Using Attio for Competitive Analysis

Most CRMs store a competitor the same way: as a value inside a property on a deal. That is a string. It cannot carry a note, a document, a date you last checked something, a contact, or a list of the fourteen deals it appeared on. Anything you learn about the rival has to live somewhere else, which is why most competitive programmes end up maintaining a spreadsheet alongside a CRM that already contains half of what the spreadsheet is for.

This system is built the other way round. Its data model starts from objects, records and attributes rather than from a fixed set of sales entities, so a competitor can be a record in its own right, joined to deals through a reference rather than described by a label. Once that is true, the competitive set stops being an export and becomes a queryable part of the same database as your pipeline.

Two things follow, and the page spends most of its time on them. The record can carry data about the competitor as a company, some of which arrives automatically, which is unusual and genuinely useful. And that automatic data behaves differently from everything else in the workspace when you try to take it anywhere, which is the constraint to know before you build a process on it.

1. 1Decide where a competitor lives before you create anything. Attio enables People and Companies by default and allows a total object count that runs from three on the free plan to unlimited on enterprise, with custom objects reserved for the paid tiers. So the first decision is whether rivals become company records alongside your customers or their own object. Both work. Only one of them keeps competitors out of every list and filter by construction.
2. 2Join deals to the competitor with a record reference. Add a reference attribute on the deal pointing at the competitor record. Attio documents these as bidirectional, so updating either side updates the other, which means the competitor's record shows the deals they contested without anybody maintaining a second list. This single attribute is what the rest of the method rests on.
3. 3Add the attributes a deal field could never hold. On the competitor record, not on the deal: their entry price, the segments they win in, the claim your sellers hear most, the date each was last checked. That last one matters more than it looks, because a competitive record with no last-checked column tells you nothing about whether you are reading a current fact or a two-year-old one.
4. 4Read the enriched figures as bands, never as numbers. Attio fills in estimated ARR, employee range, funding raised and foundation date automatically from its data providers. Treat the employee range and the funding figure as usable, since both trace to observable filings and profiles, and treat estimated ARR as a model output. It is a reasonable way to sort a list of forty rivals and a poor thing to put in a slide.
5. 5Point a research attribute at one question across the whole set. The web research agent answers a prompt about each record and writes the answer back as a column, with its reasoning and sources attached. Ask one narrow, checkable question rather than a broad one: what does this company's pricing page show as its entry tier, not what is their strategy. Then open the sources on three records before trusting the other thirty-seven.
6. 6Chart the competitive set, then export what the charts cannot hold. Reports cover the counting. The export covers everything you want to keep or share, in CSV or Excel, one column per visible column in the view. Save the view's filters before exporting, and check the file rather than assuming: the enriched columns come out empty, which is the single most surprising thing about this system.
7. 7Give the set an owner and a review date. A competitor record decays silently, because nothing in the interface distinguishes an attribute refreshed yesterday from one written a year ago. Name one person, put a quarterly review in the calendar, and re-run the research attributes in a batch on that date rather than whenever somebody happens to notice a record looks stale.

The first decision is where the rival lives, and it is worth making deliberately because migrating later is tedious. Two standard objects are enabled by default, People and Companies, with Deals, Workspaces and Users available to switch on. Beyond those, custom objects are documented as available on the Pro and Enterprise plans only, and the total object allowance runs three on free, five on Plus, twelve on Pro and unlimited on Enterprise.

The reference attribute is what does the work

Whichever object you choose, the join is a record reference attribute on the deal, pointing at the competitor. Attio documents these as pointing to records of the same object or of a different one, usually constrained to a specific object, and describes the relationship as bidirectional: updating the attribute on either side automatically updates the other. A single-select reference holds one rival, a multi-select holds several.

That bidirectionality is the whole reason to prefer this over a picklist. Name a competitor on forty deals over two years and their record accumulates forty links without anybody maintaining a list, so the question “which deals did we contest against them” is answered by opening the record rather than by building a report. It also survives the things that break a text field: renaming the company, merging two records, correcting a spelling. Where the competitor is an option in a list instead, every one of those is a manual clean-up job, which is the recurring maintenance cost described under Pipedrive.

The attributes worth adding by hand

The point of having a record is that it can hold what a deal field never could, so put the competitive substance on the competitor rather than on the deal. Their entry price and the date you checked it. The segments where they beat you, in your own words. The single claim your sellers hear most often from buyers who talked to them. Their positioning sentence, copied verbatim from the page it appears on, with the date.

This is the part with almost no equivalent in a conventional CRM. Add a company and the workspace fills in a set of attributes on its own. Attio’s documentation lists them for companies as name, logo, description, categories, location, estimated ARR, employee range, funding raised, social profiles, foundation date, team and follower count, with estimated ARR, employee range and funding raised excluded on the free plan. The data comes from Attio’s providers and updates when they have something new.

For a competitive set that is a genuinely useful first pass. Forty rivals sorted by size and money raised, in the time it takes to type forty domains, without opening a single external database. The question is what each of those attributes actually is, because they are not all the same kind of statement and the interface presents them identically.

Observed, modelled, and the difference between them

Funding raised and foundation date trace back to things that were announced or filed, so they behave like most third-party firmographic data: broadly right, sometimes months behind, and occasionally attached to the wrong legal entity where a company trades under a name that is not the one on its registration. Employee range is a band derived from public profiles and similar signals, which is why it is a range rather than a number, and a band is exactly the right shape for it.

Estimated ARR is a different animal. It is a model output about a figure the company has not published, produced by a method that is not documented, so there is nothing to check and no way to reproduce it. That does not make it useless. It makes it a sorting key rather than a fact: fine for deciding which fifteen of forty rivals deserve attention, wrong for a slide. The same distinction between recorded and modelled fields runs through Crunchbase, and it is worth applying here with the same discipline, because a number that appears automatically in a CRM carries an air of authority that a number on a profile page does not.

What a sparse enrichment result is telling you

When a competitor comes back with almost nothing filled in, the instinct is to conclude they are small. Read it the other way first. Enrichment providers know most about companies that raise money, publish, and are written about, so a thin record is evidence about a rival’s publicity and fundraising habits rather than about their business. In mature markets the most dangerous competitor is frequently the profitable, quiet one that no data provider has any reason to describe.

AI attributes are columns whose values are generated rather than typed, and Attio documents four kinds: summarise a record from data already on it, classify a record into select values, complete a custom prompt, and a web agent that generates answers by researching external web sources in real time. The web agent is the one that matters here, because it is documented as designed for information that lives outside the workspace.

Configure it with a prompt and it writes an answer back as structured data you can filter and sort on. Where confidence indicators are enabled you can hover a value to see the agent’s reasoning and the sources it used, which is the part that makes it usable for research rather than merely impressive: an answer you cannot trace is not evidence, and here you can trace it.

Ask one narrow question of the whole set, not a broad one

The failure mode is predictable and expensive. A prompt like “summarise this company’s competitive strategy” produces forty paragraphs of plausible prose that nobody reads and nobody can check. A prompt like “what does this company’s pricing page show as its lowest paid tier, and on what date was the page last updated” produces a column, and a column can be sorted, compared and disproved.

• Narrow beats interesting. One fact with a source, asked of every rival at once, is the shape that makes this feature worth the credits. Breadth is what the record and your own reading are for.
• Verify three before trusting thirty-seven. Open the sources behind three answers you already know the truth of. That calibration takes ten minutes and tells you whether the column is a dataset or a decoration.
• Never let it answer a question the company has not published. Asking an agent for a private rival’s revenue produces a confident number assembled from estimates. It has the same status as any other model output and should be treated the way the enrichment section above treats estimated ARR.

The limit that shapes the whole workflow

Attio states that AI attribute values must be generated manually, either for individual records or in bulk, and that they do not run automatically, with recalculation triggered from the cell, a bulk selection or the column header. So this is a snapshot instrument, not a monitor. It answers a question on the day you ask it, and then the answer sits there looking exactly as current as it did that morning for as long as nobody re-runs it. Everything in the maintenance section further down exists because of that one sentence.

The nine graded rows above fall into three groups, which is how they are taken here: what holds the rival, what describes them, and what carries the result out.

The three structures that hold a rival

A Companies record is the route that works on every plan and inherits enrichment, at the cost of sharing an object with your customers. A custom Competitors object buys structural separation on the paid tiers and one of your object slots. A record reference attribute is the join itself and the reason either route beats a dropdown, since it maintains the link in both directions with nobody curating a list. Underneath all three, a list is where the competitive set you actively track lives, carrying list-specific attributes such as tier or owner that describe the rival’s place in your programme without touching the underlying record. What it takes to make any CRM produce a competitive dataset is worked through under CRM data.

The three ways a record acquires description

Enriched company attributes arrive on their own and are graded low here deliberately, not because they are bad but because their provenance cannot be inspected and one of them is a model output. The web research AI attribute answers a question you wrote, on the open web, with reasoning and sources attached, at ten credits per record. Classify and summarise attributes work from what is already in the workspace at one credit each, which makes them the cheap way to tag a competitive set by tier, segment or threat level once somebody has written the criteria down.

The two places a result becomes readable

Reports cover the counting, with five documented types of which insight, funnel and time in stage do the competitive work, and availability varying by plan. The view export takes any table or kanban view out as CSV or Excel, one file column per visible column, and carries the constraint that the next section is mostly about.

Export here is simple and there is one surprise in it, which is significant enough to design around rather than discover. Attio documents exporting any table or kanban view as CSV or Excel, with every column visible in the view becoming a column in the file, and advises saving filters before exporting so they are reflected in what you get. It is available on all plans; on Enterprise, admins can restrict exports to admins only.

That inverts the usual advice. On most systems the export is the complete version and the interface is the convenient one. Here the interface is the complete version, and the file is the subset you typed in yourself plus anything an AI attribute wrote back. Two practical consequences follow.

Type in what you intend to keep

If a figure has to travel, into a board pack, a spreadsheet model, a shared competitive review, it needs to be in an attribute you created rather than one that filled itself in. The workable habit is to copy the enriched value you actually rely on into your own attribute at the moment you rely on it, with the date. That doubles as the provenance record the enrichment does not give you, and it means the file matches what people saw on screen.

The sheet worth building

One row per tracked rival: name, tier, the segments they win in, their entry price with the date you checked it, the claim your sellers hear most, the number of open deals they are contesting, and the date the record was last reviewed. Seven columns, all of them yours, all of them exportable. Bring the enriched size band across as a typed value if the audience needs it. The competitive landscape template gives that set a shape somebody outside the revenue team will read.

Reporting, and where a platform connects

For the counting, stay inside the workspace. Insight reports give the current state of a metric, which is open deals per rival; time in stage measures how long records sit at each stage, which is where a competitor’s effect on your process appears before it appears in outcomes; funnel reports compare progression against your baseline. For the watching, none of it helps. A record only knows what somebody put on it, so teams that need the competitive set to correct itself pair the workspace with a competitive intelligence platform, which is a different kind of instrument rather than a better version of the same one.

The object allowance is the first constraint you meet

Objects run three on free, five on Plus, twelve on Pro and unlimited on Enterprise, and that count includes the standard objects you have switched on. Since People and Companies are enabled by default and Deals is the one most revenue teams turn on next, a free workspace has already spent its allowance before any competitive modelling begins. Custom objects are documented as Pro and Enterprise only, which is why the Companies-record route matters: it is not a compromise, it is the only route below Pro, and it works.

What the credits mean in practice

AI attributes are documented as available on all plans with workspace credit limits depending on the plan, and the web agent is priced at ten credits per record against one credit for summarising, classifying and prompt completion. The tenfold gap is what should shape how you use it. Running a broad web research prompt across two hundred companies is expensive and produces material nobody reads. Running one narrow, checkable question across the fifteen rivals that actually appear in your deals is cheap and produces a column somebody uses on Monday. Decide the question before you decide the list.

Reports and exports by plan

Report availability differs by type and tier: insight reports on every plan, with funnel and historical values on the paid ones, and the total number of reports capped per plan. Exports are documented on all plans, with all workspace members able to export from records pages and members with view access able to export from lists, and an Enterprise setting restricting exports to admins. Check what your plan includes before promising a dashboard, because the difference between charting the set and exporting it is a tier boundary rather than a preference.

What it costs in time

An hour to model the object and the reference attribute. An afternoon to add the fifteen rivals that matter with the attributes you care about. Then the recurring cost, which is the one nobody budgets: because nothing refreshes itself, somebody has to re-run the research attributes and re-check the typed facts on a schedule, and that is roughly half a day a quarter. A competitive set nobody owns is worse than no competitive set, because it looks maintained.

Two of the features that make this system good for competitive work also change the legal picture slightly, and both are worth thinking about once rather than discovering later. Data arrives on your records that you did not collect. And a prompt you write is sent, with context from a record, to a model that answers it.

On the first: enrichment means personal and company data reaches your workspace from third-party providers rather than from the person or company concerned. For company attributes that is ordinary business information and unremarkable. For people it is not, and the obligations that attach to holding personal data you did not obtain from the individual are worked through on competitor leadership team. For competitive work the rule is narrow and easy to hold to: the company is the subject, and its employees are not.

On the second: a research attribute is a question about a record, answered externally. Keep the prompts about published, company-level facts, which is what they are good at anyway, and do not write prompts that ask for information about named individuals. Nothing about the feature makes this dangerous by default; writing careless prompts at scale is what would.

The rest is the ordinary boundary that applies to any CRM, and it is worth restating in this system’s own terms because a competitor record here is more inviting to fill in than a picklist value ever was. A rival’s confidential document does not belong in an attribute, however it reached you. What a competitor said to you directly about where their pricing is going does not belong in your process at all, which settles the question of whether it belongs in an attribute. And every attribute you type should read as something you would be content to have quoted back, because a record designed to accumulate is a record that will still be there long after the deal that prompted it.

A well-built competitive set here looks more complete than one held in an ordinary CRM, which is precisely why the gaps are worth naming. Four of them, each with somewhere better to look.

• What they filed rather than announced. Enrichment reflects what has been published and written about. Documents filed under compulsion say different things, and often earlier, which is the subject of SEC filings.
• What their page said last quarter. A record holds the current value of an attribute and no history of it, so a repositioning that happened between two of your reviews leaves no trace at all. Dated captures of how a page looked before are covered under web archives.
• What is coming next. Nothing in a CRM sees a product before it ships, and an agent reading public pages sees the same things everyone else does on the same day. Ranked by how far ahead each one looks, the sources are under competitor roadmap.
• What is happening inside the company. A hiring freeze, a reorganisation or a wave of senior departures changes how a rival competes months before it shows in anything you can enrich, and employees describe it publicly. Glassdoor covers how far that testimony can be pushed and where it stops.

And the structural gap, which is the important one. Every record in your competitive set exists because a person decided to create it. Enrichment describes companies you have already named and the research agent answers questions you already thought to ask, so nothing in the workspace can surface a rival nobody has heard of. In a model this good at organising what you know, the absence of what you do not know is unusually easy to miss.

Because nothing here refreshes on its own, maintenance has to be a scheduled act rather than a habit, and it needs a named owner. That is a stronger requirement than a conventional CRM imposes, since there at least the deal data updates itself.

Quarterly, one person re-runs the research attributes across the tracked list in a single batch, re-checks the facts that were typed in by hand against their sources, and stamps the last-reviewed date. Doing it as a batch matters: refreshing records one at a time as somebody happens to notice them produces a set where a third of the values are current and nothing indicates which third.

Annually, review the set itself rather than the records. Which rivals have not appeared on a deal in a year and can be archived. Which names have been showing up in the reference attribute without ever being promoted to the tracked list. And whether the attributes you defined eighteen months ago are still the ones that decide deals, since a competitive set that was designed once and never re-scoped ends up answering last year’s question very well.

Off cycle, act immediately on three signals. A rival starts turning up in deals from a segment where they had never appeared. An enriched figure moves by a step rather than a fraction, since a change in the funding or headcount band usually reflects an event rather than a revision. Or a competitor is acquired, at which point the record and the company behind it have quietly become two different things and every attribute on it needs re-dating.

What ages here does so in a way worth stating precisely, because the system is otherwise so well suited to this work. Every attribute on a competitor record was written at a moment somebody chose. Enrichment updates when a provider has something new, which is not the same as when the rival changes something. Research attributes are documented as running only when a person runs them. Typed attributes change when a person retypes them. And a value produced eleven months ago renders in exactly the same colour, in exactly the same cell, as one produced this morning.

So the failure mode is not a record that looks stale. It is a record that stays confident: a competitive set the team consults weekly, quotes in deal reviews and gradually stops interrogating, describing prices changed in the spring and a positioning the rival abandoned before that. Nothing in the interface raises its hand. Which is why a workspace this good at organising competitive knowledge still needs competitive intelligence software beside it rather than instead of it. Flares tracks what a rival alters on the surfaces they publish and stamps the date, which gives your attributes something capable of contradicting them. Interpretation is still a person’s job. Whether a repricing resets your market or barely registers depends on which segment you sell into, and the deals sitting beside that competitor record are what answer it.