By use case
Sales & Revenue Operations
Clean pipeline data and follow-ups that actually happen
Usually owned by
- Sales
- Revenue Operations
- Marketing
- Customer Success
- < 5 min
- Inbound lead acknowledged and routed
- 8 hrs/week
- Admin time recovered per rep
- 90 days
- Renewal risk visibility, from reactive
The change
What actually differs afterwards
Not a maturity model. The concrete difference in how the work happens.
Today
- CRM data entered inconsistently, so forecasts are guesses
- Inbound leads sit for a day before anyone looks
- Renewals discovered in the month they expire
- Proposals rebuilt from scratch each time
- Reps spend a third of the week on admin
Afterwards
- Records enriched and normalised automatically on creation
- Inbound leads scored, routed and acknowledged in minutes
- Renewal and churn risk surfaced 90 days ahead
- Proposals drafted from the deal record, then edited by a human
- Reps get time back for conversations
The data problem underneath
Every revenue operations project turns out to be a data project. Forecasts are wrong because stages mean different things to different reps. Territories overlap because account records are duplicated. Renewal dates are unreliable because nobody updates the field.
Automating on top of that produces faster wrong answers. So the first workstream is almost always hygiene: deduplicate, normalise, enrich on write, and enforce the definitions.
What we automate, in order
Enrichment and normalisation on creation. Company size, sector, technology signals and geography attached automatically, in a consistent format, at the point the record is created — not in a monthly clean-up.
Inbound triage. Score against fit criteria, route to the right owner, and acknowledge within minutes. Response time is the single strongest predictor of inbound conversion and the easiest thing to fix.
Renewal and risk surfacing. Usage signals, support history and engagement combined into a 90-day forward view. Renewals should never be a surprise.
Proposal drafting. Assembled from the deal record and your standard language, with the relevant case studies pulled in. A rep edits and prices it.
Explainable beats accurate
For lead scoring specifically, a transparent rules-plus-fit model that a sales leader can inspect and argue with gets adopted. A slightly more accurate opaque model gets ignored within a quarter. Adoption is the constraint, not accuracy, and we optimise for the one that actually changes behaviour.
Adoption is the whole engineering problem
This is the use case where the software being correct is least sufficient. Reps are measured on quota, not on data quality, and anything that costs them time gets routed around within a fortnight — quietly, and without anyone reporting it as a failure.
Four things that decide the outcome, none of them about models:
Enrichment has to be invisible. Data that appears on the record without anyone filling in a form gets used. A required field added to improve reporting gets filled with a full stop.
The scoring rule has to be arguable. A sales leader who can look at why a lead scored 82 and disagree with the weighting will engage with it. A score with no explanation gets dismissed as "the tool" within a month, regardless of accuracy.
Drafts must be genuinely better than a blank page. A proposal draft that needs heavy rewriting is slower than starting fresh, and reps only need to experience that twice before they stop opening it.
Remove what is not used. Usage should be tracked per feature and the unused ones deleted rather than mandated. Mandating a feature nobody wants produces compliance theatre and poisons the next thing you ship.
The data layer underneath
Enrichment, deduplication and consistent definitions are data platform work applied to revenue records. On smaller estates this stays inside the CRM and its native tooling. Once revenue data has to reconcile with finance and product usage — which is where renewal risk scoring gets its signal — it needs the modelled layer, and pretending otherwise produces the familiar situation where three systems report three different ARR figures.
Where this shows up by sector
- SaaS and technology — renewal risk and expansion signals from product usage, which is the highest-value version of this work and the one most dependent on a real data layer.
- Retail and eCommerce — wholesale and B2B pipeline alongside the direct channel, usually with a longer tail of small accounts than the CRM was configured for.
- Logistics — rate quoting and tender responses, where the proposal drafting work has more structured input to draw on than most sectors.
FAQ
Questions we get asked
Will AI write our sales emails?
It can draft, and a human should send. Fully automated outbound at volume damages your domain reputation and your brand, and recipients can tell. Drafting from real deal context so a rep edits rather than composes is the version that works.
Our CRM data is a mess. Fix that first?
Yes, and it is usually the highest-return item. Deduplication, normalisation and enrichment on write, plus a backfill pass. Automation built on unreliable records produces confident nonsense, and forecasting on it is worse than not forecasting.
Can it do lead scoring?
Yes, and worth being clear about how. Rules and firmographic fit get you most of the value with an explanation a sales leader can argue with. An opaque model that says "0.73" and cannot say why does not get trusted or used.
What about our proposal process?
Drafts assembled from the deal record, the relevant case studies and your standard scope language — then edited and priced by a person. Pricing and commitments stay human, always.
What does a first engagement cost and how long does it take?
Four weeks to production for the first workstream, which is almost always CRM hygiene plus enrichment on write. That ordering is deliberate and occasionally unpopular, because the visible wins are in lead routing and proposal drafting. Building those on unreliable records means automating your existing data problem, so hygiene goes first and the routing work lands in weeks five to eight.
Will this work with our CRM?
With HubSpot, Salesforce and Zoho, yes, and those cover most of what we see. Anything with a reasonable API is workable. The question worth asking is not whether we can integrate but whether your CRM is configured in a way anyone can automate against — inconsistent stage definitions, three fields meaning the same thing, and required fields nobody fills in are more common blockers than the platform itself.
We tried automating revenue operations before and adoption failed. Why would this differ?
It might not, and the reason matters more than the tooling. RevOps automation usually fails on adoption rather than engineering — reps route around anything that makes their week harder, and a system they distrust gets ignored within a quarter. What changes the outcome is involving the reps who will use it in defining the rules, making the enrichment invisible rather than another form to fill in, and being willing to remove a feature that is not being used rather than mandating it.
Who owns this once you are gone — sales ops or engineering?
Sales operations should, and we build toward that. Scoring rules, routing logic and field mappings live in configuration a non-engineer can read and change, not buried in code. If every rule change needs a developer, the system ossifies within two quarters because sales reality changes faster than an engineering backlog moves. We name the owner during the engagement and hand over to that person specifically.
Can it forecast?
It can improve the inputs to your forecast substantially, which is usually the real problem. Forecasts are wrong because stage definitions are inconsistent and close dates are aspirational, not because the arithmetic is hard. Fix the data and your existing forecast improves. A model predicting deal outcomes on top of unreliable stages produces a confident number with the same underlying error, which is worse than an obviously rough estimate.
Let's find out what is actually automatable
Bring a process that annoys you. In 30 minutes we will tell you whether AI helps, what it would cost, and where it would fail — even if the answer is don't bother.
Or email [email protected] · we reply within 1 business day