What a baseline is
A baseline is a written record of where things stood on the day work began. Not an impression, not a figure somebody half-remembers from a meeting — an exported, dated set of numbers, each with its definition attached, stored somewhere both sides can see it.
For a marketing project it usually covers organic and paid traffic and where that traffic came from, enquiries or orders and what each one cost, revenue where revenue can honestly be measured, current positions for the keywords you intend to target, and the technical state of the site: indexed pages, speed, obvious errors. The point is not to collect everything available. It is to fix, in advance, the small set of numbers that will later decide whether the work is judged a success.
Why a baseline matters
Without one, every conversation about results becomes a memory contest, and memory reliably favours whoever is talking. A dated export ends that argument. It protects in both directions, too: when a client says traffic was better before, the baseline either agrees or it does not, and either answer moves things forward.
There is a quieter benefit. Writing the baseline forces you to define terms while nothing is at stake. What counts as a lead? Does an unanswered phone call count? Are we reporting sessions or people? Those arguments are cheap in the first week and expensive in the sixth month.
Common mistakes with baselines
The oldest trick in reporting is choosing the starting point. A baseline taken during a quiet month makes almost any later figure look like progress, and anyone who checks the dates can see it. Take a period that fairly represents normal trading, and state plainly which period you took.
The second mistake is trusting a platform to keep the data for you. Analytics tools change definitions, adjust their models and discard detail after a retention window, so a figure you never exported may simply not be there when you go looking. The third is quietly redefining a metric mid-project — adding new form types to a conversion count, or switching attribution model — which makes the comparison meaningless even when nobody set out to mislead. If a definition has to change, restate the baseline under the new definition and show both.
How to set one properly
Take a period long enough to absorb normal variation, and take the equivalent period from the previous year alongside it. In Nepal this matters more than in most markets: the festival calendar moves against the Gregorian one, so the same calendar month can contain Dashain in one year and not the next, and a year-on-year comparison made carelessly will mislead you.
Export the raw figures to a file, date it, note anything unusual in the period — an outage, a sale, a campaign that ended — and keep it outside the platform it came from. Then record annotations as the project runs, so that when a line moves months later you can say why rather than guess. Reporting built this way survives scrutiny, and that discipline is what separates useful dashboards and reporting from a monthly slide nobody trusts.