Blog

Notes on the product as it is built, not case studies:there are no customers yet to write those about.

app.salescrew.io/today

Posts here cover how a feature works, why a design choice was made, and what shipped in a given week. Anything describing a feature that is not live yet is marked as a roadmap item, with a link to the roadmap page.

A Sales Team of One Person Plus Agents

One person cannot draft every reply, log every call and research every account. What changes when an agent does the first pass and a person keeps the judgment.

A Voice Agent That Books Into Nothing Is a Demo

A voice agent that cannot write the booking into a real pipeline is a phone trick, not a sales tool. The call has to land somewhere a rep or a stage sees it.

Agents Need a Queue, Not a Toggle

A single on/off switch for AI cannot tell a safe action from a risky one. What agents actually need is a policy per action class and a queue for the risky ones.

Audit Logs Are a Feature, Not Plumbing

Most products treat an audit log as a compliance checkbox added late. Once agents act on their own, the log becomes the only way to answer what happened and why.

Building a CRM a Sales Team Will Actually Use

Adoption fails when a CRM is a data-entry chore layered on top of how a team already works. What changes when the CRM meets reps where they already are.

Can a Daily Briefing Replace the Standup?

A daily digest can replace the status-reporting half of a standup. It cannot replace the conversation where a blocker gets noticed and solved on the spot.

Changing a Live Schema Without Downtime

Additive migrations and background backfills keep one database online during a schema change. Single-tenant means doing that reliably across every client's project.

Check Suppression at Freeze, Not at Send

By send time a message is already written, personalised and queued. Suppression belongs at the moment the audience is frozen, with the send check as a backstop.

Cost per Lead Should Be a Query

Most teams keep ad spend in one tool and deals in another, so cost per lead is a monthly manual join. It should be a query against one database.

Deliverability Is an Operations Problem

A better subject line does not fix a sender reputation problem. Deliverability is warmup, authentication, caps and bounce handling done consistently.

Do Not Trust the Verified Flag on Bought Data

A bought contact list's verified flag reflects a check made at some point in the past, not a guarantee that the mailbox works today. Treat it as a starting point.

Dual Control: Why Every Button Is Also a Tool

SalesCrew builds every UI action and its MCP tool twin together, with identical effect. That rule is what keeps automation from becoming a separate, riskier path.

Every Agent Should Ship in Review Mode

An agent's default mode on day one should queue its work for a human, not run it. Review mode is how trust in an agent gets earned instead of assumed.

Hiring When Agents Write the First Draft

When an agent drafts the reply, the rep's job shifts from writing it to judging it fast. Hiring and onboarding should test for that, not typing speed.

How to Measure Whether AI Engines Mention You

Answer engines cite pages, not sites. Measuring visibility means asking the same set of real buyer questions repeatedly across models and logging what gets cited.

How We Price AI Without Credits Theatre

A vague credit system hides what a credit actually costs or converts to. SalesCrew meters AI in model calls, with a plain allowance, a soft alert, and a hard cap.

Marketing Data Belongs Next to Deal Data

Knowing which channel produced a lead is not the same as knowing which channel produced a closed deal. That answer needs spend and deal data in one place.

MCP Is the Integration Layer That Will Stick

Every vendor building a one-off AI chat feature is solving the same problem separately. MCP is the shared protocol that means one integration works everywhere.

Provisioning a Client Instance in One Click

Single-tenant means every new client is a new database, not a new row. That is a harder provisioning problem, and automating it well is still on the roadmap.

Put Ad Spend in the Same Database as Deals

Marketing spend and CRM deal data usually sit in separate tools, so channel economics gets rebuilt by hand. Keeping them in one database changes what is answerable.

Row-Level Security Is Not Enough

Row-level security is a real control, not a complete one. It only isolates customers who share a database in the first place, and one wrong policy defeats it.

Six Things an Agent Should Never Do Alone

Send, delete, merge, move a deal, spend money, or act outside the CRM. The line is not how smart the agent is, it is whether the action can be taken back.

Snapshots Are the Thing Agencies Actually Buy

A single feature helps one client a little. A proven configuration exported as a snapshot and applied to the next client is what turns setup into a repeatable asset.

The 'Interested' Label Is Lying to You

An automated reply label carries no confidence score, so 'Interested' can mean the reply had a question mark. Treat it as a sort order, not a qualification.

The Agency Model That Runs on One DB per Client

An agency reselling a CRM needs to tell each client their data is really theirs. A database per client, plus a portable configuration, is how that claim stays true.

The Config Is the Product

The same codebase running for every client is not what makes a CRM feel tailored. What each instance is configured to do is what actually differs.

The CRM Market Is Splitting in Two

One branch of CRMs stays a system a human clicks through. The other is being built so an agent can operate it directly. The two need different architecture.

The Kill Switch Nobody Else Ships

Per-agent toggles are not a stop button. When something goes wrong and you do not yet know which agent caused it, you need one control that stops all of them.

The Outbound Metrics That Predict a Quarter

Closed revenue tells you how the quarter went. Reply rate, meetings booked and cadence completion tell you how it is going, while there is still time to act.

The Trade-Offs Behind a Single-Tenant CRM

One database per client buys isolation and a real export. It costs more to run and to provision. Here is the honest trade a single-tenant CRM is making.

The Unsubscribe Line Cold Email Misses

CAN-SPAM asks for a clear opt-out, honoured within 10 business days, and a real postal address. Cold email that drops these is breaking a rule, not saving space.

Two Tools for Outbound Is a Design Failure

A CRM plus a separate sending tool means suppression, reply state and deal stage live in two places. The sync between them is where the mistakes get made.

Vertical Tools vs Configurable CRMs

A vertical CRM ships trade-specific defaults out of the box. A configurable one takes longer to set up but does not lock you into one trade's assumptions.

What "AI-Native" Actually Means

AI-native gets used for a chatbot bolted onto old software. The real test is whether every action in the product is also something an agent can call directly.

What 140+ MCP Tools Taught Us

Building an MCP tool for every CRM action, not a chat wrapper on top, changes what you build first: scoping, naming and the approval queue before the tools.

What a CRM Actually Needs on Day One

Five things: contacts, companies, a pipeline with stages, tasks, and a timeline. Automation added before those are right just moves bad data around faster.

What a Voice Agent Actually Costs Per Minute

A voice agent's per-minute cost is four line items stacked: hosting, model tokens, speech synthesis and transcription, and telephony. Here is the formula.

What an AI Receptionist Should Do When Stuck

An AI receptionist fails the moment it cannot answer, not the moment it starts talking. What a good failure path looks like, and why it needs a human backstop.

What an Inbox Agent Should Touch

An inbox agent can draft, classify and log all day without much risk. The moment it sends, changes a stage, or touches a record, the rules have to change.

What Answer Engines Actually Cite

Answer engines tend to lift a direct answer, a table, or a named number over a paragraph of marketing prose. Here is the structural reason, and what to do about it.

What Belongs in a Sales Prompt Pack

A sales prompt pack needs persona context, an objection list, tone constraints, and explicit limits on what the agent may never claim. Not only a tone of voice.

What Breaks Past 100 Mailboxes

Sending capacity scales with mailboxes. Authentication, warmup, reputation and bounce handling do not. Past a hundred, the operations become the constraint.

What GoHighLevel Got Right

GoHighLevel bundled CRM, funnels, SMS and calendars into one white-label product agencies could resell fast. That bundle is the thing worth understanding.

What the AI in Our CRM Actually Costs to Run

Folding AI cost into a flat 'AI included' line hides which features are expensive to run. Metering per feature and per tier is the harder but more honest choice.

What We Got Wrong Going From Tool to SaaS

SalesCrew started as one team's CRM before it became a product other companies configure. Turning an internal tool into a SaaS rewrites more than the pricing page.

When Multi-Tenant Is the Wrong Choice

Multi-tenant is the right default for most software. It is wrong when a customer needs a real database export, a residency choice, or a hard isolation boundary.

Why 'Verified' Email Data Still Bounces

Email verification checks a mailbox at one point in time. Mailboxes get deactivated later, which is why a real bounce should count more than a stale verified flag.

Why Every Agent Tool Call Needs a Log Line

A record-level change log tells you what changed. It does not tell you which of 145 tools an agent called, with what arguments, or which run did it.

Why Marketplace Bids Belong in Your CRM

A bid on Upwork or Fiverr is a deal like any other. Tracking it in the marketplace's own dashboard instead of your pipeline is why it gets lost.

Why Real CRMs Ban Cold Email

Most major CRMs prohibit cold outbound in their acceptable-use terms to protect shared sending infrastructure. That policy, not a technical limit, is the real wall.

Why Sales Teams Abandon a New CRM

Adoption fails when logging a deal costs a rep more time than it saves them. The fix is removing the entry step, not adding a reminder to use the tool.

Why We Gave Every Client Their Own Database

SalesCrew runs one Supabase project per client instead of one shared database. It costs more to operate. Here is the mechanism reason it was worth it.

Writing for Machines That Quote You

An answer engine lifts a sentence, not a page. Writing so the first sentence survives being quoted alone is a different discipline than writing to be read in full.