Do I Need a CRM? Where Notes and a Spreadsheet Stop Holding

The quote was in a text. The follow-up date was in your head. The customer's second message landed in an inbox somebody else checks, and by the time it surfaced they had hired the other company. That is the moment this search usually starts, and it comes with a second fear as strong as the first: buying a system built for a sales team, watching nobody enter anything into it, and ending up with the spreadsheet plus a subscription. Both fears are reasonable. The way through them is to name the thing that actually failed before you look at a single product, because a CRM fixes one kind of failure very well and three other kinds not at all.

The situation has a shape, and it is the same one the Small Business Administration describes in its CRM implementation article: emails, messages and calendar reminders that are disconnected and hard to find, with leads and commitments falling through the gaps because nobody has an end-to-end view. Yours is a notebook, a phone, an inbox, a calendar and a sheet, and the gap is between any two of them.

The cost of the gap has one well-known measure, and it is worth knowing its limits. A Harvard Business Review audit from 2011 checked how 2,241 US companies responded to a web lead: 37% replied within an hour, 24% took more than a day, and 23% never replied. In a separate dataset the same article found that contact within an hour was associated with nearly seven times the likelihood of qualifying the lead compared with an hour later. Qualifying meant a real conversation with a decision maker, not a sale, and one of the authors ran a sales-software company at the time. What the figure supports is narrow and true: a follow-up that happens the same day is worth more than one that happens next week, and a system where follow-ups depend on memory will keep producing the 23%.

Whether a CRM is the fix depends on what dropped the follow-up. If nobody owned it, or the history lived in somebody's phone, that is a CRM-shaped problem. If the technician was double-booked, that is a scheduling problem, and a CRM will not touch it.

A CRM is the right category to test when customer follow-up, ownership of the next action and the history of every interaction need to be shared reliably between people, and the current setup cannot pass that test even after a bounded repair. Two people who both need to know what was promised to a customer, and cannot, are the classic case.

The SBA's advice is to identify two or three primary pain points before evaluating any provider, and it warns that a pricey system can add complexity that does not suit a small company. Headcount is not the threshold. A one-person business whose follow-ups live in six places can need one, and a five-person business with one trusted sheet and a clear owner for every next action may not. "Just use more spreadsheets" fails in the opposite direction: a sheet can be tightened a great deal, and it still cannot make anyone own the next call.

Before any demo, make the mess visible. List every place a customer, lead, quote, appointment, status or next action gets created or changed: paper, phone notes, texts, email, the calendar, documents, the sheet. Then reconstruct two or three recent customer journeys from first enquiry to last action, and make one of them the job that went wrong. Mark every point where information was re-entered, handed off, owned by nobody, or could not be found. That map is the specification, and no vendor can write it for you.

From the map, define the minimum record in plain words. A stable identifier. The customer. A status from a short controlled list. An owner. A next action. A promised date where one exists. The approved quote or scope. One consistent place for the supporting conversation. Then turn the failure into tests you can pass or fail: "the owner can see every quote with no next action", "either of us can find the promised scope from a job number", "a reschedule updates one record and nothing else has to change". Those tests are how you will judge a repaired sheet, a CRM, or anything else, and the decisions inside them belong to you: which fields are required, who may change a status, when a quote counts as approved, who owns what.

A surprising amount of the failure can be fixed inside the sheet you already have, and it is worth doing first because it tells you whether the problem was the tool or the discipline. Give every active record a stable ID. Make one row the master for each customer or job. Replace free-text statuses with a short controlled list. Add an owner column and a next-action column and refuse to leave them blank. Keep related documents in one predictable place. Google Sheets supports protected ranges for formulas and reference lists, editing restrictions by person, and version history with restoration, so accidental damage becomes recoverable. The earlier piece on when spreadsheets stop being enough covers where the limits of that repair sit.

The repair leaves two gaps. Google says plainly that sheet protection is not a security measure, because anyone with edit access can copy or export the whole thing, so a sheet holding anything sensitive is a sheet with a security problem however tidy it is. And no column makes a person own the follow-up. If the repaired sheet passes your tests during a normal working cycle, keep it for now. If it fails on ownership and history, you have your answer, and you have it from evidence rather than from a sales call.

When the category is a CRM, the pilot is small, reversible and run against the real failure. Pick a handful of live records rather than the whole customer base. Run a real enquiry, a real follow-up, a real quote, a real change and a real handoff through the tool, and ask the vendor to demonstrate your specific missed-follow-up scenario rather than their standard walkthrough, which the SBA also recommends. Judge the pilot against the tests you wrote, not against the feature list.

No fixed number of weeks fits every business, because record quality, active jobs, integrations and staff time all move it. What you can do is time-box each stage, and set the cutover date only after the pilot has passed. A vendor's promise of implementation in days describes their setup, not your data.

Two vendors' own pages, checked in September 2026, show why the price is a configuration rather than a number. HubSpot's Sales Hub pricing displayed a free tier for up to two users, Starter at $7 per seat per month billed annually or $20 monthly, Professional at $90 annually or $100 monthly, and Enterprise at $150, with one-time onboarding fees of $1,500 for Professional and $3,500 for Enterprise. Salesforce's Sales pricing page displayed a Free Suite at $0, Starter at $25, Pro at $100, Enterprise at $175, Unlimited at $350 and Agentforce 1 Sales at $550 per user per month, with add-ons shown at $100 to $220 per user per month. Those are display prices for those dates, from two of many vendors, and they are examples rather than a recommendation.

The range they describe is $0 to $550 per user per month before onboarding, add-ons, usage, tax, migration and the hours you and your staff will spend. The repaired sheet costs no subscription and is still not free, because the cleanup and the process decisions are somebody's afternoon. Price the exact configuration you would actually use, then set it beside the count of exceptions from the map: enquiries with no owner, quotes with no next action after the promised date, duplicate customers, time spent reconstructing a history. That comparison is the business case, and it is one only your own records can make.

Read the order form and not the pricing page. The term and renewal date, the seat type and count, contact or usage limits, onboarding fees, what support is included, the cancellation route, refund rules, price-change terms, the export steps, retention and deletion, sub-processors, and who is responsible for security. Two current examples of why. HubSpot's data processing agreement, last modified 14 April 2026, strongly recommends retrieving customer data before the subscription term ends, says data is deleted or returned on termination according to product procedures, and says retrieval assistance may be available at the customer's cost. Housecall Pro's terms say subscriptions renew automatically unless terminated before the renewal date. Neither is unusual, and both are things to know before the import rather than after.

The data itself has legal weight in Texas. Business and Commerce Code Chapter 521 requires a business that maintains sensitive personal information to implement reasonable procedures to protect it, and requires notice to affected people no later than the 60th day after a breach is determined, subject to its stated exceptions. The practical rule for a trial or a migration is to minimise what goes in: no payment details, government identifiers, bank information or health data in a casual account for convenience, restricted access from day one, and the processor's terms read before the first record. The Texas Data Privacy and Security Act, per the Attorney General, generally exempts small businesses as federally defined except where sensitive data is sold, and whether it reaches you depends on your business and your processing. This is general information, not legal advice.

No local rule sets a threshold for moving from a notebook to software. What the city offers is somewhere to think it through with people who have done it. The Arlington Economic Development Corporation's industry cohorts, free and connecting businesses with subject-matter experts, had served 20 Arlington businesses in professional-services and construction groups since September 2025, according to the city, and the 2026 resources page adds LiftFund's 0% interest loans, a Strategic Solutions Fund for local manufacturers, and the three chambers as networks. An operations decision like this one is exactly what those rooms are for.

Whatever you end up with, the thing that prevents the next dropped follow-up is an operating standard rather than a feature. One current record for every active customer or job. A fixed meaning for every status. One named owner and one next action on every record. A defined place for the approved scope and the conversation behind it. Every commitment made to a customer captured in that record or linked to it.

Then a weekly exception check, which takes ten minutes and is the whole discipline: records with no owner, next actions past their date, jobs with no approved scope, duplicate customers, and conflicts between the calendar and the master list. You know the system works when either of you can take a job number and answer, without opening anyone else's phone, what was promised, who owns the next step, where the customer stands and where the conversation is. Measure the same exception counts you measured before, over the same period, and a shorter list is your evidence. Nothing about it promises revenue. It promises that the next customer's second message lands somewhere with a name on it.

Do I really need a CRM if I only have two people?

Headcount is not the test. If one person owns every follow-up, the record is trusted and the workflow survives a normal week, a repaired sheet may be enough for now. If two people both need the same customer history and cannot get it, size is irrelevant and a CRM is worth piloting.

Spreadsheet or CRM?

Repair the spreadsheet first: stable IDs, controlled statuses, an owner and next action on every row, protected ranges, version history. Run it for a cycle against your tests. If it fails on ownership or shared history, pilot a CRM. If it passes, keep it and revisit when it stops passing.

What does a CRM cost for a small business?

Two vendors' own pages in September 2026 ran from $0 to $550 per user per month across tiers, with onboarding fees of $1,500 to $3,500 on some plans and add-ons on top. The figure that matters is the exact configuration you would use, plus migration and your own hours.

How do I try one without uploading all my customers?

Pick a handful of live records, run your real missed-follow-up scenario through it, reconcile the counts, keep a read-only backup, and only import the rest after the tests pass and you have read the export terms.

What customer data should stay out of a trial?

Payment details, government identifiers, bank information and health data. Texas requires reasonable protection for sensitive personal information and 60-day breach notice, and a trial account is not where that data belongs.

When is a CRM the wrong answer?

When the thing that failed was scheduling, dispatch, invoicing or an undefined status. Those need a different category or a decision, and a CRM bought for them becomes the subscription nobody uses.

The job that fell through was not lost to a missing product. It was lost to a follow-up nobody owned and a history nobody could find, and whether a CRM fixes that depends on whether those two things are what failed.

Map the mess, write the minimum record and the tests, repair the sheet, and let the results choose the category. If it is a CRM, pilot it with a handful of records before you trust it with all of them, and read the exit terms before the entry.

If the map shows that the follow-up problem sits in the enquiry form and the inbox before any record exists, Arlington Website Designer wires that first step for businesses in Arlington, TX so an enquiry lands with a name and a date on it. Describe where your leads arrive and where they go next through the contact page, and you will get back the two or three places the handoff is breaking and what would close them.