AI for Sales Operations in Telecom: Quotes, Margin and the Deal Desk

A business deal in telecoms is an assembly job before it is a sale. Somebody has to work out whether each site can be served, on which access technology, at what build cost, by when, and whether the bundle still earns anything once the discount the customer asked for is applied. That work lives across a serviceability tool, a rate card, a hardware price list, a partner portal and the memory of whoever priced the last one like it.

This page is about that commercial layer, and it stops where the order does: once an order is clean and accepted it belongs to the delivery side of the house. Everything before that, and the in-life commercial events that follow, is where an agent earns its place.

Serviceability before a price is promised

The gap between a price you can honour and one you cannot is usually a site nobody checked properly. An agent can run the checking pass: take the address list off the opportunity, normalise each entry into something the serviceability systems will accept, query the access options per site, and return the answers as one table rather than forty screenshots in a thread.

The output is an evidence sheet. Per site: the access types returned, the lead time band quoted, whether you already serve that building, and whether the answer is a clean yes, a yes conditional on survey, or an unknown.

The value of an honest unknown

The most useful thing the agent produces here is the flag it refuses to fill in. A site whose address did not match cleanly, or where the only answer was a long build, gets marked as unresolved with the reason attached.

A multi site quote with a different answer per site

On a twelve site deal, each site can carry a different access technology, a different lead time and a different build contribution. Pricing them as one average is how a deal looks fine on the sheet and loses money on the ground.

The agent builds the quote line by line from the serviceability evidence and the applicable rate card, keeping each site distinct. A line reading "Unit 4B, Westgate Park" shows its own access type, bandwidth, term, install charge and monthly rate, with the rate card version it was priced from recorded against it.

The salesperson sees a draft quote with every assumption visible and every unpriced site listed separately, not a confident document with a hole in the middle.

The bill of materials finance will recognise

Sales thinks in monthly value. Finance thinks in recurring and non-recurring, and in which costs are yours versus a third party's.

An agent assembles the full materials list behind each site: router or gateway, resilience hardware, cross connects where the service lands in a shared facility, engineer install, out of hours loading, and any third party build contribution passed through.

Splitting the numbers the way the finance team reads them

It then separates one-off charges from the recurring base, marks which recurring lines are cost of sale against a supplier rate and which are your own network, and notes anything with no cost reference at all. That last category is where surprises live.

Margin on a bundle, element by element

A bundle that mixes connectivity, hardware, a managed wrap and a term commitment can be approved on a blended margin and still be a bad deal, because the discount landed on the one element with the thinnest cover.

The agent prices the bundle twice: once as the customer sees it, once decomposed by element with the supplier cost, the discount applied and the resulting margin against each. Where a discount takes an element below a floor, it says which element, by how much in relative terms, and what the payback looks like once install cost is amortised over the committed term.

It approves nothing. It makes the case legible fast enough that the deal desk argues about a position rather than building one.

The deal desk pack, ready before the meeting

Most deal desk time goes on reconstruction: pulling the quote, hunting the cost references, finding comparable deals, then waiting a day to hear why the term is shorter than standard.

An agent can have that pack waiting: the quote, the feasibility evidence, the materials list, the margin decomposition, the non-standard terms requested, comparable approved deals from your own history, and the approvals this shape of deal requires. Open questions are written as questions, with the person who owns each one named.

The pack is evidence for a judgement, never a decision.

Where quoted intent goes missing in the paper

The order form, the schedule, the service levels referenced, the term and the renewal mechanic are where quoted intent goes missing. A site gets dropped, a bandwidth changes, an install charge that was waived reappears, a term lands as thirty-six months when the price assumed sixty.

The agent reads the signature-ready paper and the approved quote side by side and reports differences by field: sites present in one and not the other, charges that moved, service level tier mismatches, term and notice period inconsistencies, and whether the renewal clause matches what the pricing assumed. Every finding cites the clause and the quote line it came from.

Group, subsidiary, site and service identifier

Telecoms account structures are hard in a specific way. A parent group signs, trading subsidiaries consume, sites take delivery, billing accounts carry the charges, and service identifiers sit underneath, with each layer holding a different name for the same customer.

An agent can maintain the map continuously: link billing accounts to legal entities, attach services to sites, reconcile the sites on the account record against the sites that actually carry live services, and raise conflicts as conflicts. Where two records plausibly describe one entity, it proposes the merge with its evidence and leaves the merge to a person. Pricing, forecasting and renewal preparation all fail quietly when this layer is wrong.

Forecasting revenue that starts on an install date

Recurring revenue in telecoms starts when the service goes live, not when the contract is signed, so the forecast is really a schedule of expected live dates and a slipped build moves money into a later period without anybody filing an update.

The agent rebuilds the roll-up each week from the underlying commitments: contracted monthly value per site, expected live date, current delivery milestone, and the variance against the date the forecast assumed. It reports what moved, which sites caused it, and which forecast lines rest on an install date nobody has reconfirmed.

Moves, adds, changes and ceases on a live estate

The unowned work is in-life. A customer closes a site and the cease removes revenue nobody had flagged. A service bills on for a year after the unit it served was vacated. A term auto-renews at list price and the account team hears about it from a complaint.

An agent watching the estate against the contract base can surface all of it: ceases pending and their revenue effect, services with no usage across a sustained period, commitments at risk of breach, and every term approaching its notice window with the notice date and current mechanic stated. The same material becomes the re-sign pack, so a renewal opens from the customer's actual estate and usage rather than a rep's recollection of it.

Rate cards, channel partners and clawback

Wholesale and partner motions price off rate sheets that change under you. An agent can diff a new card against the version in force, list the products whose rates moved, and flag live quotes and renewals that were priced on the superseded version before they are signed.

The same discipline applies to money going out. Commission and partner payments can be rebuilt from the order and billing record, including clawback where a service ceased inside its qualifying period, with the qualifying event, the cease date and the affected lines cited. When a partner disputes a statement the reconciliation already exists.

Where AI fails here

Build cost on an unsurveyed site is the clearest limit. No agent can tell you what the civils will cost where nobody has looked, and a confident estimate here is worse than a blank, because it gets quoted. Treat every unsurveyed site as an open question with a named owner.

Account hierarchies that are wrong in the source system stay wrong. An agent reading a broken parent-child structure will produce a tidy, authoritative, incorrect view of who the customer is, and that view is more dangerous than the mess it replaced because people trust it.

Commission plans that exist only as a slide and a verbal understanding cannot be computed. Neither can a bespoke discount scheme that is really a relationship. If the rule is not written down, the agent should state the ambiguity rather than resolve it.

The limits bite hardest where money is decided. The agent never approves a discount, never signs or amends contract paper, never commits a delivery date and never issues a credit. Adopt AI Partners builds these with the approval boundary in the design, rather than bolted on afterwards.

Hand over the quote desk before the forecast

Start where the evidence is objective and the rework is worst: feasibility packs and multi site quote assembly. Both produce something checkable on day one, so a reviewer can tell at a glance whether the agent was right. Add the bill of materials and the margin decomposition next, then contract checking, then the in-life estate watch. Leave the forecast until the account structure underneath it has been cleaned, because a forecast built on a wrong hierarchy fails in a way that is hard to see.

If you want to work out which part of your quote to contract path would survive this first, email hello@adoptaipartners.com and we will go through it on your own deal shapes.