What to fix first when conversion is bad everywhere

When nothing on a site converts, owners tend to conclude that everything needs replacing. That instinct produces a six month project, a large invoice, and a new site that often converts the same, because nobody diagnosed anything before rebuilding.

The alternative takes a week. Check things in the order of how much they cost you and how cheap they are to repair. Most of the time you find two things, both small, both fixable by Friday.

What order should I fix things in?

Four tiers, and you do not move to the next until the current one is clean.

Tier one, contact paths. Can a person on a phone reach you. Is the number a link that dials. Does the form actually deliver to an inbox someone reads. Does the email address on the site still exist. Every failure here is total, silent, and free to fix, which is why click to call is still missing from more sites than anyone would guess and why it is always the first check.

Tier two, clarity. Does the first screen say the service and the city in words a customer would use. If a stranger cannot tell what you sell in five seconds, nothing further down matters.

Tier three, traffic quality. Are the visitors people who could plausibly buy from you. A site cannot convert an audience that was never in the market.

Tier four, everything the category likes to talk about. Trust content, copy, layout, speed, offers. All of it real, none of it worth touching while tier one is broken.

A broken contact form costs more than a bad headline, and it takes less time to fix.

The reason for that order is not taste. It is that tier one failures are total. A form that does not deliver converts at zero regardless of how good the copy above it is, and the loss is invisible because nothing on your screen looks broken. Tier four improvements are marginal by nature. You are moving a rate, not repairing a wall.

Losses at the contact layer are large enough to be measured elsewhere. Patient Prism's analysis of 11,552,668 dental patient calls across 8,280 locations in 2025 found that 31 of every 100 calls were abandoned before reaching an agent, and separately attributed 9.3% of all lost opportunities to voicemail, disconnects and hold hangups. That is dental, and it is a call intelligence vendor reporting on its own customers, so it describes practices organized enough to buy call tracking. The relevant point is the shape: a large share of demand dies at the contact step, after the marketing already worked.

How do I know it is not a traffic problem?

Look at where the traffic comes from before you rebuild the page it lands on.

If most visits arrive from searches for something you do not sell, no page change will fix that. This is common after a blog post ranks for a general question, or when broad match ad targeting brings in browsers instead of buyers. Reading your search terms report settles it in ten minutes, and the fix is in the campaign rather than the website.

If visitors arrive from outside your service area, same conclusion. Check the geography before you change the layout.

And if sessions are very short across the board, on the order of a few seconds, that usually means the promise that brought them did not match what they found.

Split direct and branded traffic out from the rest while you are looking. People searching your business name convert at a different rate than strangers, and mixing them makes a mediocre site look fine or a good site look broken.

Why not just redesign the whole site?

Because a redesign changes everything at once, which means you learn nothing and you inherit new problems.

If conversion improves after a rebuild, you cannot say which of the forty changes did it. If it gets worse, you cannot say either, and you now have to argue about a site you just paid for.

There is also the risk of losing something that was working. Rebuilds routinely drop the trust content, break the tracking, change the URLs, and remove a phone number from the header. I have seen more sites converted downward by a redesign than upward.

Then there is the pricing question, and it deserves a warning. Every published range you will find for what a small business website costs to build or maintain traces back to agency pricing pages rather than to a survey. There is no study with a disclosed sample behind "$2,000 to $10,000 to build" or "$500 to $1,500 a year to maintain." Those are quotes, aggregated by people who also sell the service. Even the one survey-shaped figure in circulation, a claim that a majority of small business buyers spent under $10,000 on their most recent project, is attributed to a report whose methodology has not been confirmed. A price you were quoted is a fact about that vendor. It is not a benchmark, and nobody should present it to you as one.

The honest exception is when the site genuinely cannot be repaired. If it is built on something unmaintainable, if nobody can edit it, or if it fails on phones in ways that cannot be patched, rebuild it. That is a platform decision rather than a conversion decision, and it belongs with the wider question of what your business actually needs, including whether branding work should come before the website rather than after.

Otherwise, make one change at a time and keep the site you have.

What does the rebuild have to deliver to pay for itself?

Run the break even before you sign anything. Four lines, your numbers.

Where you are. 18 inquiries a month. You close one in three, so six customers. Average job $2,400 at a 40% gross margin is $960 of gross profit each, so $5,760 a month.

What it costs. A rebuild quoted at $12,000. Spread over twelve months, it needs to produce $1,000 a month of additional gross profit just to break even in year one.

What that means in customers. $1,000 divided by $960 is 1.04 extra customers a month. At a one in three close rate that is about three extra inquiries a month, taking you from 18 to 21.

Now ask the question. What evidence do you have that this rebuild produces three more inquiries a month? Not a hope, evidence. Nobody in the sales process offered any, because none exists, and you cannot test your way to it either. Conversions needed per variant is roughly 16 ÷ r², where r is the relative lift and the 16 is built from 2(1.96 + 0.84)², the 1.96 for 95% confidence and the 0.84 for 80% power. The familiar hundred conversions per variant is therefore exactly what detects a 40% lift (derived), and a site with 18 leads a month has nine per arm, where the smallest resolvable difference is about 133%derived.

Now price tier one against the same yardstick. Making the header number tappable across every template takes about forty-five minutes and costs nothing. Recover one inquiry a month and you have four extra customers a year, roughly $3,840 of gross profit for forty-five minutes. Recover three and you have matched the rebuild's break even without spending $12,000.

That is the whole argument for the tier order, expressed as arithmetic rather than opinion. Rerun it with your own job value, margin and close rate before your next quote.

How long before I know whether it worked?

Longer than you want, and the answer depends on your volume.

At local traffic levels, a week tells you nothing. A month is barely readable. A quarter compared against the previous quarter is usually the smallest window that means anything, which is the practical reality behind measuring conversion honestly when traffic is small.

There is one exception worth knowing. If you fix something that was completely broken, a dead form or a number that never dialed, you will see it immediately, because you went from zero to something. Total failures produce obvious recoveries.

Everything else is a slope, not a step. Track the count of real inquiries monthly, write it down outside the analytics tool, and judge on direction over quarters.

Resist the urge to keep changing things while you wait. Five changes in one month leaves you with a result you cannot attribute to anything.

What about speed, and where does it belong?

Tier four, and it is genuinely real there.

Google's own mobile benchmark page, the February 2018 edition, reports that as the number of elements on a page goes from 400 to 6,000, the probability of conversion drops 95%. That figure comes from a neural network trained on bounce and conversion data with a stated 90% prediction accuracy, published by Google, which sells advertising. Its current Core Web Vitals thresholds put Largest Contentful Paint at 2.5 seconds or less for a good rating, worse than 4 seconds for poor, measured at the 75th percentile of visits.

Two cautions on that first number. It is about element count, meaning the sheer quantity of things on a page, not seconds. And Google has published its mobile bounce statistic in three different editions with different figures, so if someone quotes you a bounce percentage without naming which edition it came from, they have not read the source.

Speed still belongs in tier four, because a slow page that loads is a rate problem and a form that does not deliver is a wall. Fix the wall first.

What to do this week

Monday, run the contact path check on a phone. Tap every number. Submit every form. Send an email to every published address. Fix whatever fails, today.

Tuesday, screenshot your homepage on a phone and crop at the fold. Ask three people who do not know your business what the company does and where. Rewrite the headline if they get it wrong.

Wednesday, open your traffic sources and your search terms. Decide whether the visitors were ever plausible buyers.

Thursday, count the fields on your form and cut every one you do not use, since each field costs you submissions and this is the cheapest tier four fix there is.

Friday, write down this month's inquiry count. That is your baseline, and it is the only way you will know whether any of this week's work mattered.

Questions about where to start?

Email me at eric@seod.com with your website address and the three questions your customers ask most often before they book. I will send back a short ordered list of what I would fix first, second, and third for your specific site, with the reason for the order.

It is one email, no call, and no proposal attached. If the honest answer is that your site is fine and the problem is upstream, that is what I will tell you.

Otherwise, keep reading the conversion library.