Set the Baseline Before You Start, or You'll Argue About It Later

Every month-six argument is really an argument about month zero

The conversation goes the same way every time. Month six, a room with a founder, a marketing lead and an agency. Somebody says traffic is up. Somebody else says it doesn't feel like anything changed. A chart appears. Then a second chart with a different date range, and the meeting quietly becomes a negotiation about which chart is real.

Nobody in that room is lying. They're arguing because nobody wrote the number down on day one, so both sides are now free to reach for whichever comparison flatters them. The agency picks the month before the last core update. The client remembers a good week in March that probably didn't happen. Six months of work gets judged on rhetoric.

A baseline is the cheapest insurance in the whole engagement. It takes one week, costs nothing, and it's the only reason a difficult month-six conversation stays a conversation about the work instead of a conversation about the reporting. We do it before any billable work happens, and we'd tell you to do it even if you were hiring someone else.

The six numbers worth freezing

One number is too easy to argue about, and twenty numbers means nobody reads any of them. Six is the point where the picture stops being deniable — because the six fail in different directions, and the pattern between them tells you what's actually happening.

Each of these lives in a system somebody already owns. None of them requires a new tool.

• Don't freeze average position. It falls while a site is winning, because new long-tail queries enter at position 40 and drag the average down. It's the worst number on most reports and it's on most reports.
• Don't freeze domain authority or domain rating. They're third-party estimates of Google's opinion, recalculated whenever the vendor changes their model. You can't hold anyone to a number a vendor can revise.
• Don't freeze total sessions. Paid, direct, referral and a bot or two are all in there. Organic-only, non-brand, or it isn't measuring the work.
• Do keep the raw exports. CSV files, dated, in a shared folder both sides can open. A number in a slide is an assertion. A number with its export attached is evidence.

Picking a window nobody can accuse you of choosing

The window is where baselines get quietly rigged, usually without anybody intending to. An agency that picks the trailing quarter containing a two-week site outage has set itself an easy target. A client who picks their best-ever quarter has made the guarantee unwinnable. Both look reasonable at the time and both poison month six.

Five rules that make the choice defensible, and all of them work because they remove discretion from whoever benefits.

1. Trailing 90 days, ending the day before kickoff. Not the last full calendar quarter, which can be four months stale by the time anyone signs. Not a hand-picked stretch. The 90 days immediately behind you.
2. Attach 24 months of monthly context. The baseline is 90 days; the sanity check is two years. Export monthly organic leads and non-brand clicks for 24 months and put the chart next to the figure. It takes ten minutes and it catches every distortion in this article.
3. Compare the window against your own 12-month median month. If the baseline period sits far above or far below that median, you've caught seasonality, an outage or a campaign. Say which, in the document.
4. Name the known distortions before you agree the number. The fortnight the contact form was broken. The LinkedIn post that went unusually wide. The month sales stopped answering the phone because two people quit. Both sides list what they know, in writing, before the number is fixed.
5. Fix the definitions at the same moment as the number. What counts as a qualified lead, which brand terms are in the regex, how a lead gets tagged organic. A number without its definitions isn't a baseline, it's a screenshot.

The Indian calendar problem, and why generic advice gets it wrong

Most baseline advice is written for markets where the year is roughly flat. Indian demand isn't, and the two largest distortions don't sit still on the calendar — they move.

The festive block, roughly Navratri through Diwali, is the biggest retail and D2C demand window of the Indian year. It's set by a lunisolar calendar, so Diwali can land anywhere from mid-October to mid-November depending on the year. An October–December baseline can therefore contain the entire festive peak one year and part of it the next. Compare those two windows as though they're equivalent and you've credited or blamed SEO for a calendar.

The second is the financial year. India's fiscal year ends on 31 March, which pulls B2B procurement, budget-flush purchasing and renewal decisions into January–March, then empties April while new budgets clear approval. A B2B business that freezes a baseline across January–March and gets measured across April–June is being asked to beat its best quarter using its worst one. That's not a hard target. It's an impossible one, and it will end the relationship in month four for reasons that have nothing to do with the work.

Other patterns worth checking in your own data rather than taking on trust from a list like this: education demand tracking board results and admission cycles across roughly April to July, wedding-linked categories concentrating between November and February, and monsoon suppressing construction, travel and outdoor categories across much of the country from June to September. None are universal. All are visible in twenty minutes of your own Search Console history, which is the only version that matters.

• If the window is materially off your median, pick one of three fixes and write down which: shift the window to a neutral quarter, keep the window but make year-on-year the real test, or measure a rolling 90-day figure against the same 90 days last year.
• Year-on-year is the honest instrument for seasonal businesses. Comparing this Diwali to last Diwali removes the calendar entirely. It costs you speed — you can't read it in 90 days — which is exactly why some businesses shouldn't buy a 90-day guarantee at all.
• Watch the festive date drift explicitly. If Diwali moved by three weeks between the baseline year and the measurement year, note the shift beside the number so nobody has to rediscover it in an argument.
• Regional variation is real. A brand selling into Kerala, Bengal and Punjab is looking at three different peak calendars. If your traffic splits sharply by state, split the sanity check too.

The baseline document, and why both sides sign it

A baseline that lives in somebody's head is not a baseline. Neither is one that lives in a kickoff deck nobody opens again. It needs to be a single page, dated, with two names at the bottom, and it needs to exist before the first invoice.

Ours runs to about 400 words. Here's everything that has to be on it.

1. The source of truth for each number, named as a specific system. "The CRM" is not a name. "HubSpot, Deals object, Lead Source = Organic Search" is.
2. The definition of a qualified lead, one paragraph, with two examples of what counts and two of what doesn't. If sales already uses disqualification reason codes, adopt those instead of inventing a parallel vocabulary.
3. The channel attribution rule. First-touch or last-touch, UTM conventions, the hidden form field, and how phone and WhatsApp enquiries get captured — because in most Indian service businesses a meaningful share of organic leads arrive that way and attribution loses all of them by default.
4. The brand regex, written out in full, frozen. An agency that adds terms to the brand filter later can manufacture non-brand growth. One that removes terms can make it vanish.
5. The window, as two dates, plus the 24-month context chart as an attachment.
6. The six numbers, as six figures, with the raw export filenames listed next to them.
7. The documented distortions and exclusions. Outages, campaigns, the TV spend planned for month two, existing customers re-enquiring, internal test submissions, job applications.
8. A named arbiter on each side — one person, not a department — who settles edge cases monthly without it becoming a renegotiation.
9. The date and two signatures. This is the part people skip and it's the part that does the work. A signed page changes what month six can be about.

What invalidates a baseline, and what only annotates it

This is the section most agreements never write, and it's the one that decides whether month six is calm. Some events genuinely break the comparison. Most don't — they just need a note on the chart. Deciding which is which after the event is how both sides end up feeling cheated, so decide now, in advance, when neither of you knows who it will favour.

Setting one for yourself in an afternoon

If you're not hiring anybody, or you're already mid-engagement with an agency that never did this, you can build the whole thing yourself in about two hours. It's worth doing before your next quarterly review rather than after it.

1. Export 24 months of monthly organic leads from your CRM. Filter to organic source only. Put it in a sheet and look at the shape before you look at any total.
2. Build the brand regex and save it as a Search Console view. Every spelling of your name, the domain, common misspellings, the founder's name if people search it. Export 24 months of non-brand clicks and impressions alongside the lead data.
3. Count distinct non-brand queries with at least one click for the trailing 90 days. One number. Write it down.
4. Compare indexed pages to published pages. Search Console's Pages report against your CMS count. If the gap is large, that's your next quarter, whatever anybody's content plan says.
5. Export referring domains from whichever backlink tool you have access to, and keep the file rather than a screenshot.
6. Calculate organic session-to-enquiry conversion for the same 90 days, and write down how you calculated it.
7. Write the definitions page and send it to your agency. Ask them to confirm it in writing. What they say next is more informative than any audit they'll ever send you — a good agency will improve it; a nervous one will explain why measurement is complicated.

What we freeze, and what we'll never promise

We spend the first week of every engagement on this and we don't bill for the delivery work until it's signed. It has cost us clients who wanted us writing in week one, and it has saved every relationship that later hit a hard quarter.

The number we commit against is one of the six: your trailing-90-day count of qualified leads from organic search, frozen on day one with its definitions attached. If we haven't beaten it within 90 days, we keep working free until we do. Not a ranking position, ever — nobody controls Google's index, and a promised position is a promise about somebody else's system. What an SEO guarantee costs the agency explains why that commitment caps us at three new clients a month.

The other five numbers go on the monthly report next to the frozen figures, because the headline number alone can't tell you why anything moved. That reporting shape is covered in how to judge your SEO agency's performance, which works fine as a checklist against an agency that isn't us.

SEO runs from ₹75,000 a month, and from ₹40,000 for smaller sites, ex-GST. Month-to-month after the first quarter, 30 days' notice, and you keep every asset including the baseline document if you leave. The guarantee in full, and what SEO actually costs in India if you want the arithmetic before the call.