Forecasting SEO ROI for SaaS without inventing numbers

Most agency forecasts are constructed backwards. Someone picks a number that justifies the retainer, then works out what traffic growth would produce it. The giveaway is a smooth curve, because real search programs move in steps as individual clusters break into the top three.

A forecast built the right way is uglier, lower, and defensible in a room with a finance team in it.

Building it bottom-up

The unit is the keyword cluster, not the site. For each cluster on your priority map you need six inputs.

Realistic search volume, with a haircut. Take the tool number and adjust it against what Search Console actually shows for terms you already rank for. On narrow B2B accounts, we typically find tool volumes are understated for long-tail and overstated for head terms, so the adjustment goes both ways.

A target position by month twelve, based on the current top ten’s link profiles and content depth versus what you will realistically build. Be pessimistic. A cluster where the top three all have 200-plus referring domains and you have 40 is not a top-three target in twelve months.

A click-through rate for that position. Use your own Search Console data by position if you have enough of it, because CTR varies enormously by query type and by how many SERP features sit above the results. Generic CTR curves published by tool vendors are a last resort and they overstate, particularly for queries that trigger AI answers.

A conversion rate, taken from a comparable page you have already built rather than a benchmark. If you have no comparable page, use a range and say so.

Your actual close rate on organic-sourced opportunities, from the CRM. Not your blended close rate, which includes referrals and outbound.

Your average contract value, and if you sell multiple products, the mix.

Multiply through, sum the clusters, and phase it across the year according to when each page ships and how long that cluster type takes to rank. That phasing is what makes the curve step-shaped and honest. Our SEO calculator runs a simplified version of this model if you want a rough number before doing the full exercise.

Three scenarios, and stating the error bar

We present low, expected, and high. The low case assumes the two largest clusters underperform and publishing slips by a quarter. The high case assumes everything ranks a position or two better than projected.

Then a sentence that has done more for our client relationships than any result: the expected case has roughly 40% error on it in the first six months, narrowing to maybe 15% by month twelve as real data replaces assumptions.

Finance people are entirely comfortable with a range and an error estimate. What they are not comfortable with is a single confident number that turns out wrong, because it costs the person who presented it internally.

The assumptions that break

From reconciling forecasts against actuals across a few dozen programs, the failures cluster.

Publishing capacity is the most common by a wide margin. The plan assumed eight pages a month, the client’s review cycle supports three, and by month five the roadmap is a quarter behind. This is almost always a review bottleneck rather than a writing one, and it is predictable at the start if you ask the right question: how long did your last five pages take from brief to live?

Search volume being smaller than projected is second, and it is a real risk in narrow verticals. This is why we haircut and why the low scenario exists.

Competitive difficulty being higher than the difficulty score suggested is third. Those scores are a function of link metrics and they miss brand strength, SERP feature dominance, and the possibility that the top result is a review site your domain will never displace.

Conversion rate assumptions are fourth and usually the least wrong, because they are grounded in your own pages.

What rarely breaks: the ranking timeline for bottom-funnel commercial pages. Those have been consistently 8 to 14 weeks across enough accounts that I would treat a big miss there as a signal that something technical is wrong.

Presenting it to finance

Three things make the difference.

Express it in the same units as everything else in the business case: pipeline generated, cost per opportunity, and payback period in months. A traffic number in a finance meeting is noise.

Compare it explicitly to your marginal cost of acquisition through paid. Organic usually loses badly for the first five months and wins decisively from month nine, and showing both halves of that is more credible than showing only the second.

Include the terminal value argument, carefully. The pages you build in year one keep producing in years two and three at a maintenance cost of maybe 20% of the build cost. That is the actual case for SEO over paid, and it is also the part that sounds like special pleading if you overstate it. We model year two at 60% of year one’s incremental gain, which is conservative against what we typically see.

When the forecast is wrong

Re-forecast quarterly against actuals and show both lines on the same chart. Hiding the original projection once it diverges is the single fastest way to lose a client’s trust, and everyone notices.

The conversation to have at month seven is not “were we right” but “which assumption was wrong and what does that imply.” If publishing capacity was the constraint, the fix is organisational and the forecast holds with a shifted timeline. If the demand was not there, that is a strategy problem and the honest move is to say the category has less search opportunity than projected and here is what we would do instead.

We have had that second conversation twice. Both times the client kept working with us, which surprised me. I think the alternative — quietly adjusting the target and hoping nobody compares — is more common and more damaging.

A rough sanity check

For a B2B SaaS company with a $30,000 ACV, a 25% close rate on organic opportunities, and a $12,000 monthly all-in spend including link budget, breaking even means producing about 1.5 additional closed deals a month by the end of year one. Working backwards through a 5% page conversion rate and a 30% opportunity rate, that is roughly 400 well-targeted organic sessions a month.

Four hundred. Not forty thousand. If a forecast requires tens of thousands of monthly sessions to work, either the ACV is small or somebody is targeting the wrong keywords.