Sellio Assistant: chat, act and approve actions

The Sellio Assistant is the smart robot in the top-right corner (and inside every record). You chat in natural language and it finds, creates and updates data, configures the CRM and answers questions, always within your organization. Sensitive actions go through your confirmation or the operator's approval.

The Sellio Assistant is the product's agentic AI assistant. Click the robot in the top-right corner to open the chat on any screen; inside a record, the same assistant appears already aware of that record's context. You ask in English (or Portuguese/Spanish) and it acts using tools: list objects, read and create records, configure objects/fields/reports, and more. It only sees and changes data from YOUR organization.

How to use it

1. Open the robot (top-right corner) or the Assistant panel inside a record.
2. Type what you need, e.g. "list my open deals", "create a follow-up task for tomorrow", "move this deal to the Proposal stage".
3. The assistant replies and, when it creates or changes something simple and reversible, executes it right away and cites the ids.

Create and update in natural language

You do not need to know the technical value names. When creating or updating a record, just talk normally: say "Proposal" and the assistant stores the right stage; give a company NAME and it finds and links the related record; give a colleague's name or email to set the Owner and it resolves the user. If a name is ambiguous or does not exist, it keeps what you typed and tells you, instead of making something up. When something is missing (a required field, say), it tells you exactly which one, by the name you see on screen.

Need a reminder? Ask "create a follow-up for tomorrow" or "schedule a call for 3 days from now" and it creates the task with the due date already set, linked to the record when you are inside one.

Analytical questions and creation shortcuts

Beyond creating and updating, the assistant answers analytical questions about your book of business, always from your own numbers (it does not make data up):

• Deals at risk or stalled: ask "which deals are at risk?" and it lists the open opportunities with no recent activity, stagnant or past due, with the risk factors.
• Churn risk: ask "which customers might cancel?" and it shows the accounts with the highest risk and the signals driving the score.
• Explain the forecast: ask "why did the forecast drop?" and it compares the current period with the previous one (target, commit, best case, gap and coverage) and narrates the change.
• Performance by rep: ask "which reps need attention?" and it brings the win rate, wins and losses per person.
• Recommended next actions: ask "what should I do on my deals?" and it suggests the next step to unblock each opportunity (follow-up, log an activity, review discount or deadline, qualify).
• Buying committee: ask "who decides on this deal?" or "is anyone missing from the committee?" and it shows the key roles present or missing (decision maker, buyer, champion) and the decision-maker coverage per account.
• Stakeholder strength: ask "who are the strongest stakeholders?" or "who is the blocker on this deal?" and it scores each committee member (role, influence and sentiment) from 0 to 100, flags who is a champion and who is a blocker, and returns the ranking.
• Committee by area: ask "what does coverage look like by area?" and it groups stakeholders by department (the contact department), with the average score, champions and blockers for each area. A contact with no department shows up as "not set".
• Priority accounts (ABM): ask "which accounts should I focus on?" or "my priority accounts" and it ranks accounts by a 0 to 100 score and A/B/C tier, combining fit (qualification), open pipeline, engagement over the last 90 days, mapped committee and recency of the last activity. Each account comes with the why (high pipeline, recent engagement, no committee, no recent activity) so you know where to invest and what is missing. Filter by tier A to see only the hottest ones. It uses internal data only (no external intent signals).
• Ideal customer profile (ICP) and fit: ask "what is my ideal customer profile?" or "which accounts look most like the ones I already win?" and it LEARNS the ICP from your won customers (industry, size, tier and revenue band) and scores each account fit from 0 to 100 by adherence to that profile. It shows the derived profile, how many won accounts formed the basis, and what matched or is missing per account. All internal data.
• Account journey (ABM): ask "how is the journey for account X?" and it rolls up, for that account, how many contacts it has, how many are reachable by email, and the state of the sequences (active/completed/paused) and automations (ok/error runs) across the account contacts. It tells you if the account has not entered any campaign yet. You can also enroll every contact of the account (or just the buying committee) into a sequence at once, idempotently (anyone already active is skipped). All internal data.
• Data cleanup suggestions: ask "what do I need to clean up in my data?" or "which fields does nobody fill in?" and it analyzes contacts, companies, leads and opportunities and lists actions ranked by severity: duplicates to review, missing email/phone, low completeness and always-empty fields (candidates to hide or remove). All internal data.
• Process bottlenecks: ask "where do my deals get stuck?" or "which stage is the slowest?" and it points out the time per stage, the bottleneck and where the most drop-off happens.
• Improve the sales process: ask "how do I simplify my funnel?" or "do I have unnecessary stages?" and it analyzes the REAL process (mining the stage transitions) and recommends what to adjust: never-used or rubber-stamp stages (where deals pass through in seconds without adding value), the bottleneck, the stage with the most drop-off and whether the pipeline is too long. Internal facts only; needs enough transition history.
• Most efficient channels: ask "which channel brings the best deals?" or "where do my won sales come from?" and it ranks the opportunity sources (inbound, outbound, referral, event, partner) by win rate, revenue and average deal size, and points out the best channel. Internal data only (the opportunity source field).
• Compare periods and teams: ask "how did this month go vs last?" and it compares won deals, revenue and win rate of the current month against the previous one (change and direction); ask "who sold the most on the team?" and it compares reps against each other by revenue and win rate, against the average, pointing out the best and worst. Internal data only.
• Accounts at risk (CS): ask "which accounts are at risk?" or "where did tickets spike?" and it compares the recent window with the previous one per account and flags interaction drop (activities plunged) and ticket surge, with the percentages and reasons. Internal data only (does not cover product usage, which needs external telemetry).
• Pipeline health and quality: ask "is my pipeline healthy?" and it gives a 0 to 100 score with a band (healthy/watch/at risk) and the factors: coverage (weighted value vs goal), on-time (overdue opportunities), momentum (stalled deals) and stage distribution; plus a quality score from data completeness, recent activity and due dates. Internal data only.
• Revenue recommendations: ask "how do I grow my revenue?" or "what should I do to sell more?" and it synthesizes the internal signals (pipeline health, channel efficiency, at-risk accounts, month trend) into actions prioritized by severity, each with the metric that justifies it: increase coverage, clear overdue deals, re-engage stalled ones, scale the best channel, retain at-risk accounts, and more. Internal data only.
• Hot leads and lost leads: ask "show me this week's hot leads" and it lists the open leads with the highest heat (score + recency); ask "build a campaign for lost leads" and it lists the disqualified ones for you to re-engage (then assemble the segment and the campaign draft). Internal data only.
• Invalid data corrections: ask "which invalid data do I have to fix?" (name the object, e.g. contacts) and it shows the before/after preview of the values that would be normalized (whitespace, email, phone to international format) without writing anything: the review step before applying the bulk cleanup (which is a separate, confirmed action). Internal data only.
• Email and website domain audit: ask "do I have bad emails in my base?" and it flags emails with an invalid domain, disposable/temporary emails, free providers (a warning on a B2B account), role-based emails (info@, sales@, which are not a person) and malformed websites, with the count per type. It is a heuristic over known lists, without checking deliverability on the network; internal data only.
• Territory strategy: ask "which territories are behind quota?" or "how is coverage by territory?" and it consolidates open pipeline and won revenue from each territory's reps, computes coverage (pipeline ÷ quota) and attainment (won ÷ quota) and classifies (on track / behind / at risk / no quota), highlighting where pipeline is missing. Internal data only.
• Legal and finance committee: the buying committee map now also identifies Legal/compliance and Finance stakeholders. Ask "who is legal/finance on this deal?" and committee coverage flags whether those roles are present.
• Suggest the next stage: ask "can I advance this opportunity?" and it moves the deal forward to the furthest stage whose cumulative required fields are already filled, and points out what is missing for the next one. Deterministic, no AI cost.
• Draft a follow-up (subject and body) for a contact or opportunity, with the tone adjusted to how long it has been since the last touch · no AI cost; you review before sending.
• MQL→SQL handoff: ask "how is the marketing to sales handoff going?" and it shows the lead funnel by stage (new, nurturing, MQL, SQL, converted, disqualified), grouped by source, with MQL→SQL and SQL→won rates, revealing where the handoff leaks. Internal data only.
• Retention playbooks: ask "what should I do to retain at-risk accounts?" and it suggests actions per at-risk account (schedule check-in, executive sponsor touch, value recap, support review, success plan review) based on the risk reason, prioritized. Internal data only.
• QBR agenda: ask "build the QBR agenda for this account" and it synthesizes the account's health, adoption, risks, tickets, expansion and renewal into a prioritized quarterly business review agenda. Deterministic, no AI.
• SLA breached: ask "which tickets breached SLA?" and it lists open tickets whose first response or resolution is past the SLA deadline and still unresolved, from most to least overdue. Internal data only.
• Suggested tasks: ask "which deals are going cold?" or "where do I need to act?" and it flags open opportunities and leads with no follow-up (no upcoming task and no recent activity) so no deal goes cold. Suggests only, creates nothing.
• A/B test improvement suggestion: when you look at the per-variant metrics of a message step, the system suggests the next move (apply the winner when there is a clear gap, keep collecting when the sample is small, try a bolder change when they are tied, or revise the subject when open rate is low across all variants).
• Create an email template by command: say "save this email as a template" or "create a follow-up template with subject X and body Y" and it creates a reusable template in your library (name, subject and body). As a write action, it goes through your confirmation.
• Create a dashboard chart by command: say "add a bar chart of opportunities by stage to my dashboard" or "create a KPI of total won deals" and it creates and saves the chart on your dashboard (object, type, measure and grouping). Write action, goes through your confirmation.
• LTV by segment: ask "which segments are worth the most?" or "which type of customer has the worst churn?" and it brings the ranking by lifetime value and churn risk by industry, size or tier.
• Why we lose: ask "why are we losing, on price or to a competitor?" or "which products do we lose most?" and it groups the lost opportunities by competitor, price range or product, with the count, the lost value and each one's share.
• Conversion by source, campaign or ICP: ask "which source converts best?", "which campaign brings the most wins?" or "conversion by ICP" and it brings the win rate per key (total, won, lost and open). Keys with few closed deals are flagged as low sample, so you do not draw conclusions from little data.
• Best channel and time to reach out: ask "what is the best channel to reach this customer?" or "what is the best time to follow up with this lead?" and it analyzes the real touch history (email, call, WhatsApp, chat) to tell you which channel and which window (time of day and day of week) got the most replies. If the record has little history it uses your organization average and says so; if there is not enough data it says it cannot recommend rather than making it up.
• Relationship strength: ask "how strong is my relationship with this contact?" or "which relationships have gone cold?" and it gives a 0 to 100 score and a band (strong, warm or cold) from the real interaction history, combining three signals: recency (how recent the last conversation was), frequency (how many interactions in the last 90 days) and reciprocity (how much the contact also replied or reached out to you). Pass an account to get the rollup of its contacts (average and the strongest tie). With no recorded interactions it says cold rather than inventing one.
• Most common objections: ask "what are the most common objections?" or "what stalls my deals the most?" and it aggregates the objections already identified in conversations (in each contact's insights), counts them and ranks them from most to least frequent, with each one's share. You can filter by period. Honest note: it only sums objections already analyzed in conversations; if you have not generated conversation insights yet, the list comes back empty rather than inventing objections.
• Trend, seasonality and sales velocity: ask "are sales growing?", "what is the revenue trend?", "is there seasonality?" or "what is my pipeline velocity?" and it computes the trend of won revenue and of opportunity volume (up, down or flat), points out the peak and trough months, and measures sales velocity (how much revenue the pipeline generates per day, from the number of open opportunities, the average deal size, the win rate and the average cycle length). With little history it warns that the sample is insufficient, and seasonality is only reliable with about two years of data, rather than inventing a pattern.
• "What if…?" simulations (what-if): ask "what if we increase the conversion rate by 10%?", "what is the impact of hiring 2 reps?", "what if we raise the price by 5%?" or "what if the proposal stage conversion goes up to 60%?" and it projects the scenario from your current numbers, without inventing the baseline. For forecast it uses the pipeline velocity formula (number of opportunities, average deal size, win rate and cycle); for hiring it computes the added revenue capacity with an optional ramp; for pricing it shows the new revenue and, if you do not provide the expected volume change, it warns that it assumed constant volume (no elasticity); for a stage it recalculates end-to-end conversion and total cycle. It is only a simulation: nothing in the system is changed.

And to build structure fast, ask in one sentence: "create a pipeline called Enterprise with the stages Discovery, Proposal and Closing" builds the pipeline with the stages in order; "create a follow-up cadence: call on day 0, email on day 2, meeting on day 5" creates the activity sequence with the right waits.

Confirming sensitive actions

Sensitive actions are NOT executed immediately. When the assistant proposes something reversible and scoped to your organization (e.g. emailing a segment, deleting some records), an "Action awaiting your confirmation" card appears with the risk and the reason. Click Confirm to execute, or Cancel to dismiss it (the proposal expires on its own after 30 minutes).

Operator approval (high risk)

High-risk actions (involving money, billing, security/RBAC or bulk deletion) go to the platform operator for APPROVAL, not to you. The assistant tells you the request is pending. If approved, the action runs with YOUR permissions (never with elevated privilege).

When your request needs development

If you ask for something the assistant cannot do with its action/no-code tools (it needs new code or a new feature), it does NOT write or ship code. Instead, it registers your request for our team to review and explains what happens next. If the idea would help all customers, it enters the improvement queue for prioritization; if it is specific to your organization, it offers you a dedicated instance and captures your interest.

Track the outcome under "Talk to the team" (Support menu), in the "My requests to the developer" section: each request shows its current status (open, in backlog, dedicated offered, declined or implemented), the decision date and the team notes. You also get a bell notification on every update.

Monthly usage and cap

At the top of the chat there is an indicator of this month's usage vs. your account's monthly cap. When the cap is reached, the assistant stops replying until next month or until you adjust the cap in settings. Usage is metered fairly and appears on your invoice.

Open this article inside the system →