Quick start: from installation to the first report
What to click after installation, what the system does meanwhile and where people usually stumble — from granted access to a report the whole team opens.
measured 25.08.2026 · first full load of the pilot account (anonymised property developer account) · docs/РЕШЕНИЯ.md, пункты 37 и 40
Permissions, the letter for the administrator and what happens after access is revoked are covered on the installation page. Here is the path to the first number. Two of the five steps have no interface yet: the stage markup and the analytics field list are switched on by us, at your decision. More on that below, in a separate block rather than in fine print.
Five steps in order
Installation and first data load
You. The account administrator opens the installation link and clicks “Allow”. The link is issued for a specific account: until the widget is in the marketplace, there is no public install button. Then just wait — nothing else to click. Which permissions are requested, and the letter for the administrator.
The system. Reference data first — pipelines, stages, users, deal fields. Then deals and the status change history, in date windows. Transitions are assembled from the history: from where, to where, when, who moved the deal and how long it sat on the previous stage. The screen shows a percentage, not a spinner; an interrupted load resumes from the same point, not from the start.
How long. Granting access — 2 minutes, an estimate. The first load — 34 minutes on 7 years of history, and that is a measurement: 69,567 deals, 1,100,000 events, 240,031 transitions. A younger account loads faster: the time depends on the volume of history, not the size of the company. After that, sync runs incrementally — 12 seconds every 5 minutes.
What goes wrong. The integration sees exactly what the person who granted access sees. If some pipelines are hidden from them or dismissed employees are hidden, the history arrives incomplete — and this does not show up as an error, the report simply lacks people and pipelines. The sign: the pipeline list is missing pipelines that exist in the CRM, or “Managers” lacks a person who definitely worked in the period. The fix is to grant access from an administrator with full visibility.
How to check. We do not show reports on half of the history: until the first load finishes, the screen shows a percentage, not half a funnel. A “last thirty days first” mode is planned, but it is not in the code — so it is not in the promise either.
Confirming the stage markup
You. Review the list of stages marked as parking and say what does not belong there. A parking stage is one where a deal is not moving towards a sale but waiting for a call, a decision or the season. The manager signs the decision, not the algorithm. What a parking stage is and how to tell it apart.
The system. The heuristic goes through the loaded history and proposes a markup. What a person has confirmed is stored separately from the heuristic’s output and overrides it: a reload or re-markup does not touch confirmed stages.
How long. 1 minute for the list — an estimate from the pilot. The conversation about what counts as waiting in your pipeline is usually longer, and that is fine: every number in the report depends on this decision.
What goes wrong. The heuristic makes mistakes, and we measured it: on the pilot’s full history it declared 7 stages as parking instead of 3 — including the one from which 98% of closings go to won. No threshold fixes this, which is why a person signs the list. The full post-mortem, with numbers.
How to check. Check yourself: in the “Funnel” tab, parking stages are taken out of the chain and listed separately. If that list contains a stage after which deals turn into money, the markup is wrong — and the conversion of the whole chain is computed on the wrong steps.
Choosing analytics fields
You. Name the deal fields you need in breakdowns: project, source, loss reason, property type. By default nothing beyond system fields is synced, and a field that could contain a name, phone number or e-mail address never makes the list at all.
The system. Completeness of every field is computed right on the deal export: the counter stores only numbers — how many deals have the field filled in — and field values are never written anywhere. So completeness is known for every field in the account, while only a handful are stored. Then the verdict: 60% and above — the breakdown is built silently, from 30 to 60% — with a warning, below 30% — not built.
How long. No separate wait: completeness is computed inside the first load, in the same pass over the deals. A field’s mark survives reloads — a reference data refresh does not erase it.
What goes wrong. A deal has many fields; only a few are fit for breakdowns. The pilot has 104; of the 8 fields examined as breakdowns, above 60% — 0, between the thresholds — 2, below 30% — 6. The most complete is “Причина отказа” at 41%. This is not a widget defect and not us being picky: a breakdown by such a field would show who fills what in, not how sales are going. What to do with a field below threshold.
How to check. The “Data quality” tab spells the verdict out in words: “can be counted”, “the breakdown will cover less than half”, “breakdown cannot be built”. Separately there is “not measured” — that is not zero, it is the absence of a measurement.
The first view
You. Pick a pipeline and a period — that is already enough to see numbers. Then narrow down: group, manager, deal field. “Compare” adds a second period, and the widget computes the deltas.
The system. The filter bar applies to every tab at once, and switching tabs resets nothing: overview, funnel, lead path, customer journey, managers, data quality and the AI review are all computed from one view. When the period is a whole calendar month, the “Overview” shows a plan card: deals, won, revenue, attainment and a pace forecast. Its actuals are taken across the whole pipeline; filters do not narrow them.
How long. No measurement. The view is computed from data already loaded — the widget does not call amoCRM at this moment, and changing a filter does not start a new sync.
What goes wrong. In places the widget will refuse to show a percentage. With a base under 8 deals it shows “not enough data” instead, and that is not zero. Conversion above 105% is not an error either: deals enter a stage not only from the previous one, and such a value is flagged, not hidden. Transitions made by automation go on their own line and do not enter the team median. Where every number comes from.
How to check. Start with “Data quality”, not “Funnel”: it determines which breakdowns are worth opening at all. Second — “Funnel”: it immediately shows whether the markup from step 2 is right.
A saved report
You. The “Reports” button → “Save current…”, a name and, if colleagues need the report, the “Visible to the whole team” checkbox. A link to the view is the “Copy report link” button in the filter bar.
The system. The report remembers the view, the open tab, the set of print blocks and the heading. Manager and project are stored as stable keys — the amoCRM user id and the field value — not as positions in a list: lists are rebuilt for every period, and a report saved by position would show someone else’s numbers under the same name tomorrow. A saved report’s link is short — the report id and the language, with no list of filters: the recipient opens the fresh version of the report from the server, not a frozen set of parameters as they were at copy time. The session key never goes into the link — sharing a link is not sharing access, the recipient goes through their own authentication. Inside amoCRM the link leads to the widget page in the CRM itself.
How long. One click. No more than 30 reports per author: on an attempt to save the next one the widget says “Limit of 30 reports per user. Delete the ones you do not need” rather than silently losing the report.
What goes wrong. A manager left, a project vanished from the data, a pipeline was deleted — the filter honestly resets to “all”, and a line appears on top: “The manager from the report is not in this period — all managers are shown”. The widget does not substitute the next name in the list. A deleted report answers by link with “The report did not open. It may have been deleted”. The demo has no saved reports at all: they work when signed in from amoCRM, and saving a view over invented data would itself be an invention.
How to check. Private and shared differ in permissions, not in appearance. Anyone in the account can open a shared one; only the author can rename, overwrite, un-share and delete it — an edit attempt on someone else’s report answers “not found” and gives no hint that it exists at all.
estimate, not a measurement · granting access — 2 minutes, confirming the markup — 1 minute: only the first load was timed with a stopwatch, the rest is an estimate from the pilot
Two steps that have no screen yet
Both decisions work and are stored in the database — what is missing is the interface for making them on your own. For now this is done together with us at onboarding and by writing to support: there is one account, and the conversation happens anyway.
Confirming the stage markup
screen in progressThe heuristic marks up on its own; what a person confirms overrides it and survives reloads. There is no “this is not a parking stage” button in the widget yet: send us the list — we correct it and recalculate. The parking list is always visible, as a separate block in the “Funnel” tab, so you can check the markup without us.
The analytics field list
screen in progressCompleteness is computed for every field in the account and shown in the “Data quality” tab. The “this field is needed in breakdowns” mark is set on our side and is not erased by a reference data refresh. Fields that could contain personal data never make the list, no matter who asks.
The full list of what the product cannot do yet has its own page — what we cannot do yet.
What to open first once the load is finished
The order is not random. Starting with “Funnel” is tempting, but if the fields are empty and the parking stages are marked up wrong, the pretty picture becomes the first number you have to retract.
1. Data quality
Completeness of every field and the verdict on it. Determines which breakdowns are worth opening and which the widget will refuse to build.
2. Funnel
Conversion between adjacent steps, with parking stages as a separate list. A check of the markup from step 2: a stage that does not belong in the parking list stands out at once.
3. Lead path
Rollbacks, skipped steps and moves to other pipelines. This is usually where the answer to “where did the leads go” is found.
Walk the route without installing
Steps 4 and 5 can be tried right now: the demo runs the same filter and the same tabs on anonymised data. There are no saved reports there — they are tied to an amoCRM account, and saving a view over demo data would be an invention.