Demand generation forecasting model

Most marketing forecasts get built backwards. Someone takes the revenue number the board approved, divides it by average deal size, multiplies by a win rate nobody has re-measured in a year, and calls the output a pipeline target. It holds for one quarter. Then sales misses, marketing points at lead volume, and the whole thing is rebuilt from a fresh set of invented ratios.

A forecast you can defend runs the other way. Start with spend, move through conversion rates you measured yourself, and land the pipeline in the month your sales cycle says it will land. The arithmetic is simple. Refusing to fill the model with numbers you like better than the ones you have is the hard part.

What goes into a pipeline forecast you can defend

Five inputs and one lag. Spend by channel, cost per qualified response, response to opportunity rate, win rate and average deal size produce the value; the sales cycle decides which month that value shows up in.

The physical shape of the model is one row per channel per month, with columns running left to right in that order. No stage should be skipped, because a forecast that jumps straight from spend to revenue gives you nothing to diagnose when it misses. You want to be able to say “responses were fine, the opportunity rate fell” in under ten minutes.

The most common reporting error in this whole category is comparing pipeline created in March to revenue closed in March. Those two numbers have almost nothing to do with each other in a business with a 90 day cycle. March revenue came from December pipeline.

Spend has to be fully loaded or the whole model lies to you in a direction you will enjoy. A content programme that costs $8,000 a month in freelance invoices and two salaries is not an $8,000 channel. If you have not split loaded cost by channel yet, the demand generation budget calculator does the allocation before you start modelling conversion.

The worked example: $20,000 ACV and a 90 day cycle

Here is a single month for a mid market business with a $20,000 ACV, a 90 day median cycle and $112,000 of monthly demand generation spend. Nothing in it is rounded to look neat.

Fifty nine opportunities at $20,000 is $1.18M of pipeline created in the month. Now apply win rates, which differ by source far more than most teams model:

Two things in that table deserve suspicion. Branded search at $714 per win is not a channel doing $714 of work, it is a harvesting surface collecting demand that podcasts, review sites and word of mouth created somewhere you cannot see. Treat it as a capture cost, not as proof that brand advertising is cheap. The demand creation versus capture split matters here because a forecast that grows branded search 40 percent without growing anything upstream is forecasting a number that will not arrive.

The second is $7,568 marketing cost per win. That is not CAC. Load the sales organisation in and cost per win for this profile typically lands between $14,000 and $18,000, which at 80 percent gross margin puts payback around 11 to 13 months.

Model the lag with two numbers, not one. Use the median for the base case and the 75th percentile for the downside case. If your median is 90 days and p75 is 140, roughly a quarter of each cohort lands a full quarter later than the base case shows, which is exactly how a team that hit its pipeline target still misses revenue. The pipeline forecast calculator applies both offsets so you can see the two revenue curves side by side.

Why 3x pipeline coverage is the wrong number for most teams

Three times coverage is the reciprocal of a 33 percent win rate and nothing more. It became a default because someone’s enterprise software business in 2009 won a third of its qualified deals, and the number outlived the business.

The formula is coverage equals one divided by your win rate on open pipeline, divided again by the share of that pipeline that closes in the quarter it was assigned to.

Run the coverage number per segment or it tells you nothing. A business with half its target in enterprise and half in self serve has a blended requirement around 5.5x, and hitting exactly 5.5x while the enterprise segment sits at 3x means the quarter is already lost even though the dashboard is green.

Coverage is also a symptom, not a virtue. Nine times coverage at a 24 percent win rate means either your opportunity definition has drifted and you are counting conversations as deals, or your AEs are keeping dead opportunities open to protect their pipeline number. Both are fixable and neither is fixed by generating more leads. Snapshot coverage on day one of the quarter against that quarter’s new business target, and never mid quarter, because mid quarter measurement quietly includes deals that were created and closed inside the period.

Where your conversion rates come from, and why not from a benchmark deck

From your own CRM, over the trailing four quarters, with the immature cohorts removed. Borrowed benchmarks are for sanity checking a number you already produced, never for populating a cell you have not measured.

The assumption set is the part teams skip and the part that makes the forecast defensible. When a CFO asks why you modelled a 24 percent win rate, “measured on 61 closed mid market opportunities from Q2 2025 to Q1 2026” ends the conversation. “That felt about right” does not.

Use published demand generation benchmarks the way a doctor uses a reference range: if your number sits far outside it, go and check your instrumentation before you go and check your strategy. A 62 percent response to opportunity rate usually means somebody is counting sales accepted leads as opportunities.

Scenario planning for hiring ramp, capacity and seasonality

Three scenarios, and the one you publish is the middle one with the low case attached. The high case exists so you know what to do if it happens, not so you can present it.

Sales capacity caps everything. An AE working mid market deals can hold somewhere between 12 and 20 active opportunities before the win rate starts falling, so 59 new opportunities a month with a 90 day cycle implies roughly 150 to 180 open deals at steady state and somewhere near ten to twelve carrying reps. Forecast 90 opportunities a month into an eight rep team and you will not get more revenue, you will get a lower win rate and a longer cycle. Cap the model at capacity and say so out loud.

Ramp is the second constraint. A mid market AE hired in month one is typically productive from month four to six, enterprise from month six to nine. That means a Q1 hiring plan changes Q3 bookings, so a forecast built on the headcount plan rather than the ramped headcount plan overstates the second half of the year every single time.

Seasonality is smaller than people think and real enough to model. Late December compresses into a few days for anything needing procurement, August drags in European markets, and the first two weeks of January produce demo requests from people building next year’s plan who will not buy for five months. Build the seasonality index from your own three year monthly history rather than a generic curve, and if you do not have three years of history, use a flat index and say it is flat.

For the low case, do not invent pessimism. Take the 25th percentile of each conversion rate from your trailing four quarters and run the same model. That gives you a downside that is grounded in things that actually happened to your business rather than a number somebody halved for comfort.

The monthly variance review that makes the model better

Thirty minutes, one meeting, one question: which input broke? A forecast review that concludes “we missed by 18 percent” has produced no information. A review that concludes “responses were at 104 percent of plan and the opportunity rate fell from 19 to 14 percent” has produced a decision.

That fourth row is the one that has changed most since 2024. Buying groups of six to ten people, which is what Gartner’s research on B2B software buying has reported for years, now routinely add a security questionnaire and a procurement gate to deals that used to close on an AE’s word. Your cycle length input drifts upward quietly, the model keeps landing revenue a month early, and nobody notices for two quarters.

Feed the variance findings into the plan rather than into a slide. If the opportunity rate on LinkedIn has fallen for three consecutive months, the demand generation plan template is where the reallocation gets recorded, not the forecast spreadsheet.

Forecast accuracy belongs on the marketing scorecard

Measure it, publish it, and be judged on it. Absolute percentage error on quarterly pipeline created, reported at the channel group level so that two offsetting errors cannot cancel into a flattering total.

Twenty to twenty five percent error in the first two quarters is normal while the instrumentation settles. By the fourth quarter you should be inside 10 percent. A team that is reliably right within 10 percent gets to ask for budget and gets it. A team that misses by 40 percent in both directions gets audited, and deserves to be, even if the average across the year looks fine.

There is a cultural payoff that is easy to miss. Once accuracy is the metric, the incentive to inflate the forecast disappears. Nobody games a number they will be measured against twelve weeks later.

Where this model will let you down

It cannot forecast demand it cannot see, which is most of the demand that matters. Podcast listeners, Slack community lurkers, people who read a comparison thread on Reddit and typed your name into Google eight weeks later: all of that arrives as branded search or direct, and the model credits the last surface rather than the cause. Run a self reported attribution question on the demo form and compare it to platform data, because dark social routinely accounts for a third or more of what the CRM labels direct.

The second failure mode is structural change. A competitor cutting price by 30 percent, a category shifting because of a platform launch, or your own packaging change will break every conversion rate in the model at once, and no amount of trailing four quarter data sees it coming. When something structural happens, throw the rates out and rebuild after two clean quarters rather than pretending the historical average still applies.

Third, the model rewards what is measurable. Channels with clean click paths look efficient and channels that create demand look expensive, so a forecast used as a budget allocator will slowly starve the things that feed it. The countermeasure is a fixed floor, typically 25 to 40 percent of budget, protected for creation activity and reviewed on branded search growth rather than on cost per opportunity. The wider demand generation strategy view is where that floor gets set, not the forecast.

What to do next

Rebuild the model once, properly, and then leave the structure alone for a year. The value comes from the same shape being refilled every quarter with fresh rates, not from redesigning the spreadsheet each time it misses.

Pair the forecast with a short list of leading indicators you check weekly rather than quarterly, because the point of the model is to find out you are wrong in week three instead of week eleven. The demand generation metrics page covers which six belong on that dashboard and which ones to delete. And if you want to see what the shape looks like in practice before building your own, the B2B SaaS demand generation examples library has the teardowns.