KAISHA action plan

The numbers in Analytics tell you what happened. The Action plan tab in KAISHA tells you what to do about it.

Open KAISHA → Action plan and press Build my plan.

Where the advice actually comes from

This matters, because it is the difference between advice you can act on and advice that sounds good.

Kasenso works out the findings, not the AI. Every comparison, every peso figure, every "this is X% below that" is calculated from your own sales records by code — the same code that produces your reports. KAISHA is handed those finished conclusions and asked only to write them up in your own voice.

That means two things worth knowing:

• No figure in your plan is invented. If KAISHA writes a number that was not in the finding it was given, that action is thrown away and you are shown the plain version instead. You may occasionally see a slightly blunter sentence. You will not see a number nobody can trace.
• It can decline. A rule that does not have enough history behind it returns nothing at all. An assistant that always finds three insights is not reading your data; it is filling a page.

Every action has a Show the numbers link. That is not decoration — an insight you cannot check is one you should not act on.

What it looks at today

Weather and sales. What rain, snow or storms cost you per day, and — more usefully — which dishes take a bigger share of a wet day, so you can prep for the forecast instead of the plan. Needs your shop location set and at least six wet days and six dry ones on record.

Your forecast, when it is not trustworthy yet. A falling projection is the most alarming thing on the analytics page and, on thin history, usually an artefact of one soft month. This tells you when that is the case, and decomposes the drop: whether you served fewer customers or the same customers spent less. Those need opposite responses, and the chart cannot tell them apart.

More rules are coming. They are being switched on only once their thresholds have been tested against real shops rather than chosen at a desk.

Choosing the language

The plan has its own language setting, separate from the one your Facebook posts use: English, Taglish or Formal, next to the Build button.

They are separate on purpose. Taglish is usually right for a caption and often wrong for a page telling you your bestseller barely makes money — plenty of owners run a Taglish Facebook page and would rather read their numbers in plain English. Pick per plan; it remembers what you used last.

No option uses emoji. English is plain English said straight, not friendly English — if you asked for English, you probably wanted to be told the thing directly.

Reading a plan

Each action carries a peso figure per month. That is the ranking currency, on purpose: a 40% swing on something selling four a week should never outrank a 6% swing on your bestseller.

Mark an action Done when you have acted on it, or Not for us when it does not fit your shop. Some good advice is wrong for a particular business — coin handling, kitchen space, the staff you actually have — and you know that where your sales figures do not.

The limits

• Premium, alongside Analytics, whose data it reads.
• Five plans a day. A plan is a weekly thing; regenerating it hourly will not produce better advice, because the underlying numbers have not moved.
• It proposes, never acts. No prices are changed, no promos created. If an action suggests a price change, you open Price Lab and decide for yourself.

When it says it has nothing

That is a real answer, not a failure. Most rules need a few months of trading, and the weather ones need a wet season. Keep selling, keep your location set, and the plan fills in.

Three rainy days is not evidence. A confident figure drawn from three days is worse than no figure at all, because you would prep against it.