How 141 accounts got re-tiered in three days, not a quarter
Backstory's customer success lead described the exercise at SaaStr AI Day: eight signals cut to four, four connectors instead of five teams, and one signal that was scoring the wrong way round.
The board asked Haya Kamola, who leads customer success at Backstory, to define what the company's best customers looked like, measure the rest against that, and hand the executive team a tiering framework.
There were 141 accounts. On her account of the project, described at SaaStr AI Day, the same exercise in a previous job took her team plus four others a full quarter. This one took three to four days.
She started with judgement, not data. Account teams and senior leaders were asked what made one or two specific customers feel different — not the biggest, not the oldest.
The answer: customers who built systems around the platform, planned five years out with it at the centre, wanted a say in the roadmap, and kept finding new use cases. Everything else was scored against that description.
Her point is that without the definition you end up scoring accounts against whichever CRM fields happen to be filled in.
Most of the signals that mattered did not exist yet. Deployment velocity, executive visibility, addressable seats inside the account, tech stack mix, AI maturity — several had never been measured consistently across the whole base.
So they were built first. Backstory had a five-level AI maturity framework, previously maintained by asking account teams to categorise customers by hand, which Kamola called grueling. It was rebuilt as a prompt run against every account, pulling CRM fields, public company information and the full conversation history, and returning a level plus its reasoning.
The cross-functional data pull — the part that used to eat the quarter — became four connectors: Amplitude for usage, Atlassian and Jira for logged feature requests, Backstory's own MCP for conversation history, and Slack, because the company runs an internal channel for every customer.
That last one is the cheapest to copy. The account team's private read on a customer is usually the earliest one, and it rarely reaches any structured system.
The only manual step was a CSV export from Salesforce: account name, executive engagement, predicted health, AI maturity, renewal date and renewal ACV. The analysis runs as an ordered sequence rather than one prompt, and a full run takes about 20 minutes.
It took three or four iterations. Eight signals became four scoring buckets — growth potential, AI maturity and its velocity, engagement level, and account health — because contradictions between signals cost more than the extra inputs were worth.
The more useful error: the first version docked points for a high volume of feature requests, on the assumption that a customer asking for a lot is unhappy. The data went the other way. Backstory's highest-adopting customers correlated strongly with request volume.
A mistake like that survives forever in a hand-built model, because nobody runs the model across the whole base and checks who landed where.
The output was four tiers. Tier A holds 8 accounts, Tier B 21. Tier D is the list of customers the company concluded may not be there in two years, and the decision to make is whether to service them differently.
In the field it has already changed the QBR cadence for Tier A, added an executive sponsor programme, and tied per-account growth targets to AE pipeline goals. Backstory plans to rerun it quarterly.
The lesson for anyone tiering a book this quarter: write the definition before you open the spreadsheet, build the missing signal rather than substituting the one that is already in the CRM, and assume one of your weights is pointing backwards until you have checked the full list against what you already know.