How accurate are MRR projections?

From the SoleOS answers series — written about our own product space; grounded in published definitions and documented behavior, never invented numbers.

A projection is a conditional extrapolation, not a prediction — it answers "if the recent trend holds, where does this line go," not "here's what will happen." How much to trust one comes down to two things: how much history it's built on, and how wide the band drawn around it is. A tight, confident-looking line fit through three weeks of noisy data isn't accuracy — it's false confidence dressed up as a chart. A wide band drawn from months of steady numbers is the honest version of the same information.

What a projection actually is

Under the hood, a revenue projection is a trend line fit to your recent MRR history and extended forward. That's the whole mechanism. It doesn't know your roadmap, your marketing calendar, or that you're about to ship the feature you think will change everything, and it has no model of churn beyond whatever churn already happened. It's answering one narrow question: if the shape of the last several weeks kept going exactly as it has, where would this line sit in 30, 60, or 90 days? That's genuinely useful — it's just a different question from "how much will I actually make," and treating a projection as the second is where most of the disappointment, or false comfort, comes from.

Why short data windows make projections unreliable

With only a handful of data points, a trend line can look just as clean as one built on months of history — the math doesn't know the difference. But a fit through two or three weeks is really just connecting whatever happened during those weeks: a launch spike, one big annual subscriber landing on the same day, a slow week because you were heads-down fixing a bug instead of shipping. A projection can't tell noise from signal on its own — it needs enough history for the noise to average out and the underlying shape to actually show.

More data doesn't just make a projection prettier — it makes the extrapolation itself better calibrated, because a trend fit through ten weeks has survived more billing cycles, more good and bad stretches, than one fit through two. It's a large part of why a tool like SoleOS declines to render a projection at all until a project crosses a minimum amount of history: a short window can still produce a chart, but that chart would be more confident than the data supports.

The assumption every trend line makes: a constant growth rate

Every simple projection rests on one assumption — whatever growth rate produced the recent data will keep producing it. That rarely survives for long, and it's least reliable early. A product that went from $200 to $600 MRR in six weeks isn't compounding at some stable rate destined to continue; it's more likely riding a launch or a single early-adopter cohort. Early growth is lumpy — step changes from one customer, plateaus while you build the next thing, drop-offs when a channel that was working stops working. Even mature products rarely hold one constant rate for a full year: pricing changes, seasonality, and market saturation all bend the curve eventually. A trend line doesn't know any of that is coming — it draws the smoothest line consistent with what already happened and keeps going, reasonable close in and increasingly shaky the further out you push it.

Illustrative only: a product growing $400 to $500 to $625 MRR over three months is compounding at roughly 25% a month. Extrapolate that flatly and month seven lands around $1,900. If the real rate settles closer to 8% a month — still solid, just past the launch window — month seven is closer to $850. Both numbers came from the same three data points; only the assumption about what happens next differs.

What a confidence band actually means

A confidence band is a range, not a hedge on a single number. It reflects how tightly the fitted trend explains the actual data points behind it, and how much history that fit is drawn from. A wide band on three weeks of data is the model correctly telling you it doesn't have enough evidence to narrow things down yet — that's honesty, not uselessness. A narrow band on months of consistent numbers is earned confidence. A tool that shows one clean number with no band is quietly picking a single point out of a range and hiding the uncertainty around it; the band is the same forecast with that uncertainty left visible. It's why SoleOS holds back a confidence label — and holds back a projection entirely under a short window — rather than showing a smooth line with a straight face from day one.

Fitted trend vs. wishful straight-lining

A fitted trend is derived from actual numbers, with a band reflecting how much those points scatter around it. Wishful straight-lining looks similar on a chart but runs backward — someone picks a growth rate they want, or a milestone they're hoping for, and draws a line to it. "If I keep growing 20% a month, I'll hit $10k by December" is a spreadsheet cell, not a fit to anything that happened. One is honest math extended forward; the other is a goal wearing a trend line's clothing. A useful tell: if a projection lands suspiciously close to a round number you've been saying out loud, ask whether it was fit from your data or reverse-engineered from the target.

How to actually use a projection

Treat it as a directional gut-check and a milestone ETA, not a promise. As a gut-check, it's useful for noticing whether the business — given what actually happened, not what you hope happens — is pointed up, flat, or down. As a milestone ETA, "at the current rate, you'd cross $1k MRR around early Q4" is a useful planning input for hiring or spend, read as a rough window to revisit monthly rather than a date to circle. If you're chasing a number like $1,000 MRR, pair the projection with the work that gets you there rather than treating the line itself as the plan — this walkthrough of going from $500 to $1k MRR is more about the levers than the arithmetic. What it isn't for: promising a number to a co-founder or investor, or serving as the sole input into a big financial decision.

Red flags that mean don't trust this one

• Projecting off two or three weeks of data. Statistically closer to guessing than forecasting — not enough time for a real trend to separate from noise.
• Ignoring seasonality. Many products see real seasonal swings — a January bump, a summer lull, a back-to-school uptick. A trend fit through one side of a swing will overshoot or undershoot once the season turns; a projection has no idea a season exists unless the tool accounts for one.
• A number that suspiciously matches your stated goal. If it lands exactly on the milestone you were hoping for, check whether it's a fit or a wish.
• No band shown at all. A single confident line with no stated range is hiding, not eliminating, the uncertainty underneath it.

SoleOS published this guide as part of its content library on portfolio and revenue tracking for solo founders — it also declines to project a product's MRR until a minimum data window has passed, and labels confidence by how well the trend explains recent history rather than showing a line regardless. If you run one product and already eyeball your own growth rate weekly, a dedicated projection tool probably isn't solving a problem you have yet — a spreadsheet and your judgment get you just as far. It earns its keep once you're running several products and want projections, and their confidence, side by side without refitting a trend line by hand for each one. SoleOS's metrics dictionary documents how each figure — including projections — is calculated and where it can mislead, and the guides library has more on reading revenue numbers honestly.

Frequently asked questions

How many weeks of data do I need before a projection means anything?

There's no universal number, but a few days or a single week isn't it — not enough time for a real trend to separate from ordinary noise. A reasonable rule of thumb is waiting until the trend has survived more than one billing cycle and more than one unusually good or bad week. SoleOS, for instance, won't render a projection until a project has roughly three weeks of history, and tightens its confidence labeling further as that window grows.

Why did my projection change a lot when I added just one more data point?

Because the window behind it is still short. With only a handful of points, one new point can be a large share of the total evidence, so it can swing the fitted line noticeably. As more history accumulates, any single point becomes a smaller fraction of the whole, and the projection stabilizes. If one data point is still meaningfully moving your number, that's itself a sign the window is too short to lean on yet.

Does a wide confidence band mean the tool is bad at forecasting?

No — often the opposite. A wide band means the tool is accurately reporting that recent data doesn't yet support a narrow answer, which is the honest response to limited or noisy history. A tool that always shows a narrow, confident band regardless of how little data it has is the one to be more suspicious of.

Should I project ARR the same way I project MRR?

Not directly. ARR is just MRR annualized, so projecting ARR on its own risks compounding the same growth-rate assumption twice — once in the trend, again when it's multiplied out to a yearly figure. It's more reliable to project the MRR trend first, then annualize the result. For more on when each metric is the right one to reach for, see this comparison of ARR and MRR for small SaaS businesses.

Can I share a projection with an investor as a firm number?

Treat it as a scenario, not a commitment. If asked for a number, state the assumption behind it plainly — "at our current 30-day trend, holding steady" — along with the range around it, rather than presenting the midpoint as a promise. A projection that turns out wrong because the underlying trend changed isn't a failure of the math; it's exactly what "conditional on the trend holding" always meant.