Skip to the page

See · Analysis & prediction

Who to act on, what comes next, and whether it worked.

Ready-made analyses list who to act on, why and what it is worth. Forecasts give a likely range and say how far off past forecasts have been. Predictions give the chance each open offer is won or each invoice is paid late, learned from your own history.

Works with Assistant Analysis Dashboards History Workflows
Tarnwick Supply Co.Boards · Orders forecastSample data
Orders forecast · monthly, next 3 monthsUpdated today, 06:00
Ask the assistantShare
Orders, $kForecastLikely range (80%)Holiday3Holidays and your own events marked
400500600JanMarMayJulSepNovJanforecastHoliday$498k
1A likely range, not one number
November, likely$512kRange $478k–$546k
December, likely$534kRange $490k–$578k
January, likely$498kRange $448k–$548k
  • 1A likely range, not one number
  • 2Says how far off it has been
  • 3Holidays and your own events marked

Four questions it answers.

Who should we act on?

Ready-made analyses read your systems on a schedule and list the rows worth acting on, each with its reason, such as customers going quiet.

Ready-made analyses

What comes next?

A forecast of any figure, by day, week, month or quarter, with a likely range.

Forecasts

What will happen to each open record?

The chance each open offer is won or each invoice is paid late, learned from your own history, with its reasons.

Predictions

Did acting on it work?

Mark what you acted on, and see how many of those rows left the list, against a small group held back.

Did it work?

Ready-made analyses: who to act on, and why.

Analyses that ship with nexibi:

Customers going quiet

Who has stopped ordering against their own usual gap, what their business is worth a month, and a next step, such as "Call today with a win-back offer".

Stock that isn't moving

Money tied up by store and item, each line with what to do: sell it off, stop reordering, check demand or order less.

Sold but not billed

Two systems compared customer by customer: sold and never billed, billed with no sale.

Plans at risk

Each plan against its actuals: at risk, losing money, delayed or on track.

  • Each run lists who, what to do, the value and one sentence on why. Earlier updates are kept.
  • They run when you choose: every day, twice a day, every week, or only when you press Update now.
  • Act on a row from its menu: send it through a follow-up flow, hand it to a team or a person, mark it handled, or ask the assistant about it.
  • You choose who sees each analysis: only you, every row, or each person their own rows.
  • Ready packs map the fields for Xero, QuickBooks, Stripe, HubSpot and Odoo, so there is nothing to map by hand. Xero and QuickBooks first ask your company to register an app with them.
  • Analyses need the data they read to be connected: customers and orders, stock, or plans.
Customers going quietUpdated todaySample data
9 customers to act on$38,400 a month
Brevin Hardware
No order for 41 days, usually every 14
$6,200 a month Call today with a win-back offerAct
A reason on every row
Send through a flow Hand to a person Mark handled Ask the assistant
Kessler Electrical
33 days, usually 12
$4,900 a month Send a reorder reminderAct
Halden Builders
29 days, usually 10
$3,700 a month Call today with a win-back offerAct

Forecasts that say how far off they have been.

  • Forecast any figure ahead: a sum, count or average from a saved query, an analysis, a history copy or a system's list, by hour, day, week, month or quarter.
  • Split it by up to two columns, such as product line and branch. The parts always add up to the total.
  • The engine tries several methods on your own history and keeps the one that did best. Methods that do worse than a simple benchmark are dropped.
  • Every forecast has a likely range, built from its own past errors.
  • It says how right it has been: off by how much on average, running high or low, and whether it beats "the same month last year".
  • It checks the data before every run and says what it found: missing periods, an unfinished period, duplicates. With fewer than 4 points it refuses.
  • A saved forecast refreshes itself on the schedule you choose, and says when it fell behind and why.
  • Money is forecast one currency at a time; currencies are kept apart.

says how far off
it has been

Tarnwick Supply Co.AssistantSample data
How many orders will we take in November?
Assistant

Likely $512k, between $478k and $546k (80% range).

$478klikely $512k$546k

Past forecasts were off by 6% on average. Off by how much, said every time

Orders forecastRefreshes daily at 06:00Share

Your calendar is part of the forecast.

  • Public holidays in 33 countries are taken into account, from 2015 to 2035, with festivals the calendars miss, such as Black Friday, Singles' Day and Mother's Day, and Ramadan and Eid in their phases.
  • Add your own events: promotions and price changes, once or every year.
  • Branches in different countries each follow their own holidays and time zone.
  • What-if. Ask what a week's promotion or a price rise would do, and see the forecast with and without it. If your history has never seen anything like it, it says so instead of guessing.
  • Overrides with a reason. Raise or lower a period and say why. The model's own forecast is kept beside yours, and later you see which came closer.
What ifOrders forecastSample data
15% off for one week in March
Assistant

Likely $41k more orders that week, with a wider range.

WithoutWith 15% off
Wk 1Wk 2Wk 3Wk 4+$41k
40% off for three weeks
Assistant
Says so instead of guessing

Predictions on each open record.

  • The chance each open record ends "yes": an offer won, an order lost, an invoice paid late.
  • Learned from your own history: one model per prediction, retrained every week, and sooner when it drifts. No model is shared across companies.
  • It needs 200 decided records to start, and tells you how many more it needs.
  • Every chance is compared with a simple baseline, and the baseline is used when the model is not better.
  • A column that gives the answer away, one filled in only after the result, is left out until a person confirms it.
  • Each record shows its reasons, with the line "These are patterns in your history, not proof."
  • Expected value: value times chance for each record, and the expected pipeline total with a range, each currency apart.
  • Each customer's next order: due, overdue or most likely stopped, ranked by the yearly value at risk.
  • Each person sees their own: share a prediction so each rep sees their own offers and each team lead sees the team's.

patterns in your history,
not proof

Open offersChance of winningSample data

When one offer was won and one lost, the won one came higher 73 times in 100.

OfferCustomerValueChance
OF-2231Lumora Hotels$14,20078%70–85%
Ordered again within 30 days twice beforeA reason on every chance
OF-2244Quellmark Foods$9,85064%55–72%
OF-2250Fenwright Builders$7,30058%49–66%
These are patterns in your history, not proof.
Share "Open offers"Sample data

Did it work?

  • Mark what you acted on. nexibi shows how many of those rows left the list within 30 days, and the money on them.
  • Keep a comparison group, a small share held back at random, and it compares what you acted on against what was held back. Until the difference means something, it says "not enough data yet". It gives no rate below 30 acted rows.
  • Same data, same answer: a second run on the same day changes nothing, and too little data ends as "not enough data", never as a thin list.
Did it work?Customers going quietSample data
Customers going quietLast 90 days
Acted on4226 left the list within 30 days$61,800 on them
Held back5not enough data yet
Against a group held back

It works together with the rest.

Set up by asking

The assistant finds the list, proposes the forecast or the prediction in plain words, and you approve a card.

AI Assistant

On a dashboard

A forecast is drawn with actuals solid, the forecast dashed and the likely range as a band. An analysis can sit on a board as a tile and keep its Act menu.

Dashboards

From the history engine

A forecast can read a history copy, and new data from the copy runs it again.

History engine

To workflows

Send a row through a follow-up flow from its Act menu.

Workflows

Matching records

The same customer found in two systems, by tax number, e-mail or name, and confirmed by a person.

Starter boards

My morning: what changed and what is worth doing today, with "Did it work?" first. A company: one customer across every analysis and what was done about it, before a call or a visit. The team: for each person on the team, what is urgent, its value and what was done.

For project businesses

Plans against actuals: for each project, job or contract on your plans, where it will finish, what it will cost, and when the money arrives.

Who uses it.

SalesWhich customers have gone quiet, and what are they worth a year?

It learns each customer's usual gap between orders and lists who is overdue and who has most likely stopped.

For sales
FinanceWhich open invoices will most likely be paid late?

Each one gets a chance with its reasons, and a flow can send the reminders you approve.

For finance
CEO or ownerWhat will next month's orders be, and how far off have past forecasts been? For CEOs and owners

What it does not do yet.

So you know before the demo.

  • Forecasts and predictions are made and changed through the assistant. There is no form, a saved forecast cannot be reshaped, and none can be deleted yet.
  • The forecast page shows the company total; splits show in the chat and on boards. The prediction page shows each record's chance, not its likely date.
  • A prediction needs 200 decided records. A company with less history gets none yet.
  • It does not explain why a figure changed, and it does not watch every figure for surprises: only actuals that fall outside a forecast's range are flagged.
  • New kinds of analysis are written by our staff, not by your company.
  • Material and dispatch plans are built but not offered until one has run on a customer's own data.
  • Market and tender watches, and research on the web, are not offered yet.

Bring the figure you want to see ahead.

In the demo we connect one of your systems, and you ask the questions.