: committed deals that never ramp, and accounts that blow through their commit mid-term.YouTube
So I'd add a second layer under the pipeline forecast:
1️⃣ Forecast ramp curves by cohort. Group customers by use case, industry, and size, and track how usage grows in months 1, 3, 6 and 12. A new deal gets the curve of its closest cohort, and the rep's estimate doesn't count.
2️⃣ Trust the aggregate more than the individual account. Usage across all customers tends to behave fairly predictably, even when any single customer swings around a lot. It's a process of getting less wrong over time as you watch your existing customers.YouTube
3️⃣ Forecast expansion as its own line. At New Relic, about 85% of expansion came from usage, with no upsell motion attached. If expansion is a big share of your growth, it needs its own model and shouldn't ride along as a percentage on new bookings. Cassie's advice to lean on the installed base when coverage is short only works if you can forecast that base this way.YouTube
This connects back to the NRR term in the expected-ARR-per-100-opps math I shared earlier. Two segments with the same win rate and ACV can be worth very different amounts once you see how each one ramps.
This also pays off with investors. Early-stage companies often undersell the health of their business model because they can't forecast usage well, and a credible cohort model fixes that.YouTube
Peter KazanjyFounder of Founder-Led Sales Factory. Co-founder of Atrium and author of Founding Sales
Cassie's cohort curves tell you whether there's enough pipeline in aggregate. The miss usually comes from somewhere else: a commit deal that slips because the only person you were talking to was the champion. Single-threaded deals are where forecast error hides, because the champion's optimism is the only data you have.
So at the deal level, I'd make multithreading part of the definition of commit. On an X thread I described how this worked for us. Deals don't close on the group call. They close in the gaps between calls. Once we built separate relationships with each stakeholder and pinged them individually between meetings with something like "how's the team feeling about everything?", people started telling us every objection, who was raising it, and what it would take to move forward. We heard a lot less "we'll get back to you" and a lot more "let's loop in legal to start contract review."X
A practical way to run it: right after the first call, text each stakeholder so you have a direct line that isn't inbox noise. SMS gets you much better signal in deal communications than email does.X
The forecast rule I'd use: a deal can only sit in commit if you've talked 1:1 with the economic buyer and at least one other stakeholder, and each of them gave you the same close date. If you only have the champion's word, the deal stays in best case.
Cassie YoungPartner at Primary Venture Partners. Former Chief Commercial Officer at Sailthru
Kyle's expansion model covers the upside of the installed base. The downside, churn and contraction, is where I see the most fiction in board decks. Retention is the cheapest growth you have. At 125% net retention you could never sign another new logo and still grow 25% a year, so a bad renewal forecast can break the number even when sales hits its target.YouTube
Pete's rule about stages applies to renewals too. Customer health has to be defined by things you can verify. I have an allergy to the red/yellow/green board slide that's really just how the CSM feels about the account. If a customer went through an outage or shows other known risk signals, they can't be green, no matter how friendly the last call was. Work with finance to figure out which behaviors actually come before churn, then forecast renewals from those signals.YouTube
Forecast gross and net separately. A healthy expansion number can hide ugly gross churn, and I lived exactly that: great net retention, horrible gross churn. So I paid the account team on both, part of the plan on net and part on gross. If the comp plan only rewards absolute bookings, reps ignore churn and contraction, and the renewal forecast quietly rots.YouTube
Last, cohort the renewals the way Andy and I cohort pipeline. Kyle and I dug into this on Unchurned. Plenty of AI-native companies brag about hitting $10M in a month before any cohort has come up for renewal. Sub-$50/month AI-native products showed 23% gross revenue retention, meaning more than three quarters of the revenue left within a year. Until a cohort has renewed, its expansion belongs in the upside case. Keep it out of commit.Apple Podcasts
Andy MowatCMO at Inflection.io. Former VP of Revenue Operations at Carta
Kris is right that the inputs have to come from buyer activity, and I'd go one step further: the forecast is only as trustworthy as the architecture decisions you made two years earlier. At Culture Amp I joined at $5M as the only RevOps person and built the team to 150. We put attribution and CPQ in early, which meant every opp carried real product + pricing data from day one. That's why the second-product mess I described was fixable at all: the line items existed, we just needed history on top of them.
dealhub
So my practical rule for founders: make one team own the forecast data end to end, and that team is RevOps. I get really frustrated when companies set up GTM engineering as a separate department next to RevOps. You end up with two groups patching the same pipeline data, and nobody can tell you which number is right.dealhub
On automatic capture, it's finally real. Tools like Momentum now record every customer conversation into the CRM, and an agent can scan a rep's whole pipeline and push deal insights into Slack. That kills the Thursday-night field update Kris described. But be picky... when a vendor tells me they've got agents for me, I'd say 75 to 90% of the time it's an if-statement with a new label.YouTube Before you let any AI tool feed your commit, backtest it: run it against last quarter's snapshots and see if it would have called the slipped deals earlier than your reps did.
Kris RudeegraapCo-CEO and co-founder of Sendoso. Former Senior Account Executive at Talkdesk
Kyle's separate expansion line needs a single owner, or two teams will forecast the same dollars two different ways.
At Sendoso, customer success handles onboarding and adoption, and account management owns the commercial side. On Lea Tharin's podcast I described the friction that split creates: the CSM is trying to add another user to drive usage, while the account manager blocks it because they want a commercial signature first.
Spotify Put that into a forecast and the CSM is projecting usage ramp while the AM is projecting an upsell booking on the same account. Sometimes the same dollars get counted twice. Other times each team assumes the other one has the number.
The fix I'd use for Kyle's usage cohorts:
- Pick one owner for each account's expansion number, usually whoever carries the commercial quota. That person also owns the contraction risk Cassie described, so gross and net come out of the same forecast.
- Make the usage curve an input to that person's forecast. Usage growth tells you when to expect the upsell. It shouldn't become a second line item.
- Set a trigger, for example an account reaching 80% of its commit, that moves the account from the CSM's adoption view into the AM's pipeline as a dated opportunity. After that it gets the same deal-age and close-plan checks as new business.
Specialization helps you scale. BUT... if the customer journey gets split across teams, the forecast gets split the same way.
Peter KazanjyFounder of Founder-Led Sales Factory. Co-founder of Atrium and author of Founding Sales
Kris's deal-age cutoff tells you which deals are stale. The next question is what to do with them, because when a stale deal stays parked in the CRM, your stage conversion rates go wrong too. My answer is to force a moment of truth on every one of them.
Run a short follow-up sequence on any deal past its segment's median cycle. Start with a friendly check-in. Then share a new case study or ROI data. Then ask directly, "still interested?" End with a breakup email.LinkedIn Each deal comes out of the sequence either re-engaged with a real next step or closed-lost. Both outcomes help the forecast. A closed-lost deal is a data point Andy's warehouse snapshots can use. A zombie deal sitting at stage 3 for 140 days makes your stage-3 conversion rate look worse than it is and makes your coverage look better than it is.
Founders should apply this most strictly, because early pipelines are small and one fake deal is a big percentage of the number. When a deal shows two or more red flags, like "no rush" timelines, "need to talk to my boss," or no current spend on similar solutions, move on to better opportunities.LinkedIn A smaller pipeline you've scrubbed is easier to forecast from, and your rep's call becomes the last input, like Kris said, after the buyer evidence.
Kyle PoyarFounder of Growth Unhinged. Former Operating Partner at OpenView
Everyone's rules so far (commit needs a close plan, multithreading, deal age) depend on what reps type into the CRM. Reps type whatever their comp plan rewards, so the comp plan is a forecasting input too. Cassie showed it on the renewal side with her gross/net split. The same logic applies to new business.
In our episode on sales comp, the line that stuck with me was that nobody in a company is more rational than a sales rep. Give them a goal and they'll find the way to hit it or make the most money from it. Complicated plans make that worse, because reps optimize for the confusing parts.YouTube
Two checks I'd add:
1️⃣ Check whether quota passes the physics test before you trust the bottom-up roll-up. The healthy quota-to-OTE ratio we discussed was 4-5x, with 6x+ being great and rare. If the plan needs reps at 8x, expect sandbagging early in the quarter and hope-casting late in it.YouTube
2️⃣ Forecast the middle of the pack separately. On the same episode, the guest's point was that top performers will blow out the number no matter what, and most of the upside comes from moving the middle few percentage points. I suspect most misses live in that middle tier too, so I'd apply Cassie's cohort curves by rep tier and avoid one blended team rate.YouTube
Cassie YoungPartner at Primary Venture Partners. Former Chief Commercial Officer at Sailthru
Andy's point about architecture lines up with what I'm seeing in AI-native CRMs right now, and founders should read it as a warning about timing. Across the AI-native solutions we studied, the pattern was the same: strong early signal with teams under 15 sellers, then consistent breakdown at 50+. They were missing parent-child account hierarchies and territory management, and they had data accuracy issues. Customers ended up running a parallel HubSpot or Salesforce instance to compensate. Once two systems both hold pipeline data, you don't have one forecast anymore. You have two forecasts and an argument about which one is right. So pick your system of record for the team you'll have at 50 reps, even if you're at 8 today.
When you take the forecast to the board, show rates and ratios instead of absolute numbers. "$12M of pipeline" tells a board member nothing. "2.1x coverage against a 4x requirement on a six-month cycle" tells them exactly where you stand. Same with support tickets: raw volume always goes up as you add customers, but tickets per $1M of ARR shows whether the business is actually getting healthier. Normalize every forecast metric the same way so a director can compare it quarter to quarter.YouTube
Andy MowatCMO at Inflection.io. Former VP of Revenue Operations at Carta
Kris's single-owner rule works at the account level. Someone also has to own it at the system level, so the new business, expansion and renewal forecasts all roll up from one data model. At Carta my team covered sales ops, marketing ops, CS ops, enablement + GTM analytics for over $400M ARR, and at Culture Amp it was the same setup. Having all of those under one roof meant the CSM's usage view and the AM's upsell opp lived on the same account record with the same definitions, so the double counting Kris described showed up in a reconciliation report before it showed up in a board deck.
The practical version for a growing company:
- Write one definition doc for every forecast category (new logo, expansion, renewal, contraction, churn) and every stage in each. RevOps owns the doc, and sales + CS sign off on it.
- Run one weekly forecast call where all three numbers get presented together, with Cassie's gross and net retention as separate lines.
- Have RevOps publish its own number next to the leaders' call, built from the snapshot history and cohort curves. Leaders still make the call. RevOps' job is to show the gap and ask which deals explain it.
The same logic applies to AI. I've been saying fragmented AI will silo and break your GTM even more, and RevOps is the only function that can own the orchestration layer. If sales buys one forecasting agent and CS buys a separate health-score tool, you've rebuilt the two-teams-two-numbers problem with more expensive software. It's the same failure as Cassie's parallel HubSpot instance, one layer up.
Question:
The group now has a full stack of methods: Kyle's two competing versions, your cohort curves, Andy's history snapshots, and Peter's and Kris's commit rules. None of them says how to tell whether the forecast has earned trust. Cassie, what accuracy did you hold the team to, for example the week-2 call against final bookings, and how did you score a miss with the board? Please cite public work only.
Question:
Andy, your setup has RevOps publish its own number next to the leaders' call each week. How did you score which one was right over time? What accuracy band against final bookings, and at which week of the quarter, did you treat as a forecast the board could trust? Please cite public work only.
Andy MowatCMO at Inflection.io. Former VP of Revenue Operations at Carta
I scored it as a head-to-head over time instead of anchoring on one magic accuracy band. A fixed band means very different things for a 30-day SMB cycle and a six-month enterprise cycle, so the useful question is which number, the leaders' call or the RevOps model, landed closer to final bookings quarter after quarter.
The weekly snapshots make that scoring possible. You pull each leader's week-2, week-6 and week-10 call and the model's number at the same points, then compare all of them to final bookings, by segment and by rep. In Salesforce the opp only shows what it says today, so last month's call is gone unless you stored it.YouTube
I'd score three things every quarter:
- Call accuracy: leader call versus model number versus final bookings at each checkpoint week.
- Slippage: the share of commit deals that pushed a close date or cut an amount, by rep.
- Coverage conversion: how much pipeline you started the quarter with and how much actually closed, which is the coverage math I've used as a company-wide metric.
The forecast is board-ready when the early-quarter call has stayed inside the same tolerance for two or three quarters in a row and the miss can be explained deal by deal. Set the band yourself from your own spread of historical misses and tighten it each quarter. After a few quarters you also know whose call to haircut and whether the model or the leaders deserve the weight.
Ask a follow-up
+1B2B Sales
Ask B2B Sales anything...
Frequently Asked Questions
How should a startup set up its CRM stages?
Use one deal pipeline where each stage needs a buyer action, keep booked first meetings on the lead or account, and open a deal once the buyer confirms the pain and a next step.
How do you book sales meetings at a conference?
Pick the target accounts first, start outreach about a month out with a specific time and place, use small dinners or side events as the reason to meet, and book the follow-up on site.
What funnel conversion rates should a B2B team aim for?
One working range is 40 to 60% of first conversations becoming opportunities, and 25% opportunity to win is a planning case. Set the other steps from your own cohorts, split by source.
What cadence works for founder cold outreach?
Four to six touches across email, LinkedIn and phone over 10 days to three weeks, each adding something new and ending with a direct ask. Then wait for a signal.
How many SDRs should you hire per AE?
Work it out from each AE's pipeline gap rather than a fixed ratio, measure SDRs on opportunities AEs accept, and hold off on SDRs until the sales motion is proven.
