How Klosure works, in full.

01

What is Klosure?

Klosure is the deal execution engine for B2B sales. Klo — its AI — reads every email, meeting and silence on a deal, tells the rep the next move, drafts the buyer-facing question for the rep to approve, and evaluates each deal with a VP-grade verdict. The forecast comes out true because the work actually happened.

Klosure is not a CRM, and it is not a note-taker. It sits on the signals a deal already generates — inbox, calendar, meeting transcripts, deal chat — and turns them into three things a sales organisation can act on: a deal record nobody types into, a confidence score that can be defended in a board review, and a specific next move for the rep.

02

What problem does Klosure solve?

Every forecast review in B2B runs on self-reported optimism. Two records of the same deal exist at the same time — the one the rep typed, and the one the buyer's behaviour actually describes.

The CRM is a paper map. Klosure is the live route. Three consequences follow, and they define the entire product:

1. No manual logging. If a signal exists in an inbox, a calendar or a meeting, Klo has already read it. Adoption isn't enforced, because there is nothing to enforce.
2. The forecast is defensible. Every confidence factor carries the conversation evidence that grounds it — a quote, the message it came from, and the date — so a manager can defend the number rather than relay it.
3. Coaching is specific or it is nothing. Klo's own instructions state that a reply which could apply to any deal at that stage is a failure to be rewritten.

03

How does Klosure work? Four layers, one engine

Klosure runs four layers on every deal. Reading, coaching, forecasting, and the buyer room. They share one deal state, so nothing is computed twice and no two surfaces can disagree.

04

What does Klo actually read?

Klosure connects to the tools a sales team already uses and reads them continuously. There is no activity logging, no CRM overlay to maintain, and no browser extension to install.

How does Klosure decide which deal an email belongs to?

An email or meeting is attached to a deal if and only if one of that deal's known buyer-side contacts appears on it by exact email match. The routable set for a deal is the union of its champions, its buyer-side stakeholders, the buyer-side people Klo has extracted, and the deal's primary buyer address. Champions are matched first, across every deal, before mere contacts.

There is deliberately no company-domain fallback. "Same company" is not "same deal" — a different contact at the same company working a different project must never cross-contaminate. Matching stays exact-email only.

What happens when a webhook is dropped?

Every real-time path has an idempotent reconciliation backstop, because webhook delivery is best-effort everywhere:

• Dropped mail is backfilled by a sweep that lists the mailbox's recent messages and inserts only what is missing.
• Emails that arrived before their deal existed are re-matched. An email logged as "no deal match" is re-evaluated against current deals, so creating the deal later still pulls its history in.
• Large mailboxes drain safely. The Microsoft sync claims a per-mailbox lock, checkpoints its cursor after every page, stops at a wall-clock budget, and re-invokes itself to continue — so a mid-run interruption loses at most one page.
• Finished meeting transcripts are polled directly, so a misconfigured webhook can't silently swallow a call.
• Missed meetings are re-scanned. If a meeting synced before its deal had a champion, locking that champion later pulls Klo into the meeting retroactively.

05

The deal record nobody types into

Every deal carries a single structured state document that Klo rewrites whenever something material happens. This is Klosure's equivalent of a CRM record — except no human fills it in.

Is the deal record auditable?

Yes. Every field mutation is written to a history table carrying the message that triggered it, the role that triggered it, the kind of change (extracted, removed, or corrected), the before and after values, and a reason. A manager can trace any statement about a deal back to the message that produced it.

What if Klo gets something wrong?

The rep removes it and gives a reason. That removal is recorded permanently, and Klo is instructed to read the removals list on every subsequent turn. A user never has to delete the same wrong extraction twice.

06

How does Klosure's confidence engine work?

Confidence is a 0–100 read of how likely a deal is to close by its deadline. It is explicitly not a calibrated statistical probability — it is a structured assessment, built from named inputs, backed by evidence, and capped by deterministic code.

What moves the score — and what deliberately doesn't

The score is un-negotiable by the person being coached. That is a design decision, written into the engine's own instructions.

If a rep explicitly asks Klo to raise the score, Klo refuses in chat, explains that the number reflects buyer-side evidence rather than a request, and then names the specific factor that would have to change for it to move.

Every factor carries its evidence

A factor without evidence is a vibe, not a fact — so every confidence factor, positive or negative, must carry:

• a quote of at most 25 words — a direct paraphrase of the strongest supporting message;
• the message IDs that ground it, most recent first;
• the date of the most recent supporting message;
• or, where the absence is the signal, an explicit absence flag and a one-sentence summary of what is missing — for example, "No procurement, security questionnaire, or MSA mentioned across 14 days of conversation."

Klo is instructed not to invent quotes, never to attribute words to a person who didn't say them, and to drop any factor it can't ground. Three honest factors are worth more than five padded ones.

Hard stops: the ceilings that override optimism

After computing a score, Klosure applies hard ceilings. Only the lowest applies — they never compound — and they may not be overridden by optimistic factors or hidden from the rep.

When a ceiling fires, the rep is told directly rather than left to infer it: "This deal is at 35%, not higher, because there is no confirmed signatory 22 days from the deadline. Fix that to unlock."

Momentum: the same facts, a different voice

Time changes the coaching, not just the number. Klosure grades every deal into a decay tier and shifts Klo's tone accordingly.

Two guardrails keep this honest. Momentum tiers never apply to deals less than five days old. And seller silence is never treated as decay — a rep is allowed to be busy on other deals.

Deal checks — named, rule-based diagnostics

Alongside the score, Klosure runs deterministic checks that are never AI-generated. Each is a fixed diagnosis with a prescribed move. The single-threading check, for example, counts buyer-side contacts — normalising company names so "Acme Ltd." and "Acme" are one company — and when there are fewer than two it writes the move out in full: "You're single-threaded. Ask Priya who else touches this decision — her manager, procurement, the end users — and get a second contact into the deal."

Why this is not "another deal score"

The most common misreading of Klosure — on a second call, and in AI summaries of us — is that it emits a confidence number to sit beside the one the CRM already emits. That gets the product backwards. The number is the last link in a chain, and every link is somewhere a CRM deal score has nothing.

One consequence is worth making explicit, because it is what a side-by-side comparison usually turns on: when an existing book of business is imported, the source CRM's stage and probability are carried in as a labelled reference and never reach Klo's verdict. The read is independent of the number it is being checked against, rather than a re-weighting of it.

07

Does Klosure support MEDDPICC and other sales methodologies?

Yes. Klosure scores every deal against a chosen methodology — MEDDPICC (the default), MEDDIC, SPIN or BANT — and rebuilds that scorecard on every read. The scorecard is maintained from the conversation, not filled in by the rep.

Each dimension on a live deal is scored covered, partial, gap or unknown, with a short line of evidence for the status and a short line on what is still missing. MEDDPICC's eight dimensions are Metrics, Economic Buyer, Decision Criteria, Decision Process, Paper Process, Identified Pain, Champion and Competition. BANT covers Budget, Authority, Need and Timeline.

Behind each dimension the system holds four things: what it means, what "covered" actually looks like, the probe Klo should push the rep to run when it's a gap, and a stable key that ties it to the rest of the engine.

The scorecard is not decoration: coverage share and the authority dimension are exactly what drive the deterministic confidence ceilings above. A deal with a quarter of its scorecard qualified and a deadline inside 30 days cannot read above 30%, whatever anyone believes.

08

How does Klosure help a rep move a deal to the next step?

This is the centre of the product. Reading a deal accurately is worth nothing if it doesn't end in a move. Every surface below exists to convert "what is happening" into "what I do next" — and most of them end in something the rep can send, book or decide today.

The day starts with a decision, not a list

• Today's focus. Klo reads the rep's entire active pipeline in a single pass and writes one paragraph on where the day should go. It refreshes automatically when the pipeline shifts meaningfully — a ten-point confidence swing, a status change, or a new deal — rather than on a timer.
• Today's priorities. The Focus tab always answers "what do I do next" with an opinion and a call to action, not a neutral list to interpret.
• This week's commitment. The one thing the rep committed to, carried across surfaces so it doesn't evaporate by Wednesday.
• Dormant suggestions. Deals quiet for 60+ days are surfaced for a deliberate park-or-push decision, instead of quietly inflating the forecast.

Inside a deal: Deal Pulse

Every deal opens to Deal Pulse, with four sub-tabs.

Proven Plays — the Diagnose → Decide → Do loop

Proven Plays is the engine that turns a stuck deal into an executed move. It is a stateful loop, so nothing is lost when the rep navigates away and a decaying deal can always land on a terminal move instead of being re-diagnosed forever.

A diagnose turn reads the live deal state and returns coaching plus two or three typed next moves the rep can tap. Tapping one executes it:

Two invariants make this safe to give a rep: the model proposes, deterministic code executes every state change through an authorisation-checked path, and every state change is reversible, with Undo offered as a first-class move.

Klo drafts the email — and the bar it has to clear

Reps used to leave the deal room to write the email Klo had just coached them on, losing all the context in the switch. Now they describe the email they want and Klo drafts the subject and body from the deal state, the stakeholders, what needs to happen, and the recent conversation. Refining threads on the previous draft rather than starting over.

Every buyer-facing draft must clear four bars:

1. Write for the forward. Assume the champion forwards it unedited to someone who was not in the room — their CEO, their CFO, a peer who has to sign off. Give the champion one line they can forward that makes them look sharp for surfacing it.
2. Mirror the buyer's own words. Reuse the exact terms the buyer used — the system they named, the metric they care about, the number they quoted. Sharpen the frame; never swap their vocabulary for synonyms that read like marketing copy.
3. Introduce no new concern. Delete any sentence that raises a worry, a stakeholder or a risk the buyer never raised. Inventing "you'll want your CEO aligned early" hands the buyer a new problem to manage. That's friction, not value.
4. Respect the buying sequence. Push for the smallest real next commitment the buyer can make from where they stand — not the biggest one the rep wants.

Talking a deal through with Klo

The in-deal chat is not a record-keeper. Klo is three experts in one — a sales coach, a Gulf/India B2B domain expert, and a deal-desk reviewer — and how it engages is governed by explicit rules.

Reply length follows substance

A "thanks", a logistics-only email, or an FYI with nothing to act on gets no reply at all — Klo is forbidden from manufacturing coaching nobody asked for. A routine update gets a sentence or two. A material event — an objection, a named competitor, a slipped date, a stakeholder change — gets a paragraph explaining what it means and what to do. And a rep working a real problem gets as much depth as the question needs, with no upper limit.

Diagnose before tactic

When a rep asks for a tactic and the deal is actually decaying — momentum tier 2 or worse, an active hard stop, or confidence below 50 — Klo is forbidden from handing back a polite tactic. It has to diagnose first and make the rep see they're asking the wrong question:

"Before tactic — three flags on this deal. Demo was 20 days ago. Only Abdul has engaged. No forcing function named. The Eid week isn't your problem; the single-threading is. Pushing harder won't help. The real question for Monday isn't 'when can we sync' — it's 'who else at Vinsys feels this is urgent?' If Abdul can't name someone, this is a nurture, not a Q2 deal. Want me to draft a message that surfaces that, or one that just keeps the conversation warm?"

Advance or expose

Every email, message or meeting Klo advises on must do one of two things: advance the deal — lock a commitment, a date, a meeting, a signature, surface a stakeholder — or expose its fiction: test whether urgency is real, force a yes or no, ask the question whose answer reveals whether the deal exists at all. A "just checking in" touch on a decaying deal does neither, so Klo refuses to suggest it.

Rehearsal, wins, and venting

Ask Klo to play the CFO and it will — in character, pushing back on the per-seat maths and asking why not the incumbent — then break character to debrief what landed and what to tighten. Report a win and it names it briefly, then surfaces the next risk in the same breath: "Hashim's signature is the unlock. Next: who countersigns on legal — get that name before Friday." Vent without asking a question and Klo won't produce a tactic; it reflects the situation, then asks the diagnostic question the rep isn't asking themselves.

The banned phrases

Klo is explicitly forbidden from writing "Great question!", "It depends…", "Have you considered…", "pivot to your value proposition", "highlight your differentiators", or "leverage your unique strengths". And the governing rule: every non-empty reply must reference at least one specific detail from this deal — a stakeholder by name, a specific number, a named competitor or blocker. If the reply could apply to any deal at that stage, it's a failure and gets rewritten.

Scripts, not principles

Klosure keeps a searchable library of sales moves. On a given turn it retrieves the tactics closest to what the rep just said, filtered by deal stage and buyer signal — and then lifts the script. Reps want a copy-pasteable line, not a principle to figure out, so Klo adapts the move's dialogue to the deal and puts the adapted line in quotes:

Not "tell her you can drop a feature instead of price" — instead: Say to Reem: "I can get to $25K — at that level we'd drop the cultural-specifics module. Does that fit?"

One move per reply, two at most. Klo never names the move or cites the library — the rep wants the play, not a lesson. The shape is always the same two beats: one sentence of strategy, then the literal script.

Deal realism by region

Klosure encodes how deals actually behave in the markets its users sell into, rather than assuming a US calendar.

• Working weeks. Friday–Saturday in Saudi Arabia, Bahrain, Kuwait and Oman. Saturday–Sunday in the UAE and Qatar (the UAE moved in 2022), with Friday afternoon a half-day. Friday is dead for buyer outreach in Friday–Saturday markets.
• Holidays are treated as silence, not decay. Ramadan shrinks working hours by roughly a quarter and slows procurement. Eid al-Fitr and Eid al-Adha freeze the Gulf for 7–10 days. Diwali, US Thanksgiving, Christmas–New Year and the European August all do the same. When buyer silence overlaps a major local holiday, Klosure downgrades the decay tier and tells the rep why — then resumes normal rules the day after.
• Procurement floors. Saudi enterprise and government: 60–120 days minimum to signature. UAE enterprise: 30–60, government 60–120. India SMB: 2–4 weeks; India enterprise: 30–60 days. US mid-market SaaS: 2–4 weeks; US enterprise: 30–90. Europe adds 2–4 weeks for data-protection review on anything touching data. When a deal's deadline is tighter than its market's floor, Klo names the gap and pushes the rep to escalate, get an exception, or reset the date.
• Cross-border friction the rep may not have planned for: local-entity invoicing, withholding tax, VAT/GST registration, data-residency clauses, payment rails. A common stall Klo raises early — procurement insisting on a local entity to invoice from.
• Channel norms. Gulf B2B runs on WhatsApp for the live conversation and email for the formal trail. India SMB runs on email, LinkedIn and WhatsApp. US B2B runs on email, LinkedIn and shared rooms, rarely WhatsApp.

The voice guardrail is as important as the knowledge: state the specific implication — "Eid week starts Apr 9 — get the proposal in by Apr 5 or this slips to May" — never lecture a rep about a market they work in every day.

Before the call: Discovery Prep

Before a discovery call, Klo hands the rep a tailored brief: an opener tuned to how the lead arrived, the questions that matter for this buyer (built from the methodology's own probes), what a green-flag versus a red-flag answer sounds like, the two or three walk-away disqualifiers, and who still needs to be in the room. It is grounded in the rep's own profile — what they sell, their ICP, their personas, their most common deal-killer.

Qualifying: go, park, or keep probing

On an opportunity, Klo renders a verdict on top of the automatic scorecard: qualify, nurture or keep probing, the single risk that kills it, a suggested value band, and the one move — plus a manager-only read on how the rep is qualifying.

The opportunity workspace is stateful rather than a flat checklist: it leads with where the lead actually stands — new → prepped → no-show or call logged → assessed → nurturing or disqualified — and the single next move, collapsing completed steps. Once a call has genuinely happened, the assessment runs itself.

Arming the champion: the business case

Klosure's business-case generator is built on a piece of B2B reality most tools ignore: the buyer is passive. They won't do homework, hand over numbers, or articulate "why now". The rep pulls the deal. So Klo arms the rep — it proposes the urgency angle, builds a defensible ROI model, and suggests who likely signs and a path to them — instead of listing things to extract from a disengaged buyer.

Nine components: executive summary, the problem, the cost of inaction, the proposed change, why now, ROI in three scenarios, total cost of ownership, payback, risks, the ask, and the rep's own actions.

Numbers integrity is enforced. Figures the buyer actually stated, the deal value, and recorded decisions are cited as real. Everything else is a clearly labelled assumption to validate — never dressed up as the buyer's confirmed number. Klo can also run a short set of optional prep questions first — the high-leverage gaps a CFO will probe — and anything the rep answers is then treated as confirmed fact rather than assumption.

Getting through procurement

Three capabilities cover the work that stalls enterprise and IT-services deals after the buyer has already said yes.

• Security questionnaires. Klo drafts answers to an inbound security or vendor questionnaire from the team's own prior responses. Every answer a rep promotes back into the corpus compounds it, so Klo answers more of each new questionnaire in the team's own words over time. Grounding is strict: answers come only from the corpus, and any question the corpus doesn't cover is flagged as needing input, never fabricated. Klosure will not invent a security claim or a certification.
• Contract review. A first-pass commercial review of an inbound MSA, order form or DPA. Klo flags clauses that hit a team red line or diverge from the team's preferred position and proposes a redline. Well-known deal-killers — uncapped liability, broad IP assignment, one-sided termination, unlimited indemnity — are flagged even without a playbook entry, but never invented: only clauses actually present in the document are flagged. The raw contract text is not stored; only clause summaries and proposed redlines are kept.
• Procurement tracker. The steps between "verbal yes" and "signed" tracked in one place on the deal.

Competitive defence and commercial levers

Managers import the team's battle cards — how we win, their landmines, proof points per competitor — and the deal desk's commercial levers, such as a paid pilot, a phased SOW or milestone billing, each with when to use it and its guardrails. Reps then consume both through Proven Plays, where Klo fuses the team's own moat with the tactic library, under a no-invented-discount guardrail.

The approval queue

Everything buyer-facing that Klo authors waits in one place for a human. The rep approves, edits, or dismisses. Approval sends it; dismissal sends nothing. Only the deal's owner or a manager who manages them can decide.

Meetings

Klo joins meetings as a visible notetaker across Zoom, Google Meet and Microsoft Teams. Dispatch is automatic, and the rep can always override it: "Ask Klo to join" forces Klo into a meeting the matcher skipped, and "Remove Klo" cancels the bot and marks the meeting opted out so nothing re-adds it later.

When the transcript is ready it is posted into the deal and the deal is re-read — so a meeting updates confidence, stakeholders, blockers and next actions with nobody typing a note. Klosure then distils the call into call signals: talk ratio, discovery questions asked, whether a next step was secured, whether a multi-thread ask was made, and whether the rep actually practised the skill they're currently working on. Meeting minutes are metered from the real recorded length, not the calendar slot.

After the deal closes

• Deal retrospective. When a deal closes — won or lost — Klo writes a structured brief on that deal: days to close, days stuck, missed commitments, the stage path, what went right and what almost killed it with evidence, the turning points and whether each moved the deal forward, backward or stalled it, and a numbered list of what to repeat next time. It runs automatically, and again on demand if the rep wants to debrief with Klo.
• Win/loss interview draft. Klo drafts a short, buyer-safe interview email and queues it for approval. Buyer-safety is structural, not merely instructed: the drafting step is fed only buyer-safe inputs and never sees the seller-side retrospective, confidence or blockers, so none of that can leak.
• Patterns across closes. Retrospectives are read back into the weekly reflection, which names the pattern that repeats rather than recounting any one deal: "Three of your last five losses were no-decision, and in each the economic buyer never joined a call." That is the finding. "You lost Acme" is not.

Getting better: Reflect, scorecards and Growth Focus

Every week Klo writes a 30-day rolling reflection — private to the rep — showing what's working, where time is leaking, and the one deal that needs them, ending by holding them to a single commitment for the week ahead. Managers get a team-level reflection built from raw data; a manager's reflection never reads another person's reflection.

Alongside it sits a measured weekly scorecard across six dimensions, each 0–100 with an evidence-cited note and a week-over-week delta:

The manager sees the scorecard and coaches on it, without ever reading the rep's private reflection narrative.

09

What do sales managers get?

Managers get their own workspace built on the same deal state the reps see — so a 1:1 starts from evidence instead of anecdote, and no two screens can disagree about a number.

The cockpit — a five-minute screen

1. A Klo headline — a four-sentence VP soundbite on the week.
2. The number bar — the call, percentage to target, the delta, days left, coverage.
3. The motion strip — Won, Lost, Added, Slipped, Dark, week over week.
4. Patterns — the top three cross-deal patterns that matter.
5. Rep spotlight — rep of the week, and who needs help.
6. A decisions queue — persistent and checkable, so decisions don't evaporate.

Motion is computed, not narrated. Against last Sunday's snapshot: won and lost are real status transitions in the last seven days; added is created in the last seven and absent from the prior snapshot; slipped means the deal was active in both snapshots and confidence dropped by ten points or more; dark means active with no buyer message in five or more days. In week one there is no prior snapshot, so the interface subdues the "vs last week" comparison rather than inventing one.

Patterns are detected by code, not imagined: deals stuck in a stage past a threshold, pipeline generation down week over week, reps slipping confidence across multiple deals, buyer activity dropping across a set of deals, one deal carrying an outsized share of the open call, and deals past their close date. The AI may rewrite the wording in VP voice — never the label, the deals, the severity or the kind.

Weekly Review — the Monday meeting, run properly

A shared-screen ritual with two modes. The Overview is the team opener — the number, what moved, what to celebrate. By Rep is the round-robin: click through a rail of reps, each opening that rep's session — last week's performance, their number, the deals to inspect, and this week's commitments captured live.

Klo prepares each rep in advance: two or three proposed commitments (optionally tied to a specific deal), one probing question the manager should ask, and a private coach note — the candid read on sandbagging, where they're stuck, whether they need a 1:1 — which never appears on the shared screen.

Staffing: who should own this deal?

The manager describes a deal or an incoming lead, and Klo recommends an owner — reasoning over a per-rep scorecard built from every deal on the team: current load, win/loss record, deal-size fit, cycle time, and profile. Klo recommends; it never assigns. The manager makes the call.

For a batch, the manager uploads a CSV of fresh leads and Klo recommends an owner for every row in one pass. The critical difference from a row-by-row scorer is that it distributes: scoring each lead independently against today's snapshot would pile every lead onto whoever happens to be lightest right now, so the batch is reasoned about as a whole.

Ask Klo, scoped to who's asking

Company structure

Klosure models Company → Teams → Reps, not a flat list. Roles are company admin, sales head, manager and rep; multiple managers per team are supported and honoured by the authorisation layer, and deals carry their team and organisation for attribution. A sales head without their own team gets the org-wide view and a cross-team roll-up.

10

What does the forecast look like?

Board-ready and deal-deep — and computed by a single engine shared by every surface, so the Monday brief, the weekly review, the cockpit and the forecast tab can never disagree.

The forecast shows Closed, Commit, Best case, the weighted call and full pipeline against a target you set — every scenario auditable down to the individual deal and the buyer signal underneath its score. A deal counts in-period if it has a real deadline falling before the period ends; overdue deals still count, and deals with no deadline are excluded rather than quietly assumed.

The risk radar

Five groups, each with the dollars exposed before the period ends:

Pipeline itself has four lenses — list, by rep, by stage, and Build, which shows where new business is coming from by source and by rep.

11

What does the buyer see?

The buyer opens a per-recipient magic link — no signup, no password — into a room co-branded to the seller. It is the only part of Klosure a buyer ever touches, and it is deliberately not a login.

Inside: a plain-language read of where the deal stands, the mutual action plan with owners and due dates, the business case, shared documents, a momentum indicator, a Q&A section, and an Ask box. Anything the buyer writes goes straight to the rep — never to a bot. The buyer side is always free and never pays.

Why each buyer gets their own link

Each recipient gets a unique 256-bit token, of which Klosure stores only a hash. The raw token travels only in that person's email, inside the URL fragment — so it is never sent to a server, never logged, and never leaks through a referrer header. It is never returned to the rep either.

That last point is a product decision, not a technicality: the rep must not hold the buyer's link, because that is exactly what makes the buyer's later actions an independent signal. When a room is opened, forwarded, or answered, the seller learns something real about the buying committee rather than something the rep could have produced themselves.

Corporate link scanners don't fake engagement

Enterprise mail security pre-fetches every link. Left alone, that would fabricate "room opened" events and corrupt the committee-presence signal entirely. Klosure only mints a session on an explicit human "Enter" click, and rejects known scanner agents as a backstop. The pre-click landing screen is served by a separate endpoint that returns only branding and the seller's company name — no token exchange, no session, no logged event, nothing sensitive.

The buyer cannot see the seller's strategy

The buyer-facing response is assembled field by field from an allowlist. It contains only the room title, the seller's name, the content sections, and a single live question for that contact. It can never contain the deal state, the confidence score, the confidence factors, blockers, deal checks, the seller's identity, token hashes, or any other contact. The buyer's momentum indicator is a different number on a different scale from the seller's confidence, by design.

Klo's questions to the buyer are constrained and rep-approved

Klo proposes one to three short questions to ask the buyer in the room. The question type is a closed set — logistical, value unlock, or objection surface — which makes a "are you still interested?" purchase-intent question structurally impossible to generate rather than merely discouraged.

By default those questions are drafts. The rep approves or dismisses each one before the buyer ever sees it, and the rep previews the buyer's entire view before sharing the room. Auto-send exists, but it is a per-deal setting the rep switches on themselves and can reverse at any time.

When the buyer answers, that answer is recorded as fact shown to the seller — it is display-only and never feeds the confidence engine. Access can be revoked instantly: every buyer request re-checks the contact against the database, so revocation and window expiry take effect immediately.

12

What does Klosure connect to?

Does Klosure replace my CRM?

No. Klosure runs alongside the CRM. The push is deliberately one-directional and idempotent: the first push creates the CRM opportunity and remembers its ID; every later push updates that same record, so a deal is never duplicated. Klosure's data is generated by the deal itself, so it stays the source of truth — the CRM is where the deal gets reported. Lead intake runs the other way: leads land in Klosure the instant they're created in the CRM, exactly once, even across overlapping sync windows.

How do I get my existing pipeline in?

A staged import: parse, validate, stage, then create explicitly row by row — so unvalidated data never lands in the system, and a created row can never be imported twice. All spreadsheet content is treated strictly as data, never instructions, and any source-CRM stage or probability is carried in as a labelled reference only. It is never fed into Klo's verdict, because importing another system's optimism would defeat the point.

13

What runs automatically?

Klosure's scheduled work is timezone-aware, idempotent and self-healing — and there is a watchdog whose entire job is to notice when it isn't.

The Monday cascade

A company's Monday morning runs in strict dependency order, so no brief is ever built on stale data:

Phase A — UPDATE (bottom-up)
  1. Deals       re-read every alive deal across every team
  2. Rep views   each rep's reflection regenerated in the same pass
  3. Manager     each team's cockpit and weekly review
  4. Sales head  the company roll-up generated LAST, on fresh teams

Phase B — EMAIL (only after Phase A completes)
  5. Rep         personal weekly brief
  6. Manager     Monday brief
  7. Sales head  company roll-up

It fires on the company's own clock, so a sales head over multi-timezone teams is briefed exactly once — after every team underneath them is fresh. Per-team deal re-reads run about two hours before the emails go out, precisely so the briefs summarise current state rather than last week's.

The watchdog

A cascade like this can fail silently — an expired secret, a timeout, or a logic error all leave no visible trace, and the only symptom is that Monday's emails never arrive. So a daily, side-effect-free watchdog checks two things: whether the scheduled credentials still match (catching drift before the next Monday), and whether every company has actually completed a run in the last eight days. A problem sends an alert the same day; no problem sends nothing. It is deliberately configured so that it cannot be disabled by the very drift it exists to detect.

Other scheduled work

• Sunday evening — deal snapshots are written so the cockpit's week-over-week motion has something real to compare against.
• Daily — trial and plan expiry, trial-ending warnings, data-retention warnings, and lead reconciliation from connected CRMs.
• Hourly — mailbox delta syncs across every connected account.
• Continuously — transcript sweeps and meeting re-scans that catch anything a webhook missed.

14

Security, safety, and how Klo behaves

Klo never goes around the rep

• Klo never sends email on its own. Drafts stay drafts until the rep presses send. This is an architectural invariant, not a setting.
• Buyer-facing questions default to approval mode. Auto-send is opt-in per deal and reversible at any moment.
• The rep previews the buyer's entire view before sharing a room.
• Everything the buyer writes goes to the rep, never to a bot.
• The AI proposes; deterministic code executes. Every change to a deal's state runs through an authorisation-checked operation, and every one of them is reversible.
• The AI never gets between the seller and their customer relationship. That is the line the product is built around.

Access control lives in the database

Permissions are enforced by row-level security in Postgres, not by hiding buttons in the interface. Reps see only their own deals. Managers see only the teams they manage. Sales heads see their organisation. Buyers see only the room they were invited to.

The Buyer Room's tables go further: they have row-level security enabled with no policies at all, which denies every direct request. The only access path is server-side functions scoped to a single room and a single contact. Manager-only content — the candid read on how a rep is running a deal — is stored in separate tables and returned only to a manager who actually manages that rep.

Secrets, tokens and webhooks

• Mailbox and CRM refresh tokens are encrypted at rest with AES-256-GCM; tokens are never logged in plaintext.
• OAuth round-trips carry an encrypted, authenticated state binding the user, organisation and provider, so a callback cannot be tampered onto another tenant. PKCE is used where the provider supports it.
• Buyer magic-link tokens are stored only as SHA-256 hashes.
• Every inbound webhook is signature-verified — including a timing-safe comparison on billing events.
• No AI or infrastructure secrets ship in the browser bundle.

Prompt injection: deal content is untrusted input

Deals contain text written by other people, so Klosure treats that text as data rather than instruction. Buyer and deal context is fenced as untrusted reference material in the prompts that produce buyer-facing output. Imported spreadsheet cells are wrapped as quoted data and explicitly marked as never-instructions. Buyer answers are display-only and never reach the confidence engine. The buyer-question generator uses a closed enumeration, so an out-of-bounds question is structurally impossible rather than merely discouraged. And the confidence engine has the deterministic ceiling behind it — even if a model could be talked into a number, code clamps it back down. Klosure maintains a dedicated prompt-hardening test suite covering injection, buyer-leakage temptation, contradictory signals, and messy multilingual input.

Data handling

Your deal data is never used to train any AI model. Klo runs on frontier large language models accessed through enterprise API terms that prohibit training on customer data, with model choice governed centrally by a two-tier policy so every surface a person reads stays consistent. Data is encrypted in transit and at rest in a managed PostgreSQL database. Deal rooms are private by default with no public discovery, and a buyer reaches a room only through a per-recipient link. Meeting capture uses a visible bot and the rep can remove it from any meeting.

When an account lapses

Trials expire to read-only rather than vanishing, with warnings before anything is removed — and the warning copy always frames the action honestly as "upgrade to keep your account", never as a misleading "click here to keep your data". Closed deal rooms are never deleted: a won or lost deal locks and stays as history, and reopening is one click.

15

How much does Klosure cost?

Growth includes: Klo coaching on every conversation; the dynamic dashboard for rep and client; the manager dashboard with team roll-up; daily focus and weekly briefs for manager and rep; Ask Klo trained on your business; the confidence engine; email and notetaker capture across Gmail, Outlook, Zoom, Meet and Teams; 15 hours of notetaker meetings per seat, pooled across the team; and Klo learning from every email and meeting.

Quotas are pooled at team level, not per rep — meeting minutes, chat and voice all draw from a shared pool sized by seat count, with alerts to the team owner at 80% and 100%. Seat changes follow the industry norm: added seats apply immediately with proration; removed seats take effect at the end of the billing cycle, with the pending count shown in advance. The buyer side is always free.

If Klosure saves even one deal a quarter, it pays for itself several times over. Book a 20-minute walkthrough for pricing tailored to your team.

16

Who is Klosure for?

B2B sales teams, sales managers and revenue leaders who want to move deals forward, not just log them — running considered, multi-stakeholder deals where a single quiet stakeholder can cost a quarter.

• Industries: B2B SaaS, IT services and systems integration, and project-based revenue.
• Team shape: typically 10–50 reps — a sales head with two to five first-line managers underneath.
• Deal shape: roughly $20K–$500K, cycles measured in weeks to months, with real procurement involved.
• Geography: especially strong fit in the Gulf (UAE, Saudi Arabia, Qatar, Bahrain, Kuwait, Oman) and India, because the regional realism is encoded rather than assumed — but sold globally, with US, UK and EU norms covered by the same engine.

17

How is Klosure different?

The eight things that are hard to copy

1. A deal record no human types into.
2. A confidence score the person being coached cannot talk up.
3. Deterministic ceilings in code that outrank the model's own judgment.
4. Evidence attached to every factor, so the forecast survives a board review.
5. A human approval gate on everything that reaches a buyer.
6. A buyer room whose security model makes buyer behaviour an independent signal rather than a rep-reported one.
7. Deterministic dollar figures everywhere, with the AI narrating above them and never inside them.
8. Regional deal realism — working weeks, holiday freezes, procurement floors — encoded rather than assumed.

See the full teardown on the Klosure comparison page.

18

What Klosure doesn't do yet

Stated plainly, because a vendor that only lists strengths isn't describing a product. These are current boundaries, not permanent ones.

19

Glossary of Klosure terms

20

Frequently asked questions

See it on your own pipeline

Twenty minutes, one of your real deals, and Klosure's honest read of it — including the parts you won't enjoy hearing.

Book a demo →  ·  Compare Klosure to your current stack  ·  Read the research