Your target account list has 1,800 companies on it. Two reps can work maybe 40 of them this week. Account scoring is what decides which 40, and on most teams that decision is either a gut call or a number nobody trusts enough to argue with.
The standard advice is to buy a platform that scores accounts for you. That advice quietly assumes three things: a marketing automation stack, enough closed-won history to train a model on, and someone whose job is to maintain it. Plenty of B2B teams have none of the three and still need a defensible ranking by Monday morning.
This is a model you can build in a spreadsheet. The weights are written out, the recency rule is included, and the disqualifiers come first.
What Is Account Scoring?
Account scoring is the practice of ranking companies, rather than individual contacts, by how well they match your ideal customer profile and how much public evidence suggests they are in a buying window right now.
The output is an ordered list with a cutoff. The number itself does not matter much. What matters is the line it draws between accounts a rep works this week, accounts the team watches, and accounts nobody touches.
Lead scoring answers a different question, namely which of the people who already raised a hand is worth a call. Account scoring runs before anyone raises a hand, which is the scoring problem outbound teams actually have.
Why Most Account Scores Are Really Fit Scores in Disguise
Most account scores lean almost entirely on fit, because the inputs that were supposed to balance fit are empty.
The category default splits a model into three parts. Demandbase's account scoring guide breaks a modern account scoring model into account fit, intent, and account engagement, and most vendor guides describe some version of the same three.
Engagement is where outbound breaks. You are contacting companies that have never opened an email, visited a pricing page, or sat through a webinar. Their engagement component is zero by definition, not because they are a bad account.
Platform defaults carry that zero straight through. Adobe's documentation for Marketo Engage describes account scoring as aggregating the lead scores of people at the account, with example formulas that sum person-level product interest and web engagement scores. If nobody at the company has engaged with you, the sum is zero no matter how obviously that company is in market.
Take engagement out and a three-part model collapses into fit with rounding error. That is why a scored list can feel identical to the static filter it was supposed to replace.
The Three Inputs a Small Team Can Score by Hand
Replace engagement with external evidence. You cannot measure how an account behaves toward you, but you can measure what it is doing in public.
- Fit. Industry, size band, and whether the company runs the function you sell into.
- Signal strength. What the observed signal actually proves about a problem, an owner, or a deadline.
- Signal recency. How long ago that signal appeared.
Signal strength is the input teams get wrong most often, because it is tempting to score a signal by how exciting it feels rather than by what the signal actually proves.
How to Weight Fit, Signal Strength, and Recency
Give fit 40 points, give signal evidence 60, and let recency cut the signal side rather than the fit side.
Fit, out of 40 points:
- Industry match: 15
- Size or headcount band: 15
- Runs the function you sell into: 10
Signal strength, scored on the strongest single signal, out of 45 points:
- Proves a deadline, such as a posted RFP, a procurement notice, or funding tied to a stated initiative: 45
- Proves an owner, such as a leader hired or promoted into the function you sell to: 30
- Proves a problem, such as job postings for roles that do the work your product does, or public complaints about an incumbent: 20
- Proves only activity, such as a generic growth announcement or a press mention: 5
Add a second signal bonus of 15 points when an independent signal appears on the same account within 60 days. Two modest signals that agree usually beat one loud signal on its own, which is the whole argument for stacking signals.
Fit plus signal strength plus the bonus gives you 100 points. Set the tiers at 70 and above for accounts worked this week, 45 to 69 for a watch list, and below 45 for accounts left alone until something changes.
Why Recency Gets Its Own Multiplier
Recency should multiply the signal side of the score rather than add to it, because a stale signal is not a weaker reason to reach out. It is not a reason at all.
Multiply the signal points, bonus included, by the age of the signal that earned them:
- 0 to 14 days: 1.0
- 15 to 45 days: 0.6
- 46 to 90 days: 0.3
- Over 90 days: 0
Fit is left alone. A company's headcount band does not go stale in a fortnight. Its hiring spike does.
Scoring tools do support decay, usually at a coarser grain than signals live at. HubSpot's lead scoring documentation describes score decay set to intervals of 1, 3, 6, or 12 months. A one-month minimum is slower than the useful life of most hiring and launch signals, which is the practical case for tracking signal age yourself.
Set Your Disqualifiers Before You Set Your Weights
Disqualifiers are hard zeros rather than point deductions, and writing them first stops you tuning weights around accounts that should never have been scored.
- Already a customer, or an open opportunity
- Contacted in the last 90 days with no reply
- Below your headcount or revenue floor
- In a region you cannot service or support
- On a legal, compliance, or do-not-contact exclusion
The second one matters more than it looks. A 92-point score on an account a rep burned six weeks ago is worse than a blank row, because it sends someone to repeat a conversation that has already failed once.
A Worked Example: Four Accounts, One Monday Morning
Four accounts run through the model. The companies are illustrative rather than customers.
- Account A. Strong fit at 38. Posted an RFP nine days ago, which proves a deadline, so 45 points at a 1.0 multiplier. Total 83, worked this week.
- Account B. Perfect fit at 40. Hired a VP into the relevant function 70 days ago, so 30 points at a 0.3 multiplier, which is 9. Total 49, watch list.
- Account C. Medium fit at 24. Three job postings naming the workflow you replace, 11 days old, so 20 points at 1.0. A new integrations page went live 30 days ago and earns the second signal bonus, 15 at 0.6, which is 9. Total 53, watch list and close to the line.
- Account D. Perfect fit at 40. Nothing public in 90 days. Total 40, left alone.
B and D have the best fit on the list and the weakest claim on anyone's Monday. A static filter would have put both at the top, and a rep would have spent the week on the two accounts carrying the least evidence.
That inversion is the point of scoring, and it is also the first thing to check when a team reports that scoring changed nothing. If the ranking still matches what the old filter produced, recency is probably not doing any work.
How to Calibrate the Model Without Historical Data
You do not need a regression. You need twenty closed-won deals and an afternoon.
Pull the last twenty wins. For each one, write down what was publicly observable about that company roughly thirty days before first contact: a job posting, a funding round, a leadership change, a new page on their site, a conference talk.
Count which signals keep appearing. Raise the strength score on the ones that recur across several wins, and cut the ones that never show up at all. Rerun it once a quarter.
Twenty deals will not give you statistical significance, and it is worth being honest about that. What it gives you is a sanity check on your own assumptions, which is a more modest thing than a trained model and still better than weights nobody ever questioned.
Where a Hand-Built Score Breaks Down
Hand scoring holds up to a few hundred accounts refreshed weekly. Past that it fails on upkeep rather than on arithmetic.
The recency column rots first. Nobody rechecks 600 accounts every Monday, so multipliers sit at 1.0 long after the signals behind them have aged out, and the ranking quietly inverts: the accounts at the top become the ones you scored longest ago.
The model also cannot see signals you are not watching, and it only ever encodes assumptions you already hold. It will not find a pattern you have not thought of.
Running the Model Without the Spreadsheet
The upkeep problem is a monitoring problem. Someone has to check job boards, funding announcements, new product and pricing pages, review sites, and community discussion for every account every week, and then update one column.
Alsona is built for that step. You define your ICP, and it monitors unstructured sources including hiring, funding, technology, and community activity, then scores prospects on those signals together with ICP fit rather than on fit alone.
Every record shows the signals behind its score, so a rep can see why an account surfaced before deciding whether to open it. That same context feeds the individualized LinkedIn and email outreach that follows, which keeps the reason an account ranked highly and the reason given in the first message the same reason.
Build outbound around real buying intent rather than guesswork. See how Alsona scores accounts on unstructured buying signals.
